The resilience of road infrastructure is vital for maintaining community mobility and ensuring the continuity of critical services, particularly in the face of escalating challenges posed by climate change. Among these challenges, the increasing frequency and intensity of extreme weather events often manifest as floods, posing a substantial threat to urban road networks in low-lying coastal areas. These regions are especially vulnerable to multiple flood drivers, including tidal surges, streamflow, and precipitation. The co-occurrence of extreme rainfall with high tides and elevated streamflow levels amplifies flood inundation depths, yet the compound effects of these flood drivers on road infrastructure damage remain underexplored. This study proposes a quantitative framework to assess the dynamic interaction of compound flood events and their impacts on road infrastructure systems, with a focus on damage assessment. Using the extreme weather events of 2018 in Kozhikode, Kerala, India, as a case study, we integrate disparate flood hazards—pluvial (rainfall-induced), fluvial (streamflow), and coastal (storm tide)—to evaluate flood risk and road damage. A 1D-2D hydrodynamic modeling approach, coupled with depth-damage curves, quantifies the repair and maintenance costs for roads affected by compound flooding. Our findings reveal that pluvial flooding accounts for 93% of road damage, while fluvial and coastal flooding contribute 5.6% and 1.4%, respectively. This framework highlights the disproportionate impacts of different flood drivers and enables the identification of the primary contributors to road damage. Such insights can inform targeted adaptation strategies tailored to the unique needs of specific regions, enhancing infrastructure resilience against future flood events.

How to cite: Dave, R., Sen, S., and Bhatia, U.: Disproportionate Impact of Compound Flood Events on Road Infrastructure Damage, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-834, https://doi.org/10.5194/egusphere-egu25-834, 2025.

Coastal cities are increasingly affected by flooding due to the combined impacts of surface and subsurface water processes. Compound flood events, driven by changing groundwater levels, tidal surges, and riverine, pose substantial risks to urban infrastructure and livelihoods, particularly in low-lying coastal cities like Mumbai. Despite its critical importance, the role of groundwater dynamics in flood severity and its contribution to comprehensive flood risk assessments remain underexplored. In this study, we quantify flood risks induced by multiple drivers and their contributions to compound events. We integrate surface and subsurface flooding using MIKE+ and FEFLOW to simulate 1D-2D coupled hydrodynamic models, respectively. Mumbai serves as the study area due to its susceptibility to tidal surges, riverine, and groundwater flooding. By incorporating tidal, well, and streamflow data, our study quantifies the contribution of groundwater to surface flooding, offering a deeper understanding of the interplay between subsurface and surface water processes. Our findings lay the foundation for proactive groundwater management strategies and promote the development of resilient urban infrastructure, ultimately mitigating the impacts of flooding in vulnerable coastal areas.

How to cite: Sutrave, M., Kumar, A., Dave, R., and Bhatia, U.: Quantifying Subsurface Contributions to Compound Flooding in Coastal Urban Areas for Enhanced Resilience, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-867, https://doi.org/10.5194/egusphere-egu25-867, 2025.

Surface drainage system within the polder is designed to collect internal inflow, resulting
from rainfall, and external inflow, resulting from sea water infiltration, that gravitates into
the canal network. Excess water that impedes agricultural production is cyclically
pumped out of the polder, lowering the water level below the root zone. Water level
monitoring in the main canal of the drainage network is set-up as continuous real-time
measurements of surface water levels and index water velocity using radars. The
Automated Continuous Monitoring System installed under the DELTASAL project
consists of surface water regime monitoring, water quality monitoring, soil salinity
monitoring, and the monitoring of weather conditions with sensors integrated to provide
synchronized real-time data. The data collected is available to the stakeholders, polder
users, and public through an online platform. Co-creation is central to the DELTASAL
project, involving stakeholders (research-oriented community, public administration, local
authorities, and farmers) in every phase, from problem identification to development of
guidelines to optimize the water regime for agricultural use. Through workshops
stakeholders exchange requirements of water quality and quantity, validate findings, and
discuss the potential solutions that are aligned with agricultural goals specific to the
polder. The overall goal is to provide functional prognostic model as a platform for polder
management. The participatory approach aims to foster collaborative decision-making,
improving the sustainability of the drainage system. The objective of this research is to
determine the relationship between canal surface slope, water volume, and pump flow
rate under different water regime management scenarios. The research uses surface
water levels in the canal and associated pump flow rates as primary inputs to develop a
drainage optimization model as a part of the water quantity/quality prognostic model.

How to cite: Perokovic, D. and Gilja, G.: Water surface slope variation in Vidrice polder resulting from pumping operation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2538, https://doi.org/10.5194/egusphere-egu25-2538, 2025.

Climate change mitigation strategies face disruption from multiple sources: extreme climate events, socioeconomic crises, geopolitical conflicts, technological breakthroughs, as well as the abrupt transitions and disruptive actions entailed by achieving the stringent Paris Agreement goals. While these disruptive events can fundamentally alter long-term mitigation scenarios, the current literature does not sufficiently assess their implications. Existing long-term mitigation scenario narratives and modelling frameworks, using Integrated Assessment Models (IAMs), lack systematic approaches to analyse their impacts. To address this gap, we introduce the Disruptive Events-Resilient Pathways (DERPs) framework, which provides structured narratives to systematically explore and assess the resilience of climate actions to the impacts of external disruptions and entailed abrupt transitions. To operationalise this framework, we employ multiple IAMs to analyse case studies of distinct disruptions: intensifying heatwaves and droughts affecting energy systems, and the rapid uptake of Direct Air Carbon Capture and Storage (DACCS) technology. Our analysis highlights the inherent limitations of IAMs in capturing the full complexity of disruptive events. We offer novel methodological approaches to overcome them. Our results provide insights into the interplay between the impacts of disruptive events and mitigation scenarios.

We introduce a conceptual framework, alongside qualitative narratives and use cases for validation, to guide the development of Disruptive Events-Resilient Pathways (DERPs). This framework systematically explores the impacts of disruptive events on mitigation and adaptation strategies, allowing to evaluate their resilience to such disruptions. Similar to the widely-adopted SSP scenario framework, which maps socioeconomic developments onto the extent of challenges to mitigation and adaptation (O’Neill et al., 2017), the DERPs framework comprises two dimensions, thereby enabling breaking the developed spectrum into four blocks of narratives, plus an intermediate narrative that reflects current trends. We further reflect on the connection between the DERP and SSP frameworks in the discussion section below. 

The DERP dimensions and underlying narratives draw on concrete examples, to make the framework more comprehensive and comprehensible, but remain sufficiently generalisable to allow the framework to serve as a blueprint for conducting similar types of mitigation and adaptation analyses in the future. In determining the two dimensions of the DERP framework, we benefit from van Ginkel et al. (2020), who had proposed two dimensions for exploring how climate change tipping points can cause socioeconomic tipping points (SETPs). In the DERP framework, the focus shifts from climate change tipping points to disruptive events as the drivers of socioeconomic impacts, and on assessing societal resilience to these impacts. Accordingly, the two dimensions are defined as follows:

• climate action effectiveness: this refers to significant and deliberate change in the way societies and systems transition towards mitigating, or preparing for (i.e. adapting to), climate change
• resilience to socioeconomic impacts: this refers to the capacity to withstand unintended shifts in socioeconomic structures that may occur due to abrupt transitions or insufficient mitigation or adaptation failure, and the resulting climate change impacts. 

How to cite: Al Khourdajie, A., Nikas, A., Frilingou, N., Mittal, S., van de Ven, D. J., Fragkos, P., McJeon, H., Byers, E., Keppo, I., Peters, G., McCollum, D., Zisarou, E., Hawkes, A., and Gambhir, A.: Navigating the unexpected: The impact of disruptive events on mitigation scenarios, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6022, https://doi.org/10.5194/egusphere-egu25-6022, 2025.

Planktonic bacteria and archaea play a key role in river nutrient biogeochemical cycling; however, their respective community assembly and how to maintain their diversity are not well know in dammed rivers. Therefore, a seasonal survey of planktonic bacterial and archaeal community compositions and related environmental factors was conducted in 16 cascade reservoirs on the Wujiang River and the Pearl River in southwest China to understand the above mechanisms. The result showed that deterministic processes dominated bacterial and archaeal community assembly. Planktonic bacteria and archaea in dammed rivers had different biogeographic distributions, and water temperature was a key controlling factor. Water temperature can directly or indirectly affect the microbial diversity. Planktonic bacterial diversity increased with increasing water temperature, while archaea showed the opposite trend; the overall diversity of bacteria and archaea was no significant changes with changeable water temperature. Abundant microbes had a stronger distance-decay relationship than middle and rare ones, and the relationship was stronger in winter and spring than in summer and autumn. The different responses of planktonic bacterial and archaeal diversity to water temperature could be due to their different phylogenetic diversity. This ultimately maintained the stability of total microbial community diversity. This study reveals the different responses of planktonic bacteria and archaea to water temperature and perfects the theoretical framework for planktonic microbial biogeography in dammed rivers.

How to cite: Liu, N., Wang, B., and Yang, M.: The different responses of planktonic bacteria and archaea to water temperature maintain the stability of their community diversity in dammed rivers, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6325, https://doi.org/10.5194/egusphere-egu25-6325, 2025.

Humans’ attachment to the place where they live is widely recognized. Nevertheless, landscapes can be characterized by aspects that make their communities prone to natural hazards. When a disaster occurs, the relocation of the people involved can be necessary. Such a relocation, however, can be opposed by interested communities, given the place attachment (PA) to the environment at risk. For this reason, a clear understanding of this aspect is mandatory to better calibrate risk adaptation measures involving relocation. In this work, a systematic review of the role of PA in the management of relocation measures was carried out. The review followed the PRISMA protocol (Preferred Re-porting Items for Systematic reviews and Meta-Analyses). The findings indicate that generally, strong PA is associated with a low propensity for relocation, regardless of risk perception or awareness levels. This low propensity is related mainly to the fact that the place of relocation cannot satisfy the symbolic needs associated with the place of origin. On the other hand, PA is often linked to a greater propensity to take care of the place in which people live. Therefore, it can lead to the realization of adaptive behaviors. From this perspective, among scholars, there is consensus that PA needs to be considered in the construction of strategies for natural hazard management involving relocation. In addition, the literature shows that there have also been attempts to develop attachments to new relocation sites. These attempts have had mixed results. Therefore, it is even more important to further investigate the role of PA as a nonstructural measure to improve the resilience of populations affected by natural disasters.

This work was funded by the Next Generation EU - Italian NRRP, Mission 4, Component 2, Investment 1.5, call for the creation and strengthening of ‘Innovation Ecosystems’, building ‘Territorial R&D Leaders’ (Directorial Decree n. 2021/3277) – project Tech4You - Technologies for climate change adaptation and quality of life improvement, n. ECS0000009

How to cite: Carone, M. T., Vennari, C., and Antronico, L.: Place attachment and relocation, a difficult combination, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6835, https://doi.org/10.5194/egusphere-egu25-6835, 2025.

Citizen science in marine biodiversity monitoring encounters several challenges such as obtaining unstructured data which may lead to underestimate species presence or introduce spatial bias. In addition to those inherent to the marine environment. To address these challenges, efforts must be directed towards (1) enhancing participant engagement to increase the volume of data collected, and (2) developing methods to standardize the unstructured data obtained. 

As of December 2024th, over 260.000 observations have been recorded during the course of four years by more than 870 volunteer participants documenting over 2900 species, including some historical observations, in the Coastal region of Catalonia, located in the northeast of Spain. These observations have been reported and validated in the citizen science observatory MINKA (minka-sdg.org).

This presentation will highlight the lessons learnt through the past four years, the opportunities and the remaining challenges to address. 

How to cite: Companys, B., Alvarez, A., Salvador, X., Liñan, S., and Piera, J.: Citizen Science in marine biodiversity unstructured monitoring , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8787, https://doi.org/10.5194/egusphere-egu25-8787, 2025.

The Water-Energy-Food (WEF) nexus has emerged as a critical framework for addressing resource interdependencies and building resilience against climate change impacts. Despite its growing prominence, significant knowledge gaps remain, particularly in quantifying resilience and integrating cross-sectoral dynamics into actionable policymaking. This review synthesizes existing literature on the WEF nexus, focusing on its evolution, current trends, and resilience frameworks. Employing meta-analysis, this study quantifies key trends in WEF nexus resilience research, identifying dominant methodologies, geographic patterns, and gaps in policy and practice.

The findings reveal a global emphasis on conceptual frameworks and modelling approaches, with limited application to localized contexts, especially in India. To bridge this gap, this study highlights the need for policy coherence analyses and system dynamics modelling to assess resilience of the WEF-nexus under various climate scenarios. Thus, providing actionable insights for researchers and policymakers, emphasizing the importance of integrated, scalable, and data-driven approaches to enhancing the resilience of WEF systems.

How to cite: Cheranda, T. M., Santhanam, H., and Murthy, I. K.: A Systematic Review and Meta-Analysis of Water-Energy-Food Nexus Resilience: Global Insights and Implications for India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9262, https://doi.org/10.5194/egusphere-egu25-9262, 2025.

The invasive alien, African giant snail, Lissachatina fulica, considered as a pest, poses a significant threat to ecosystems, human health and agriculture across tropical and subtropical regions. Therefore, in order to address the challenge posed by the presence of this animal outside its original habitat it is essential to to understand its current and potential future distribution. Thus, this study, which highlights the critical role of regional bioclimatic datasets in improving the predictive accuracy of species distribution models (SDMs) particularly for invasive species in ecosystems with complex orography or climate, takes advantage of global and regional bioclimatic datasets to model the future distribution of L. fulica in the Canary Islands, emphasizing the influence of the archipelago’s complex orography and unique microclimates.

Our approach integrates two distinct datasets as input of the SDM Maxent. First, we used the global distribution of L. fulica from GBif and a list of bioindicators taken from the WorldClim and Chelsa datasets to train the model, which allowed us to capture the environmental niche of the species under various climatic conditions. We then applied the BICI-ULL dataset, a high-resolution bioclimatic dataset specifically developed for the Canary Islands that accounts for the intricate topography and varied microclimates of the archipelago, providing an unprecedented resolution for regional analyses. This dataset allows us to perform our projections in two different future periods, mid- (2041-2060) and end-of-century (2081-2100) and under two scenarios for greenhouse gas concentration, namely the CMIP5 representative concentration pathway 4.5 and 8.5 (RCP4.5 and RCP8.5).

The results indicate that while the current distribution of L. fulica in the Canary Islands is limited to the wetter areas of the archipelago, namely western islands such as La Palma and El Hierro and the north of Tenerife, future projections under the CMIP5 RCP4.5 and RCP8.5 scenarios reveal notable changes. For both temporal periods and driven by warming temperatures and changing precipitation patterns, habitat suitability shows a greater shrinkage remaining only a small favorable area in the northern part of La Palma. However, future projections performed with the global datasets show opposite results, that is, a large number of high-suitability zones throughout the entire archipelago in which the probability of the presence of L. fulica is very high.

The use of BICI-ULL allowed us to identify future patterns in the high-suitability zones that would have been overestimated using global datasets. This underscores the need of incorporating region-specific data when modeling species distributions in topographically complex areas such as oceanic islands. The findings highlight the importance of developing regional datasets, like BICI-ULL, that can capture microclimatic variability. Besides, this approach serves as a model for addressing similar challenges in other biodiversity-rich but vulnerable regions, contributing to the broader understanding of invasive species dynamics.

How to cite: Barragan, R., Sosa-Guillén, P., Tondreau, P. S., Pérez, J. C., Expósito, F. J., and Díaz, J. P.: Evaluating the Importance of Region-Specific Bioclimatic Datasets in Projecting the Future Distribution of Lissachatina fulica in Complex Landscapes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11483, https://doi.org/10.5194/egusphere-egu25-11483, 2025.

Tea, Camellia sinensis (L.) Kuntze is a globally significant crop, with approximately 6.6 billion cups consumed daily, making it the second most consumed beverage after water. It supports millions of livelihoods and contributes significantly to regional economies, particularly in Africa. Despite its importance, monitoring tea plantations in the continent remains manual as there are no spatially-explicit maps, thereby hindering efficient quantification of forest and biodiversity changes associated with tea cultivation, for instance. Here, we present the first high-resolution map of tea plantations in Africa, developed using computer vision techniques integrated with high-resolution satellite imagery and ground-truth polygons. Our approach achieves unprecedented spatial accuracy in delineating the area under tea cultivation with an overall accuracy of 97%. This milestone lays a foundation for spatially-explicit monitoring of tea plantations, enabling applications such as yield estimation, pest and disease detection, protected area encroachment analysis, carbon stock assessments, biodiversity impacts investigations, and evaluation of climate-driven range shifts, among others. 

How to cite: Oluoch, W. A., Drees, L., Wegner, J. D., and Wuepper, D.: Mapping Tea Plantations in Africa with Computer Vision, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11581, https://doi.org/10.5194/egusphere-egu25-11581, 2025.

Environmental communication is crucial in shaping the narrative around the aggravated issues like climate change, global warming, sea-level rise, air-pollution and pushing for impactful actions. With the burgeoning economy and rising population, the large democracy of India is facing a critical issue of air pollution in major cities, with Delhi consistently ranking first due to its persistently high air quality index (AQI) throughout the year. The Commission on Air Quality Management (CAQM) of India, a statutory government body, works diligently to improve the air quality of Delhi and the National Capital Region (NCR). In July 2022, in accordance with the directives of the Honourable Supreme Court of India, CAQM launched the CAQM policy to find a permanent solution to air pollution, aiming to develop an inclusive policy addressing all sectors contributing to and affected by pollution. This study aims to explore the ways in which environmental challenges, such as air pollution, are conveyed to both citizens and policy makers through environmental communication. Upon analysing 277 news articles from eight leading news agencies, including four newspapers and four news channels, over a six-month period prior to the emergence of the CAQM policy in 2022, it was observed that news coverage is heavily concentrated during the post-monsoon season (with 68% of the analysed news articles concentrated in October and November). This period in North India is prominently under focus due to 'stubble burning' activities mainly occurring in the states of Punjab and Haryana. Hence, a strong connection between the news articles and active number of fire locations has also been found. However, it was found that news articles do not proportionately reflect fluctuations in the Air Quality Index (AQI), for example, no news articles were noted on December 23rd and 24^th 2022, despite AQI values reaching 524 and 522 respectively. Similarly, from January to March 2022, news coverage was minimal despite high AQI levels, indicating that coverage is more linked to periods of higher fire activity in Punjab and Haryana, rather than AQI levels in Delhi. We also attempted to find a possible connection between the issues raised in the news media during the period of our interest and the CAQM policy that was formed in April 2022. It is noticed that, while the CAQM policy aimed to improve air quality and, consequently, public health, media coverage paid relatively less attention to the health implications (barely 56 articles mentioning health or mortality). The recommendations for a new policy did rise but again from November 10th to December 5th, with 96 articles published during this period, suggesting a period-specific coverage. This indicates that media reporting focuses heavily on stubble burning, whereas the CAQM policy treats it as just another pollution source, without special emphasis. Hence, for our case study, it is noted that the media's coverage of environmental communication seems to be less comprehensive and lacks depth compared to the detailed measures outlined in CAQM policy addressing air pollution.

How to cite: Pandit, S., Govardhan, G., and Ghude, S.: Analysing the role of environmental communication with respect to CAQM policy on Delhi air quality, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11694, https://doi.org/10.5194/egusphere-egu25-11694, 2025.

As a pioneering pan-European distributed research infrastructure, DANUBIUS-RI is unique to Europe and the international community, focused on interdisciplinary research on River-Sea systems.  At a time of unprecedented environmental and climate driven change, it is critical we understand the influence of land-sea interactions on coastal and ocean areas, how these will change with the intensification of extreme events and seek sustainable climate adaptation solutions. DANUBIUS-RI is a trans-disciplinary research gateway, enabling land-sea researchers with access to data, expertise, training and key study sites, along with their associated local assets. 

DANUBIUS-RI responds to the following major Research Priorities to assess:

• Water Quantity: water stores and flows across River-Sea continua for sustainable water resource management and mitigate against extreme events.
• Sediment Balance:    sediment dynamics in source-to-sink systems, to support sustainable sediment management.
• Nutrients and Pollutants: independent and combined effects of nutrients and pollutants (in both water and sediments) at River-Sea System scales, to establish the critical thresholds needed for tracking progress towards good status.
• Climate Change: ongoing impacts of Climate Change, and improve adaptation measures within and across River-Sea Systems.
• Extreme Events: extreme event occurrence and impact severity on River-Sea Systems, for floods and droughts, to support cost-effective nature-based solution development, disaster mitigation, and management.
• Protecting and Restoring Ecosystems and Biodiversity: how changing River-Sea Systems affect future ecosystem service provision, and their sustainability. Understand the relationship between biodiversity and connectivity across River-Sea Systems and its response to multiple stressors and support climate adaptation.
• Digital Twin: and build high resolution, multi-dimensional digital representations of River-Sea Systems, that stakeholders.

Besides access to Open and FAIR Data, DANUBIUS_RI provides key interdisciplinary services encompassing in situ measurement, satellite EO observations, numerical modelling and management scenario development.

The Services are grouped in 7 major categories, working with you or on your behalf:

• Digital and non-digital data, including metadata, data and archived samples .
• Tools, methods and expert support, including access to facilities and equipment.
• Measurement and analytical support, including physical, chemical, biological, biogeochemical, ecotoxicological, hydromorphological, sedimentological, and bathymetric sampling and analyses.
• Diagnosis and Impact, through modelling and impact assessment analysesthat harness data from previous or expected results (diagnostic) or through forecasts and ‘what if’ scenarios (from models).
• Solution Development, connecting you with the right partners across wide-ranging scientific expertise to develop solutions for your specific challenges.
• Tests, Audit, Validation and Certification: We validate and quality assure outputs, and provide DANUBIUS Commons accreditation and Accredited Service Providers certification services.
• Build capacity through the design/co-design, development and delivery of training courses for companies, innovators, authorities and researchers in the four areas of expertise (Observation, Analysis, Modelling, and Impact), and partner with you to organize bespoke conferences and workshops to address River-Sea System challenges.

The Research Infrastructure, accepted on the ESFRI Roadmap in 2016, is expected to become an operational ERIC during 2025.

How to cite: Stanica, A., Tyler, A., Scarrott, R., and Consortium, D. R. I.: DANUBIUS-RI, The International Centre for Advanced Studies on River-Sea Systems:  An RI for the river-sea challenges of the 21st Century , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11716, https://doi.org/10.5194/egusphere-egu25-11716, 2025.

Climate change has led to the shrinkage of ice sheets and glaciers, contributing to sea level rise, particularly in regions like West Antarctica. Over the past several decades, this area has experienced one of the most pronounced increases in temperature and precipitation. Projections suggested increase in extreme precipitation by the end of the 21st century. Together with the expected deepening of the Amundsen Sea Low (ASL), these changes play a significant role in Antarctic ice sheet's mass in the future. This study aims to analyze a spatio-temporal precipitation variability and its extreme values in West Antarctica as a response to ASL characteristics. To analyze the relationships, we used a number of parameters describing ASL (average pressure field, the central pressure, the relative pressure at the center, longitude of the ASL, and the distance to the ASL center), and parameters for precipitation (daily totals and the 95^th percentile) derived from the historical European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis data. The study is focused on natural zones along the coast corresponding to glacial basins such as Getz Ice Shelf, Thwaites Glacier, Pine Island glaciers, and Abbot Ice Shelf. Relationships between precipitation and ASL characteristics were assessed using Spearman rank correlation coefficients in each grid cell of the studied domain. Overall, the highest 95th percentile values, approximately 35 mm, were observed along the western coast of the Antarctic Peninsula. These values decreased to 15 mm along the remaining coastline of West Antarctica and further to 5 mm over the continental areas. Extreme precipitation had well-detected seasonality, with maximum precipitation totals during the austral autumn/spring seasons. In average, extreme precipitation events covered approximately 4.7–4.9% of basin areas. Over the last 30 years, the tendencies of extreme precipitation intensified the observed spatial differences: the 95th percentile increased over more humid areas with a trend of 4 mm/decade and decreased in continental regions by 2 mm/decade. The meridional position of ASL impacts weather and precipitation over the region much more than changes in its latitudinal remoteness to the coast. The ASL movement towards the west caused decreased precipitation near the Amundsen Sea and increased over the Antarctic Peninsula. Extreme precipitation was more sensitive to changes in ASL location than total precipitation. This study will contribute to understanding the occurrence of extreme precipitation events under climate change.

How to cite: Pysarenko, L. and Pishniak, D.: Change in precipitation as a response to the Amundsen Sea Low characteristics in region of West Antarctica, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12275, https://doi.org/10.5194/egusphere-egu25-12275, 2025.

Severe weather events increasingly threaten the safety of mass gathering events (MGE), particularly open-air exhibitions and artistic performances. In 2015, Milano, Italy, hosted the World Expo from May to October. The exhibition was located 15 km away from Milano, covered 1.1 km2, and shaped as a long boulevard of 3 km length. Pools and waterways in and around the Expo area were elements of primary importance. During the 184 opening days, the attendance reached 21 mln visitors, with a daily average of 115,000 visitors. In 2017, the artists Christo and Jeanne-Claude created the temporary, site-specific artwork known as The Floating Piers, built in 2016 at Lake Iseo, 75 km from Milano, Italy. It was made up of 70,000 square meters of yellow fabric supported by a modular floating dock system. These walkable piers connected Monteisola Isle to the lake coast. The floating piers exhibitions attracted 1.2 mln visitors over its 16-day run, with peaks of more than 100,000 visitors per day.

This work describes how the regional weather service planned and operated a dedicated monitoring and forecast weather service to reduce the impacts of severe weather during these two MGEs, increasing safety conditions for the visitors. Finally, Arpa Lombardia will be engaged in the weather forecast assistance for the next Winter Olympic Games in 2026, hosted in  Milano, Bormio, and Livigno.

How to cite: Cremonini, R., Minardi, G. P., Bechini, R., and Cazzuli, O.: Reducing the impact of severe weather on mass gathering events: the Lombardia, Italy, experience., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13485, https://doi.org/10.5194/egusphere-egu25-13485, 2025.

This study presents the design and implementation of a comprehensive Hail Operational Aircraft Information Dashboard system. This system, which aims at user-friendliness and a customizable visualization platform, is a helpful tool for enhancing decision-making and optimizing cloud seeding operations for hail suppression. It leverages real-time and historical data, making it accessible and easy to coordinate operations with flight crews. The dashboard display system's key functionalities include real-time flight monitoring, which displays critical flight parameters such as flight duration, cloud seeding duration, and flight path for operational aircraft. The system also offers insightful data visualizations that cover weekly, monthly, and seasonal trends of hail suppression efforts, providing a wealth of information that supports the users with a comprehensive understanding of the operations. The dashboard system features a user-friendly frontend interface developed with ReactJS, a high-performance backend run by the FastAPI Python framework for efficient data handling and API development, and SQLAlchemy as the object-relational-mapper to store all flight and hail suppression data.  Additionally, and as an innovative approach, this study explores the use of Large Language Models (LLMs) for text-to-SQL (TTS) conversion, allowing users to submit natural language queries about hail operations, which the LLM translates into SQL queries to retrieve relevant data. The dashboard visual system incorporates additional parameters for operational decisions, such as flight altitude, fuel consumption data, and seeding information. This system is expected to significantly enhance situational awareness for flight crews, providing them with a comprehensive view of the operations. Previously, these users relied on disparate sources of information and less integrated tools to manage their operations. This heightened awareness will help coordinate unit and flight crews to make better decisions and improve hail suppression efforts.

How to cite: Kumar, S., Tani, S., Paulitsch, H., and Schreck, T.: "AI-Integrated Operational Dashboard for Hail Defense Operational Systems", EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14032, https://doi.org/10.5194/egusphere-egu25-14032, 2025.

Climate change poses significant challenges to Small Island Developing States (SIDS) through heat extremes, hydrometeorological extremes and sea level rise. The societal impacts of these climate hazards are closely connected to both quantifiable and non-monetary loss and damage across multiple sectors and to presently or potentially insurable risks. One type of insurance that has been explored in many developing country contexts but is particularly sensitive to the recurrence frequency of extreme events is climate index insurance (analogous to parametric insurance), in which the contract is based on a geophysical index, rather than verified material losses.

This study explores the historical risk of heat and precipitation extreme events in the small Caribbean Island nation of St. Kitts and Nevis over the period of available record (1981-2024) and the projected frequency and severity of such events over the next 50 years (2025-2075), using historical analysis, model data and Monte Carlo statistical simulation methods. Observational data will include merged station/satellite data from the products of the Climate Hazards Group at University of Santa Barbara (CHIRPS, CHIRP and CHIRTS) and may include local station data. Climate model data will include output from CMIP6 runs of the NMME and Copernicus model suites. The Monte Carlo methods used for estimating extreme event frequencies are based on earlier research (Siebert and Ward 2011, Siebert 2016). As climate risks increase, theoretical index/parametric insurance premiums are expected to increase.

            Since the frequency of threshold crossing extreme events is the primary basis for pricing index (parametric) insurance contracts, this study will explore the evolving price of relevant parametric insurance contracts for specified return liabilities (defined through recurrence interval). This project is being conducted by the company Climate Analytics and is funded by the UN Office for Project Services (UNOPS). This methodology may inform the quantification of a national loss and damage policy and plan, in coordination with multiple stakeholders in St. Kitts and Nevis and the Caribbean Climate Risk Insurance Facility (CCRIF).

How to cite: Siebert, A.: Potential Index Insurance Changes under Climate Change in St. Kitts and Nevis: A Case Study Using Monte Carlo methods, observational and GCM data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14530, https://doi.org/10.5194/egusphere-egu25-14530, 2025.

This study investigates the spatiotemporal characteristics of maximum temperatures and heat wave (HW) vulnerability across India under the context of global warming. Using high-resolution gridded surface air temperature (SAT) data (1951–2022) from the India Meteorological Department (IMD), three regions of maximum temperatures and distinct heat wave zones were identified, highlighting their divergence. Local radiative heating and anomalous wind flows from maximum temperature zones were identified as primary drivers of heat waves, with a notable increase in HW occurrences in southeast India post-1970, attributed to global warming. Machine Learning (ML) models, including Artificial Neural Networks (ANN), multiple linear regression, and support vector machines, were employed alongside CMIP6 climate models to predict maximum SAT for India (1981–2022). ANN outperformed other ML models with minimal biases and high accuracy, showcasing its capability to enhance HW predictability. Future projections (2023–2050) reveal a gradual rise on SAT during March–May, indicating heightened HW risks. Additionally, HW intensification during El Niño decay years was linked to anomalous anticyclonic circulations, reduced cloud cover, and enhanced shortwave radiation. This caused a rise in discomfort indices and extreme temperature hours, particularly in northwest and central India. Findings emphasize the critical role of ML techniques in improving HW forecasts and guiding adaptation strategies. These insights are vital for agriculture, health, urban planning, and disaster mitigation, equipping stakeholders to address escalating climate risks and societal impacts effectively

Keywords: Heat Waves (HW); Maximum Temperatures; Machine Learning (ML); Climate Change; Vulnerability Analysis

How to cite: Satyanarayana, Dr. G. C.: Assessing Heat Wave Vulnerability in India Using Machine Learning and Climate Model Insights, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15035, https://doi.org/10.5194/egusphere-egu25-15035, 2025.

Understanding the hydrological impacts of  Land use (LU) changes and climate variability is vital for effective water resource management in river basins. This study uses the Soil and Water Assessment Tool (SWAT) to assess hydrological processes in the Shakkar watershed, a sub-basin of the Narmada River in Madhya Pradesh, India. The research quantifies the effects of LU changes and climate variability on key hydrological components, such as water yield, surface runoff, evapotranspiration (ET), and base flow, utilizing high-resolution MSWEP v2.8 precipitation data corrected with quantile mapping (QM). A multi-objective calibration approach incorporating Nash-Sutcliffe Efficiency (NSE) and Relative Volume Error (RVE) ensures accurate model parameterization. Variability trends in rainfall and streamflow across three decades (1989-2019) were assessed using the Mann-Kendall test and Sen's slope estimator. This study aims to bridge critical gaps in understanding the interplay between climate and LU changes, providing insights into their cumulative and individual impacts on water resources. Anticipated outcomes include identifying areas within the watershed most vulnerable to hydrological changes and supporting the development of sustainable water management strategies. The findings are expected to guide regional water resource planning and improve resilience against climatic variability by demonstrating the utility of advanced modeling techniques and high-resolution datasets in watershed management.

How to cite: Gopalakrishnan Balamurgan, B., Rientjes, T., and Tügel, F.: The effects of Land use  changes and climate variability on hydrological changes in the Shakkar watershed and supporting the development of sustainable water management strategies, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17037, https://doi.org/10.5194/egusphere-egu25-17037, 2025.

The water engineering group of the University of Brescia is active in the Horizon Europe project LOESS (https://loess-project.eu/) since its start in June 2023, together with other nineteen European –of which two Italian– partners. The final goals of the project are to raise awareness on the importance of soil and of its functions, to increase soil literacy across Europe and to help developing innovative educational materials and practices.
To do so, we –together with the other two Italian partners– created an Italian Community of Practice (CoP) and engaged it in providing an overview of the current level of soil–related knowledge and teaching programmes and materials, in order to identify the gap between educational offer and needs amongst different levels of the society (from pupils to students to citizens). The Italian CoP, led by the University of Brescia, is composed of 60 members from both the higher education and the research community, as well as from the primary and secondary education levels (teachers and pupils), the productive sectors (farmers and spatial planners), the politics world (local administrators) and the civil society (NGOs and associations). The CoP, or various sub–groups of people from it, has been involved in multiple activities since the start of the project, and this contribution intends to report on them.
In March 2022 we launched the WormEx II experiment, an ongoing educational experiment and a citizen–based participatory research (CBPR) performed in the garden of the Liceo Copernico High School in Brescia, in view of attracting the students’ attention on the hydrological role played by macropores, by observing some aspects of earthworm digging activity.
On World Soil Day 2023 we –together with the other two Italian partners– performed a widespread infiltration experiment involving classes from 5 Primary Schools over the Italian territory, i.e. one in Lombardy, two in Emilia Romagna, one in Sicily and one in Sardinia. A total of 140 students and 7 teachers took part in the experimental phase, after which they all joined a virtual meeting where around 50 students (between 4 and 20 per school) volunteered in reporting the experiment result to the CoP, which had previously contributed in the design of the experiment itself.
Between November and December 2024 we organised and conducted three co–creation events on Augmented Reality applications for soil health education in two 3rd-year classes of a local High School in the province of Brescia. The activity produced 22 projects created with a commercial app–prototyping tool from an international project partner.
Finally on 5 December 2024 we –together with the other two Italian partners– organized and hosted a dissemination event about World Soil Day, with considerations regarding the links between soil, peace and sustainability. The public event involved two Primary School classes (one in Emilia Romagna and one in Sicily) that reported on a previously–held laboratory activity on the topic, as well as university students and professors.
These activities showed how much our society is interested in taking an active part in research if allowed.

How to cite: Peli, M., Barontini, S., and Grossi, G.: Experiences of citizen science and co-creation within the activities of the Italian chapter of the LOESS project, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18702, https://doi.org/10.5194/egusphere-egu25-18702, 2025.

Abstract

Yasooj, with its rich cultural, historical, and musical heritage, stunning natural landscapes, and a young, educated workforce, is one of Iran's cities with significant potential to become a leading example of sustainable development. However, challenges such as weak urban management, lack of economic development, inadequate infrastructure, and natural constraints hinder its progress. In this context, assessing the current situation and providing solutions to enhance urban management and formulate sustainable development strategies are of great importance.

The objective of this study is to identify the capabilities and challenges of Yasooj's urban management and to develop a strategic vision for sustainable development, based on the SWOT model and the SPACE matrix. This research is applied-developmental in nature and employs a descriptive-analytical research method. The statistical population consists of 40 urban experts, and the required data were collected through field observations, reviews of comprehensive and detailed plans, and questionnaires.

Data analysis was conducted using the SWOT model, which identified the strengths, weaknesses, opportunities, and threats of Yasooj City. The findings indicate that the final score of the IFE matrix (internal factors of strengths and weaknesses) is 2.10, and the score of the EFE matrix (external factors of opportunities and threats) is 2.37, both of which are significantly below the average score of 2.5. The internal-external (IE) matrix analysis revealed that Yasooj is in a defensive (WT) position, requiring a review of management structures and the formulation of new operational plans.

Additionally, Yasooj's potential in tourism, agriculture, industry, and services has been identified as a key opportunity for sustainable development. To achieve sustainable development in Yasooj, urban management must revisit its structures and plans, focusing on enhancing inter-institutional cooperation and fostering greater citizen participation in decision-making processes.

Utilizing the city's natural, cultural, and social capacities, strengthening academic tourism through Yasooj University, hosting cultural festivals; and supporting agro-tourism are among the proposed solutions. Furthermore, Yasooj should aim to establish itself as a successful model of sustainable urban development by improving infrastructure, enhancing good urban governance, and fostering partnerships with the private sector and civil society.

Step-by-step and participatory planning, along with a thorough review of past strategies and the definition of clear visions, will contribute to achieving this goal. Establishing closer links between academic institutions and urban management will also play a key role in the successful implementation of development strategies.

Keywords: Sustainable Development, Yasooj City, SWOT Model, Urban Management, Tourism.

How to cite: salamatnia, A., balist, J., and Nahavandchi, M.: Assessment and Development of a Sustainable Development Strategy for Yasooj City Using the SWOT Model and SPACE Matrix, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3305, https://doi.org/10.5194/egusphere-egu25-3305, 2025.

Migration is one of human’s most drastic adaptation strategies against unfavorable conditions. With flows from and to origins and destinations, migration data are necessarily network data. Embedded within network data is interdependency among data points (flows) that renders some traditional statistical analyses, including causal inference techniques, inappropriate. To address this issue, we have developed a novel analysis, combining causal inference techniques with quadratic assignment procedure (QAP) to infer causal relationships from network data and applied it to the datasets that include migration flows and their potential drivers – these include socioeconomic, political, and environmental factors (e.g., flood and drought). We implemented this analysis for the African region data. The preliminary results are reported; the limitations and future work are discussed. We anticipate that this novel method will be applicable to a wide variety of network data in other fields.

How to cite: Muneepeerakul, R.: Causal linkages of human migration flow networks: A regional analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6756, https://doi.org/10.5194/egusphere-egu25-6756, 2025.

Extreme warming events in Texas have far-reaching environmental, economic, and societal consequences, including impacts on agriculture, energy demand, public health, and infrastructure. These events underscore the urgent need for reliable prediction systems that can anticipate their occurrence and inform mitigation and adaptation strategies. In this study, we develop machine-learning-based models to predict extreme temperature events across Texas by identifying and modeling the key drivers of these phenomena. The predictive framework incorporates the influences of large-scale climate modes and processes from both the Pacific and North Atlantic Oceans, including the El Niño–Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), Warm Pool (WP), and Atlantic Multidecadal Oscillation (AMO). By integrating these climate indices with regional atmospheric and surface data, the model captures the complex interactions between large-scale climate variability and regional temperature extremes. The contributions of each climate mode are quantified and analyzed to determine their relative importance in driving warming events across different temporal and spatial scales. To ensure the robustness of the predictions, the model outputs are further validated against physical mechanisms linking large-scale climate modes to atmospheric circulation patterns. This validation process provides a mechanistic understanding of the statistical relationships uncovered by the machine-learning models, ensuring that the predictions align with established climate dynamics. The findings from this study enhance our understanding of regional climate dynamics in Texas and demonstrate the potential of machine-learning approaches for improving the predictability of extreme temperature events.

How to cite: Chen, A. and Zhao, J.: Machine Learning-Based Prediction of Extreme Temperature Events in Texas: Understanding the Role of Large-Scale Climate Modes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7566, https://doi.org/10.5194/egusphere-egu25-7566, 2025.

Marine heat waves (MHWs) pose significant threats to coastal ecosystems, with particularly severe impacts in shallow waters where their magnitude is often amplified. The Chesapeake Bay, the largest estuary in the United States, is highly vulnerable to these events, which have increased in frequency and duration in recent decades. MHWs in the Chesapeake Bay have critical implications for its ecological balance, including effects on fish populations, habitat degradation, and water quality. Despite their growing prevalence, the underlying causes of these events and the factors regulating their variability remain poorly understood. Our study employs machine learning approaches to elucidate the drivers of marine heat waves in the Chesapeake Bay and to quantify their contributions to these extreme temperature events. By incorporating a comprehensive set of potential predictors, including local air temperature, wind forcing, river discharge, and Atlantic Ocean temperature, the model reveals the key mechanisms driving the onset, intensity, and persistence of MHWs in the Chesapeake Bay. Advanced feature selection techniques isolate the most relevant variables, while model outputs are validated against observed data to ensure accuracy and robustness. Our results suggest that local air temperature and ocean temperature anomalies from the Atlantic Ocean are dominant in triggering MHWs. These findings shed light on the complex interactions between atmospheric, hydrological, and oceanographic processes in shaping extreme thermal events in estuarine systems.

How to cite: Li, C., Xiong, N., and Zhao, J.: Understanding Marine Heat Waves in the Chesapeake Bay: Drivers, Variability, and Predictive Insights Using Machine Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7577, https://doi.org/10.5194/egusphere-egu25-7577, 2025.

The dynamic nature of Earth's lithosphere necessitates comprehensive tools for integrating geological data with plate tectonic frameworks across vast spatiotemporal scales. To address this challenge, EarthBank, in collaboration with the Earthbyte Group and Lithodat, has developed LithoPlates - a cloud-based deep-time reconstruction tool designed to support the visualisation and analysis of geological features within their paleogeographic contexts. LithoPlates leverages Earthbyte’s GPlates Web Service, enabling users to access pyGPlates functionalities and advanced plate tectonic models, offering researchers an intuitive platform for spatiotemporal analyses.

LithoPlates incorporates ten plate tectonic models, including the latest model published in 2024, which extends reconstructions back to 1.8 billion years. These models are seamlessly integrated into EarthBank’s public geochemistry data platform, enabling researchers to explore the tectonic settings and geological histories of their area of interest. By applying age-specific filters, users can visualise data within any chosen reconstruction timeslice within 1Ma steps, facilitating precise spatio-temporal analyses of geological processes such as formation, deformation, and material transport across Earth’s surface.

The platform’s dual capability to analyse data in both present-day and palinspastic geography significantly enhances its utility for geoscientific research. LithoPlates supports the reconstruction of geochronological and thermochronological data, providing a robust framework for investigating the evolution of Earth’s lithosphere. Its integration with EarthBank’s relational database further enables on-the-fly analysis of both data and metadata, offering real-time insights into complex geological systems. Robust export functionalities are also present including an open REST API, enabling users to seamlessly integrate their data and share results for further analysis.

Future advancements for LithoPlates include the integration of additional plate tectonic models, enhanced visualisation tools, and advanced filtering capabilities to refine comparative analyses across multiple reconstruction scenarios. These updates will improve uncertainty quantification, allow for more sophisticated model-data fusion, and facilitate the analysis of geophysical and geochemical datasets within a unified paleogeographic framework. 

LithoPlates represents a transformative tool for advancing Earth system reconstructions by addressing key challenges in the integration of geological, geophysical, and environmental data. Its interdisciplinary approach aligns with the broader scientific goal of developing digital twins of our planet, contributing to fields as diverse as resource exploration, paleoclimatology, and environmental risk assessment.

This tool exemplifies the potential of combining advanced modeling techniques with expanding geochemical and geophysical datasets, offering a scalable solution for analyzing the spatiotemporal evolution of Earth’s lithosphere. By providing access to comprehensive plate tectonic models and enabling precise spatiotemporal analyses, LithoPlates paves the way for groundbreaking research in understanding Earth’s dynamic geological history and its implications for modern and future challenges.

How to cite: Kohlmann, F., Noble, W., Qin, X., Higton, J., Beucher, R., Theile, M., McInnes, B., and Mueller, D.: Deep-Time Digital Twins: Integrating LithoPlates with the EarthBank Platform, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9058, https://doi.org/10.5194/egusphere-egu25-9058, 2025.

The mangrove ecosystem, a wetland forest found in tropical and subtropical coastal regions, is influenced by key factors such as tide height, salinity, precipitation, and temperature. This study focuses on understanding the impact of these factor’s decadal changes on three mangrove vegetation patches (Kozhikode, Ernakulam, and Kollam) on the southwestern coast of India. For this study, the recent past (2012-2022) rainfall and temperature data from the Indian Meteorological Department (IMD) and tide height data collected from INCOIS were used. Land surface temperature (LST) data based on MODIS and Enhanced Vegetation Index (EVI) and Salinity Index (SI) based on Landsat 8 were extracted using Google Earth Engine. The average annual rainfall at Kozhikode, Ernakulam and Kollam are 2934 mm, 3082 mm and 2305 mm, respectively.  The land surface temperature has an almost similar seasonal trend across all three mangrove sites, varying between 25 °C  to 31 °C in different seasons. The annual average tide height is observed to be highest at Kozhikode (0.97 m) and lowest in Kollam (0.49 m), whereas the annual average SI is observed to be highest in Kochi (0.13) and lowest in Kozhikode (0.10).

Evaluating vegetation changes using the EVI is essential for assessing the system’s effectiveness in protecting coastal areas from floods and guiding the planning of restoration and protective measures. The correlation coefficient between EVI and other climate variables was used to understand its impact on vegetation. Salinity, monsoon rainfall and summer LST are observed to be negatively correlated with the EVI in all three study areas. In contrast, during the southwest monsoon season, the tide height and LST positively correlate with EVI. This study revealed that optimum rainfall, salinity, and LST conditions are favourable for its growth, beyond which it negatively impacts vegetation compared to the rise in the tide height.

How to cite: Sudesan, S., Singh, U., Shyamrao Wable, P., Sasidharan, S., and Kalpuzha Ashtamoorthy, S.: Decadal Climate Variability and Its Impact on Mangrove Ecosystems of the Southwestern Coast, India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9225, https://doi.org/10.5194/egusphere-egu25-9225, 2025.

Future climate projection data are increasingly employed to evaluate the potential impacts of global warming across a wide range of domains, including meteorological variables (e.g., temperature and precipitation), hydrological processes, ecosystems, human health, and societal activities. The Coupled Model Intercomparison Project Phase 6 (CMIP6) provides an extensive dataset produced through international collaboration, incorporating multiple General Circulation Models (GCMs), diverse future scenarios, and numerous initial conditions. Despite the comprehensive nature of these datasets, most impact assessments rely on a limited subset of realizations, with no standardized methodology guiding their selection. This lack of consensus introduces potential biases into the outcomes of impact studies. This study quantitatively assesses the influence of realization selection on future climate impact assessments. Monthly precipitation and temperature data from CMIP6 were analyzed for both historical experimental periods and multiple Shared Socioeconomic Pathways (SSP) scenarios. Comparisons were conducted between outcomes obtained using all available realizations for each GCM and those derived from a single realization per GCM. Additionally, combinations of GCMs and realizations commonly used in prior studies were evaluated for their representativeness. The findings reveal that global average monthly precipitation is consistently higher when all realizations are utilized compared to scenarios based on a single realization. The inclusion of all realizations captures a broader range of variability, whereas subsets exhibit narrower variability and more localized trends. These results emphasize the significant impact of realization selection on future climate prediction outcomes. Moreover, an analysis of existing studies indicates that while selected datasets often reflect average trends, their overall representativeness requires further scrutiny. This research highlights the necessity of adopting uncertainty-aware methodologies in climate change studies. The findings offer valuable insights for improving the robustness and reliability of future climate impact assessments, paving the way for more informed decision-making in addressing climate change challenges.

How to cite: Nagata, K.: Quantitative analysis of the impact of realization selection on future climate change impact assessments using CMIP6 data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9648, https://doi.org/10.5194/egusphere-egu25-9648, 2025.

The paper investigates the potential of Artificial Intelligence (AI) and Machine Learning (ML) techniques, specifically Temporal Fusion Transformers (TFTs), to enhance the prediction of coastal dynamics along the Western Black Sea coast. We aim to bridge the gap between in-situ observations from five meteo-oceanographic stations and modelled geospatial marine data from the Copernicus Marine Service. TFTs are employed to refine predictions of shallow water dynamics by considering atmospheric influences, focusing on wave-wind correlations. Atmospheric pressure and temperature are treated as latitude-dependent constants, with specific investigations into extreme events like freezing and solar radiation-induced turbulence.  

The analysis utilizes a dataset of meteorological information collected by the Maritime Hydrographic Directorate (MHD) since 2015. The study relies on data gathered from seven automated weather stations at lighthouses along the Romanian coastline. The stations, part of the Romanian Navy - Marine Meteorological Surveillance Network, continuously gather meteorological parameters at specific ground-level heights, including wind speed and direction. The Copernicus Marine Service (CMEMS) wave reanalysis dataset for the Black Sea provides a comprehensive record of wave conditions with a spatial resolution of approximately 2.5 km and hourly temporal resolution.  

Explainable AI (XAI) is exploited to ensure transparent model interpretations and identify key influential input variables, including static, encoder, and decoder variables. Data attribution strategies address missing data concerns, while ensemble modelling enhances overall prediction robustness. The models demonstrate a significant improvement in prediction accuracy compared to traditional methods. This research provides a deeper understanding of atmosphere-marine interactions and demonstrates the efficacy of AI/ML in bridging observational and modelled data gaps for maritime safety and coastal management along the Western Black Sea coast.

Acknowledgements: The research of the M.E.M., P.P., M.G.I., and L.D. was conducted as part of the "Forecasting and observing the open-to-coastal ocean for Copernicus users" FOCCUS Project (https://foccus-project.eu/), funded by the European Union (Grant Agreement No. 101133911). Views and opinions expressed are, however, those of the author(s) only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HaDEA). Neither the European Union nor the granting authority can be held responsible for them. 

The presented results of the M.E.M., P.P., M.G.I., and L.D. have been carried out with financial support from the Sectorial Research-Development Plan of the Romanian Ministry of National Defence, PSCD 2021–2024 Project (097/2021, 092/2022, 097/2023, 097/2024): „Development of an integrated monitoring system to increase the quality of hydro-oceanographic data in the area of responsibility of the Romanian Naval Forces".
Thanks are extended to the relevant departments of INOE-2000 for their help through the "Core Program with the National Research Development and Innovation Plan 2022-2027" with the support of MCID, project no. PN 23 05/2023, contract 11N/2023.

How to cite: Mihailov, M. E., Ichim, M. G., Chirosca, A. V., Chirosca, G., Dutu, L., and Popov, P.: Temporal Fusion Transformers for Improved Coastal Dynamics Forecasting in the Western Black Sea , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9680, https://doi.org/10.5194/egusphere-egu25-9680, 2025.

The pressing demand for urbanised land due to the growing global population has led to increased land consumption, posing significant challenges to sustainable urban development. The European Union's No Net Land Take (NNLT) 2050 initiative aims to mitigate this issue by curbing urban expansion through urban densification or circular construction. Therefore, it is imperative to study the existing demand trajectory and their effects on the current development situation.

Wallonia, the southern region of Belgium, characterised by urban and peri-urban development, predominantly experience urban expansion. As a solution to that, government implemented various plans which focuses on revitalizing urban cores, addressing vacant buildings, and promoting the regeneration of central areas to prevent further urban sprawl. These ongoing urban pressures, makes it an ideal study area for conducting research on strategic planning.

In this study, we develop a Multinomial Logistic based Cellular Automata (MNL-CA) model calibrated using geophysical, accessibility, socioeconomic and spatial zoning data. Hereto, the model simulate futuristic urban growth until 2050 under two distinct scenarios:

• Business-As-Usual where urban growth continues following the historical demand trends within existing policies.
• Growth-As-Usual represents a scenario of latest observed built up demand trend along a constant rate .

The BAU scenario demonstrates a marked decline in urban expansion rates, stabilizing at 0 hectares per day by 2040. This trajectory reflects a shift toward densification and more spatially cohesive urban development. Meanwhile, the GAU scenario forecasts a sustained expansion rate of 2.51 hectares per day, resulting in a projected 49.20% increase in urban land by 2050. Together, these scenarios provide complementary insights: BAU serves as a valuable reference point for understanding controlled growth dynamics, while GAU offers a perspective for exploring the implications of constant expansion, thereby enhancing the robustness of future urban planning strategies.

While BAU offers a pathway aligned with policy goals, incorporating elements from GAU scenarios allows policymakers to "stress-test" urban strategies. This dual approach can enhance resilience and flexibility in urban planning, enabling better accommodation of future growth challenges while adhering to sustainability principles.

How to cite: Chakraborty, A., Mustafa, A., and Teller, J.: Monitoring the Impact of Urban Growth Scenarios on No Net Land Take of Wallonia, Belgium Using a Cellular Automata Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13665, https://doi.org/10.5194/egusphere-egu25-13665, 2025.

Central Italy experienced a catastrophic seismic sequence that suddenly started on 24 August 2016 at 1:36:32 UTC with an Mw = 6.0 earthquake. Buildings damaged by the shaking of this event caused about 300 fatalities, and several towns (e.g., Amatrice, Accumuli, Arquata del Tronto) were destroyed entirely. A seismic sequence started from this event, and the largest event occurred more than two months later on 30 October 2016 at 6:40:17 UTC with magnitude Mw = 6.5. On 18 January 2017, a resurgent of the seismic sequence occurred with four events of magnitude equal to or greater than 5.0 in a Southern sector of the interested region (close to Capitignano/Montereale/Campotosto Lake). Then, the sequence followed a typical multi-year decay. The impact was huge, and from an energetic point of view, the event of 30 October 2016 was one of the largest recorded in the last 40 years in Italy.

Considering this particular case study, we developed a multidisciplinary and multiparametric Jupyter Notebook which can be run, e.g. in a Virtual Research Environment (VRE). The Open Source Code and friendly environment of Jupyter Notebook permit future users to adopt the same VRE to study other earthquakes.

The Jupyter Notebooks retrieves data mainly from the European Plate Observing System (EPOS) platform (Bailo et al., 2023, https://doi.org/10.1038/s41597-023-02697-9), integrating with other sources such as climatological archives and Swarm magnetic satellites of European Space Agency (ESA). EPOS is a European research infrastructure devoted to understanding plate tectonics through multidisciplinary and multiparametric studies. EPOS has already implemented a portal (https://www.epos-eu.org/dataportal, last accessed 10 January 2024) where users can retrieve data grouped into 10 disciplines (Thematic Core Services – TCS).

The Italian seismic sequence interests the extensional plate typical of the Central Apennine Mount Chain, and multiparametric data can help to understand the physical and chemical processes that could occur before and during the earthquake. The VRE relies on the results published by (Marchetti et al., 2019) but using updated algorithms such as the one used to study the Arabian Plate earthquake doublets (Ghamry et al., 2024, https://doi.org/10.3390/atmos15111318). We will also include other atmospheric investigations of specific parameters (e.g., Piscini et al., 2017, https://doi.org/10.1007/s00024-017-1597-8). Such previous studies propose evidence for anomalies in the organised chain of lithosphere, atmosphere, and ionosphere that were identified before the Italian seismic sequence 2016-2017.

These preliminary studies contribute to investigating the relations between geo-layers in our Earth’s system and the influence of seismic activity on them. Furthermore, this VRE adds a tool to the EPOS platform with potentially several applications, such as investigations of other significant earthquakes or other natural hazards, such as volcano eruptions.

How to cite: Marchetti, D., Bailo, D., Michalek, J., Paciello, R., and Falcone, G.: A Jupyter Notebook devoted to a multiparametric investigation of the Amatrice-Norcia Italian seismic sequence 2016-2017, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13827, https://doi.org/10.5194/egusphere-egu25-13827, 2025.

The Port of Manzanillo, Colima, serves as a pivotal infrastructure for Mexico’s international trade network. In response to increasing operational demands and advance modernisation efforts, an ambitious expansion into the Laguna de Cuyutlán has been proposed. This initiative includes the construction of specialised container terminals and supporting infrastructures. However, the area’s vulnerability to geological and hydrometeorological hazards—such as earthquakes, volcanic activity, tsunamis, landslides, and tropical storms—raises critical concerns regarding the durability and sustainability of these developments.

This study introduces a hybrid risk management approach that combines the principles of the PMBOK’s plan risk management methodology with the analytical precision of fuzzy set theory. Comprehensive historical data on natural hazards were systematically gathered from risk atlases, scientific research, and official reports. The model applies fuzzy membership functions to evaluate the likelihood and impact of risks. Additionally, tools like fuzzy Delphi, fuzzy DEMATEL, and fuzzy ANP facilitate the structured analysis and prioritisation of potential threats.

The primary aim is to create a robust system for addressing the uncertainties associated with complex risk environments. By integrating advanced analytical methods with established risk management practices, the model provides a foundation for designing effective mitigation strategies. These measures are essential for maintaining operational reliability, enhancing infrastructure resilience, and minimising socio-economic impacts. This research highlights the value of interdisciplinary methodologies that link scientific advancements with practical solutions, tackling the intricate challenges posed by climatic and geological extremes in dynamic contexts.

How to cite: Cardenas, E., Delgadillo-Partida, J., Mendoza-Rosas, A. T., and Zarate-Ramirez, F.: Comprehensive Risk Assessment at the Port of Manzanillo: A Model Based on PMBOK and Fuzzy Logic., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13886, https://doi.org/10.5194/egusphere-egu25-13886, 2025.

Shipping remains a crucial element of global trade and commerce, facilitating over 90% of international trade by volume. The maritime industry’s advanced logistics chains are vital for the timely delivery of goods, supporting both economic growth and employment. However, it is also a significant source of pollution, accounting for approximately 3% of global greenhouse gas (GHG) emissions, and contributing 13% of nitrogen oxides (NOx) and 12% of sulfur oxides (SOx). Additionally, shipping emits harmful pollutants, including particulate matter (PM), black carbon (BC), and methane (CH4). These emissions not only impact the global climate but also pose severe health risks to communities near shorelines, contributing to asthma, respiratory and cardiovascular diseases, lung cancer, and premature death.

The International Maritime Organization (IMO) is actively engaged in mitigating these environmental impacts as part of its support for the UN Sustainable Development Goal 13, which addresses climate change in alignment with the 2015 Paris Agreement. The IMO has implemented several regulations to curb GHG emissions from shipping, beginning with mandatory energy efficiency measures introduced on July 15, 2011. Subsequent regulations include the Initial IMO GHG Strategy (2018) and the updated Strategy on Reduction of GHG Emissions from Ships (2023). The 2023 strategy sets ambitious targets to achieve near-zero GHG emissions from international shipping by around 2050, with interim goals of reducing emissions by at least 20% by 2030 and 70-80% by 2040. It also aims to cut the carbon intensity of international shipping by at least 40% by 2030, measured as CO2 emissions per unit of transport work. As of January 1, 2023, ships are required to calculate their Energy Efficiency Existing Ship Index (EEXI) and establish an annual operational Carbon Intensity Indicator (CII), with ratings from A to E indicating energy efficiency (International Maritime Organization).

In response to evolving regulations aimed at reducing GHG emissions, we propose a machine learning framework to improve emission predictions, with a particular focus on the Saint Lawrence River. Currently, emissions in the Canadian shipping sector are calculated a posteriori, with Environment and Climate Change Canada (ECCC) providing a national marine emissions inventory and a comprehensive visualization tool. This tool enables users to analyze shipping activities and emissions across Canada by filtering data through various parameters.

Our proposed work is designed to predict GHG emissions for vessels navigating the Saint Lawrence River, with plans for broader application across Canada. By employing a bottom-up methodology, we create a detailed emissions inventory based on individual vessel activities, leveraging Automatic Identification System (AIS) data to capture the spatiotemporal dynamics of shipping (Spire). To enhance accuracy, we incorporate vessel-specific information from CLARKSONS, including engine type, fuel type, and power, along with meteorological data such as current speed to account for external factors affecting emissions. Machine learning models, particularly deep learning techniques, are employed in the prediction phase, enabling the model to continually improve with new data. This scalable approach not only enhances environmental monitoring but also supports national efforts to reduce GHG emissions from marine transportation across Canada.

How to cite: El aissi, A. and Benabbou, L.: Predicting GHG Emissions in Shipping: A Case Study Of Canada, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14016, https://doi.org/10.5194/egusphere-egu25-14016, 2025.

The maritime transport industry faces a significant challenge: reducing its greenhouse gas (GHG) emissions by 50% compared to 2008 levels. A crucial factor in calculating and optimizing these emissions is accurately predicting ship speed through water. While various models exist, few effectively combine both physical principles and machine learning approaches, leading to limitations in prediction accuracy.

The paper proposes a hybrid model with two main components: ''Solve'' Component: A physics-based approach that uses a Physics-Informed Neural Network (PINN) to determine the theoretical speed a ship would achieve in calm water conditions, based on fundamental physical principles and equations. ''Predict'' Component: A data-driven approach that takes the theoretical calm water speed and adjusts it based on real-world conditions using machine learning algorithms, producing actual speed predictions.

The Solve Phase centers around a differential equation relating three key parameters: propulsion power (P), draft (T), and speed through calm water (V_w), the equation takes the form:

The model uses a PINN to solve a differential equation that links propulsion power (P), draft (T), and calm water speed (V_w) to generate initial speed estimates. The PINN uses a loss function that incorporates both initial conditions and differential equation residuals. A major challenge arises because V_w is theoretical and cannot be directly measured. This issue is addressed using historical data by identifying periods when sea conditions were calm to use as training data.

The model creates a bridge between its solve and predict phases. In the first approach, focused on training data generation, the system utilizes the trained PINN to generate collocation points. From these points, it creates training triplets consisting of propulsion power (P_i), draft (T_i), and calm water speed (V_wi). This approach uses a straightforward mean squared error loss function to train the neural network. The second approach takes a different path by using propulsion power (P) and draft (T) as direct inputs to the neural network. What makes this approach unique is that it incorporates the PINN directly into the loss function. This integration allows physical principles from the differential equation to directly influence the predictions, creating a stronger connection between the physical model and the machine learning component.

The predict phase begins by taking the calm water speed predictions generated from the solve phase and enhances them by incorporating various real-world factors that affect ship movement. These factors include maritime conditions, meteorological data, and current conditions, providing a comprehensive view of the actual sailing environment. To process this combined data, we use machine learning algorithms such as Xgboost. The final output of this phase is the real speed through water (V_wr), which represents a more realistic prediction that accounts for all environmental factors affecting the ship's speed.

The model offers a groundbreaking approach to maritime speed prediction by generalizing across vessel types and integrating physical principles with machine learning. By incorporating operational and meteorological data, it provides more accurate speed predictions that optimize fuel consumption and support the maritime industry's greenhouse gas emission reduction goals, bridging environmental protection with operational efficiency.

How to cite: Atanane, A., Elmimouni, Z., and Benabbou, L.: A Hybrid Machine Learning Model For Ship Speed Through Water: Solve And Predict, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14039, https://doi.org/10.5194/egusphere-egu25-14039, 2025.

The ongoing deglaciation driven by global warming and climate change has led to the occurrence of extreme weather and climate events including heatwave, cold wave, flash floods, cloud burst, GLOFs, etc. This resulted in the formation and expansion of numerous glacial lakes, particularly in the High Mountain Asia (HMA) region. Many of these lakes are at high risk of Glacial Lake Outburst Floods (GLOFs), which can release millions of cubic meters of water and debris, causing extensive damage to lives, property, infrastructure, agriculture, and livelihoods in remote and economically vulnerable downstream communities in Pakistan. This study focuses on Azad Jammu and Kashmir (AJK) in Northern Pakistan to assess GLOF risk using multi-source data. Several vulnerable sites from the Pakistan Meteorological Department’s GLOF inventory were analyzed, seven of which are highly susceptible to GLOFs. A spatio-temporal analysis of these sites considered critical factors such as lake area and volume changes, elevation, slope, aspect, temperature and precipitation patterns, land use and land cover (LULC) changes, glacier and snow cover loss, proximity to fault lines, and impact zones through geospatial techniques, GIS analysis and cloud computing Google Earth Engine (GEE) platform. Results indicate a significant decline in snow and glacier cover, coupled with an increase in land surface temperatures (LST), contributing to accelerated melting and heightened GLOF and flash flood occurrences. The study also estimates potential impacts on population, infrastructure, schools, forests, agriculture, and water quality in the Neelum Valley of AJK. The findings offer valuable insights for policymakers and disaster management authorities to devise targeted and effective risk mitigation strategies.

How to cite: Fatim Ali, S. S., Hussain Shah, S. W., and Rehman, S.: Leveraging Geospatial Techniques to Appraise the Potential Implications on Vulnerable GLOF Sites in High Mountain Asia: A Case Study of Azad Jammu and Kashmir, Northern Pakistan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15239, https://doi.org/10.5194/egusphere-egu25-15239, 2025.

Dust storms pose significant environmental challenges in arid and semi-arid regions, causing serious health, environmental, and socio-economic impacts. Traditional dust modeling approaches, like numerical methods, often struggle to balance accuracy, computational efficiency, and data availability. This study employed a Physics-Informed Neural Network (PINN) model for event-based dust storm modeling, integrating the physical principles of dust dynamics with data-driven methods. We demonstrated the applicability of the above framework in the Lake Urmia Basin, where the lake desiccation and external dust sources have triggered local dust storms. To this end, we first analyzed ground-recorded PM₁₀ and weather data to identify dusty days between 2004 and 2019. Next, we trained an initial neural network (NN) model with remote sensing data that describe meteorological and boundary layer characteristics at the locations of pollution monitoring stations. This approach allowed us to generate gridded PM₁₀ data, overcoming the limitations posed by insufficient and non-continuous data for directly training PINN. Finally, the PINN model was trained and validated on 21 selected dust events from three stations chosen for their spatial distribution and sufficient availability of PM₁₀ data throughout the events. Analysis revealed that the initial NN model achieved R² of 62% and mean absolute error (MAE) of 65  on the test data. The PINN model demonstrated substantial improvement with mean R² of 93% and mean MAE of 9  on the gridded PM₁₀, and MAE of 39  when validated against ground observations. Furthermore, the model yielded lower prediction accuracy in urban compared to rural stations, which is attributed to the bias imposed by the influence of terrestrial and industrial pollutions. This study demonstrates the effectiveness of PINNs in tackling dust transport modeling challenges in data-sparse regions, providing a novel way to combine physical principles with data-driven techniques for large-scale environmental applications.

How to cite: giahchin, K. and danesh-yazdi, M.: Event-Based Physics-Informed Neural Networks for Dust Storm Prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15382, https://doi.org/10.5194/egusphere-egu25-15382, 2025.

Different types of water bodies, such as streams, creeks, irrigation ponds, and paddies, form networks in low-elevation mountainous areas, referred to as Interconnected Small Water Bodies in Slopeland (ISWBS) in the context of Taiwan. Little is known about the ecosystem functions and conservation potential of ISWBS, and an assessment framework is proposed using an integration of remote sensing and field survey data. We analyze whether network characteristics, node characteristics, and landscape factors impact ecological functions and estimate the services related to sediment reduction, agricultural production, and biodiversity that ISWBS provides.

In this preliminary study, we focused on irrigation and natural ponds as important nodes within ISWBS. Monitoring stations were established to record the micro-climate factors of the ponds, and surveys of benthic macro-invertebrates were conducted in 2024. Using a framework of functional feeding groups, the ponds are categorized based on the relative abundance of collector-gatherers, which significantly affect the results of ordination. Community analysis shows little and non-significant relationships between community composition and environmental factors, namely variations in water depth, landscape indices, and irrigation use. However, some factors, such as water depth variation during low depth periods, total edge length, and canal connection, show potential to contribute to future analyses.

Regarding the remote sensing analysis, we find that the distance between nodes has decreased over the past 40 years. Nevertheless, no biodiversity records are available to determine the impact of landscape change. The effects of changing network characteristics on community composition and functional groups are unclear due to insufficient sampling of biodiversity data. Further biodiversity sampling and the study of network characteristics are critical to determine how ISWBS functions in ecosystem processes, especially for sediment detention and nutrient cycling at a landscape scale.

How to cite: Hsieh, C.-K., Chen, S.-C., and Liang, M.-C.: Assessment Framework of Ecosystem Services and Functions of Interconnected Small WaterBodies in Slopeland, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15571, https://doi.org/10.5194/egusphere-egu25-15571, 2025.

Environmental challenges have had a negative impact on African forest resources, which has subsequently adversely affected some ecosystem services that are required for the survival of people. We conducted a comparative study in the wet and dry woodlands in Zambia to establish the formation of tree growth rings and determine the relationship between the growth ring width and rainfall. Through the four successful Africa Dendrochronological Fieldschools that were conducted from 2021 to 2024, we collected samples from the wet miombo woodlands on the copperbelt province and the dry miombo and Baikiaea woodlands on the southern province of Zambia. From 2021 to 2023, we recorded 49 tree species from the wet miombo woodlands and found that the Fabaceae family plants had the highest species richness with 28.5%. We determined a series intercorrelation of 0.45 and average mean sensitivity of 0.465 from a master chronology of 14 tree species. The dendroclimatic study found a significant positive relationship (r-value =0.589, p-value = 0.0005) between ring width of a mixed species chronology of Brchaystegia longifolia and Julbernadia paniculata, and precipitation totals for Zambia’s wet season (October–April). In 2024, studies were conducted in the dry miombo and Baikiaea woodlands. Through this study, 16 distinct species were identified in the Baikiaea woodlands with Baikiaea plurijuga being the abundant species. We determined series intercorrelation of 0.31 and an average mean sensitivity of 0.50 from a mixed tree species from the Baikiaea woodlands. A precipitation correlation with Brachytegia longifolia from the miombo woodlands found that previous December and Current March precipitation have positive influence on tree growth. In both, dry and wet woodlands, we found that trees produce annual growth rings that are responsive to seasonal climate, and are useful for dendrochronology

How to cite: Ngoma, J. and the Justine Ngoma: A comparative study of the dendroclimatic potential of selected tree species of the tropical dry and wet woodlands of Zambia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16432, https://doi.org/10.5194/egusphere-egu25-16432, 2025.

The laurel forest (laurisilva) represents a unique and biodiverse ecosystem currently confined to subtropical regions with specific climatic conditions. In the Canary Islands, these laurisilva forests, constrained to areas with high humidity and stable temperatures as the northern slopes of Tenerife, La Gomera, and La Palma, are of particular ecological importance hosting numerous endemic species. However, climate change poses a significant threat to these fragile habitats, with potential shifts in their distribution at both regional and global scales with new regions emerging as potential refuges for laurisilva forests. The main scope of this study is to explore the current distribution of laurisilva forests in the Canary Islands and projects to the future under different climate change scenarios for mid-century and end-century, its potential range in other archipelagos of Macaronesia and selected regions worldwide with similar climatic conditions.

Using Maxent as the primary modeling tool, we first trained the model by means of high-resolution bioclimatic indicators specifically designed for the Canary Islands, the so-called BICI-ULL dataset. This dataset was generated taking into account the intricate topography and diverse microclimatic patterns of the archipelago, providing a robust framework to delineate the current distribution of laurisilva. Once the model was trained, we used the global bioindicators from WorldClim and Chelsa to project the potential future distribution of laurisilva.

Thus, this methodology based on BICI-ULL allowed us to develop a detailed understanding of laurisilva distribution in the Canary Islands, while WorldClim and Chelsa facilitated the extrapolation of projections to broader geographic scales offering a framework for identifying potential refugia and new habitats for conservation planning of the laurisilva forests. These findings underline the importance of combining regional expertise with global datasets to inform conservation strategies for biodiverse but threatened ecosystems like laurisilva.

How to cite: Sosa-Guillén, P., Simon Tondreau, P., Barragán, R., González, A., Pérez, J. C., Expósito, F. J., and Díaz, J. P.: Modeling the Future of Laurisilva Forests: Integrating Regional and Global Bioclimatic Datasets for Projections Beyond the Canary Islands, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18096, https://doi.org/10.5194/egusphere-egu25-18096, 2025.

Abstract

Climate Change Impacts in the Transboundary Prespa-Ohrid watershed.

(The strategic impact identified in the SWOT analysis)

Brisilda Stafa¹; Emanuela Kiri¹

¹ Institute of Geosciences, UPT, Tirana, Albania. 

This study investigates the impacts of climate change on open surface water in the transboundary Prespa-Ohrid Lake area using a SWOT analysis based on lake level and meteorological data. Rising temperatures, altered precipitation patterns, and increased evaporation rates have been identified as critical factors influencing water levels, particularly in Prespa Lake. The SWOT framework helps in systematically evaluating the internal strengths and weaknesses of the region’s hydrological systems and external opportunities and threats posed by climate change.

Strengths include the availability of long-term lake level and meteorological data, which provide a robust foundation for assessing climate impacts and water management strategies. Weaknesses highlight the vulnerability of shallow lakes like Prespa to evaporation and reduced inflows, compounded by inconsistent monitoring across national borders. Opportunities lie in the potential for enhanced regional cooperation, ecosystem-based adaptation, and the development of sustainable water management policies that account for climate variability. However, the region faces significant Threats, including further reductions in water availability, degradation of water quality, loss of biodiversity, and socio-economic impacts on agriculture and tourism.

The study emphasizes the need for transboundary cooperation and adaptive strategies to mitigate these risks, with a focus on integrating climate and water data to guide future decision-making in the region.

Keywords: climate change, transboundary watershed, SWOT analysis, socio–economic impact. 

How to cite: Stafa, B. and Kiri, E.: Climate Change Impacts in the Transboundary Prespa-Ohrid watershed. (The strategic impact identified in the SWOT analysis), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18927, https://doi.org/10.5194/egusphere-egu25-18927, 2025.

Stratospheric aerosol injection (SAI) is a proposed climate intervention that involves injecting aerosols (or aerosol precursors) into the stratosphere to reduce global warming and associated devastating impacts. In this study, I estimate the socioeconomic effects of future SAI using model results from the Stratospheric Aerosol Geoengineering Large Ensemble (GLENS-SAI) and the Assessing Responses and Impacts of Solar Climate Intervention on the Earth System (ARISE-SAI)  as inputs to the Climate Framework for Uncertainty, Negotiation, and Distribution Integrated Assessment model (FUND). GLENS-SAI and ARISE-SAI are an ensemble of SAI simulations between 2020 and 2100 (GLENS) and 2035-2064 (ARISE-SAI-1.5) using the Community Earth System Model, wherein SAI is simulated to offset the warming produced by a high-emission scenario (RCP 8.5) and a middle of the road (SSP2-4.5). FUND's components include agriculture, forestry, heating, cooling, water resources, tropical and extratropical storms, biodiversity, cardiovascular and respiratory mortality, vector-borne diseases, diarrhea, migration, morbidity, and rising sea levels. These aggregate impacts culminate in net damages, calculated as a percentage of gross domestic product (GDP). In both emission scenarios, global damages take a more linear trend in time, with up to 1% of global GDP loss under SSP2 - 4.5, as opposed to 6% under RCP8.5 (Figure 1). Under GLENS and ARISE SAI, damages follow a beneficial pathway, resulting in up to 0.6% and 1% savings of global GDP, respectively (Figure 1). Significant aspects of net damages include cooling and heating demand, agriculture, and water resources. Whereas cooling costs rise under both warming scenarios, savings accrue from avoided heating costs. However, SAI elicits the opposite effect. Additionally, the Dynamic Integrated Climate-Economy model, a neoclassical IAM, was tailored similarly to give further insight into damages. A nonlinear regression approach was then applied to climate and economic data to validate the results from the integrated assessment models. Finally, a cost-benefit analysis was performed on the GLENS and ARISE scenarios using operational and deployment cost estimates from Wagner and Smith (2018). SAI benefits (savings) are more than sufficient to cover the costs of operation and deployment. Even in the extreme case (GLENS-SAI), cost peaks at around 0.03% of global GDP (Figure 2). This analysis will be pivotal in advising policymakers on the economic outcomes and feasibility of SAI. 

Figure 1 ( Damages as a percentage of global GDP. Left: SSP2-4.5 and ARISE-SAI. Right: RCP8.5 and GLENS-SAI)

Figure 2 (SAI costs as a percentage of Global GDP. Blue: ARISE-SAI, Yellow: GLENS-SAI)

References

Smith, W., & Wagner, G. (2018). Stratospheric aerosol injection tactics and costs in the first 15 years of deployment. Environmental Research Letters, 13(12), 124001.

How to cite: Ansah, P.: Leveraging Integrated Assessment Models to Assess Socioeconomic Impacts of Potential Stratospheric Aerosol Injection, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19957, https://doi.org/10.5194/egusphere-egu25-19957, 2025.

For understanding localized hydrological and climatological processes, downscaling gridded precipitation data to finer spatial resolutions is a crucial prerequisite. For a densely populated country like India, accurate downscaled data is crucial for building resilience to climate change impacts, supporting adaptation efforts, and enhancing disaster management. In recent years, deep learning (DL) has emerged as a powerful tool for advancing Earth system modelling and climate data downscaling. This study presents a comprehensive intercomparison of deep learning architectures, for downscaling precipitation across India. A few efficient DL architectures from recent studies are chosen for intercomparison such as simple dense, simple convolutional neural network, Fast Super Resolution Convolutional Neural Network (FSRCNN), Super Resolution Deep Residual Network (SRDRN), U-Net, and Nest-U-Net. The experiments are designed in synthetic style by using coarsened ECMWF Reanalysis version 5 (ERA5; 1^ox1^o) daily variables as the inputs and high-resolution Indian Monsoon Data Assimilation and Analysis reanalysis (IMDAA; 0.12^ox0.12^o) daily precipitation as training labels and benchmarks for the evaluation. Training and validation are conducted for the period 1980-2014, afterwards the trained models are evaluated on data from 2015-2020. To reduce the biases induced by the highly positive-skewed precipitation data and to enhance the model performance on extreme events, a weighted mean absolute error is implemented for training. The performance of the DL models is also compared with the Bias Correction and Spatial Disaggregation (BCSD), a renowned statistical downscaling method. The results indicate that all deep learning DL models outperformed the BCSD method. Among the DL models, U-Net and Nest-U-Net demonstrated superior performance in capturing fine-scale precipitation patterns and extreme precipitation events, owing to their encoder-decoder architecture, which effectively learns spatial features at different scales. In contrast, the FSRCNN and SRDRN produced results with slightly lower precision than the U-Net models, but at a significantly reduced inference time, making them more efficient for faster data generation. The findings underscore the potential of deep learning for improving regional precipitation downscaling across India, offering a promising alternative to traditional statistical methods like BCSD in handling complex, non-linear relationships inherent in climate data.

How to cite: Murukesh, M. and Kumar, P.: Comparative analysis of deep learning architectures trained for downscaling gridded precipitation across India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-676, https://doi.org/10.5194/egusphere-egu25-676, 2025.

Renewable resources are strongly dependent on local and large-scale weather situations. Skillful subseasonal to seasonal (S2S) forecasts -beyond two weeks and up to two months- can offer significant socioeconomic advantages to the energy sector. In particular, accurate wind speed forecasts result in optimized generation of wind-based electric power. This study aims to enhance wind speed predictions using a diffusion model with classifier-free guidance to downscale S2S forecasts of surface wind speed. We propose DiffScale, a diffusion model that super-resolves spatial information for continuous downscaling factors and lead times. Leveraging weather priors as guidance for the generative process of diffusion models, we adopt the perspective of conditional probabilities on sampling super-resolved S2S forecasts. We aim to directly estimate the density, associated with the target S2S forecasts at different spatial resolutions and lead times without auto-regression or sequence prediction, resulting in an efficient and flexible model. Synthetic experiments were designed to super-resolve wind speed S2S forecasts from the European Center for Medium-Range Weather Forecast (ECMWF) from a coarse resolution to a finer resolution of data from ERA5, which serves as a high-resolution target, derived from reanalysis data. We achieve a significant increase in the quality of predictions, utilizing the proposed diffusion model for continuous downscaling and bias correction of the ECMWF forecasts.

How to cite: Springenberg, M., Otero Felipe, N., and Ma, J.: DiffScale: Towards Continuous Downscaling and Bias Correction in Subseasonal Wind Speed Forecasts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1650, https://doi.org/10.5194/egusphere-egu25-1650, 2025.

Understanding the complex relationship between trace gases as well as undestanding various source and sink pathways in the atmsophere need good qualtity continuous and reliable datasets. However, obtaining comprehensive long-term profiles for key trace gases is a significant challenge. We have initiated a new research strand to consrtuct  long term data using machine learning. Output from a  Chemical Transport Model (CTM) and observational data from satellite instruments (such as HALOE and ACE-FTS) is merged using machine learning. This integration results in the creation of daily, gap-free datasets for six crucial gases: ozone (O3), methane (CH4), hydrogen fluoride (HF), water vapour (H2O), hydrogen chloride (HCl), and nitrous oxide (N2O) from 1991 to 2021.

Chlorofluorocarbons (CFCs) are a critical source of chlorine that controls stratospheric ozone losses. Currently, ACE-FTS is the only instrument that provides sparse but daily measurements of these gases. Monitoring changes in these ozone-depleting substances, which are now banned, helps assess the effectiveness of the Montreal Protocol. We have initiated the construction of gap-free stratospheric profile data for CFC-11 as a subsequent step.

We use a regression model to estimate the relationship between various tracers in a CTM and the differences between the CTM output field and the observations, assuming all errors are due to the CTM setup. Once the regression model is trained for observational collocations, it is used to estimate biases for all the CTM grid points. To enhance accuracy, we employed various regression models and found that XGBoost regression outperforms other methods. ACE-FTS v5.2 data (2004-present) is used to train (70%) and test (30%) the XGBoost performance.

Our results demonstrate excellent agreement between the constructed profiles and satellite measurement-based datasets. Biases in TCOM data sets, when compared to evaluation profiles, are consistently below 10% for mid-high latitudes and 50% for the low latitudes, across the stratosphere. The constructed daily zonal mean profile datasets, spanning altitudes from 15 to 60 km (or pressure levels from 300 to 0.1 hPa), are publicly accessible through Zenodo repositories.

     CH4:       https://doi.org/10.5281/zenodo.7293740   
     N2O:          https://doi.org/10.5281/zenodo.7386001
     HCl :         https://doi.org/10.5281/zenodo.7608194
     HF:        https://doi.org/10.5281/zenodo.7607564
     O3:         https://doi.org/10.5281/zenodo.7833154 
     H2O:          https://doi.org/10.5281/zenodo.7912904
     CFC-11:    https://doi.org/10.5281/zenodo.11526073  
     CFC-12:      https://doi.org/10.5281/zenodo.12548528
     COF2:        https://doi.org/10.5281/zenodo.12551268

In an upcoming iteration, we are enhancing the algorithm as well as add more species in the current setup. We believe these data sets would provide valuable insights into the dynamics of stratospheric trace gases, furthering our understanding of their behaviour and impact on the climate.

References:

Dhomse, S. S., et al.,: ML-TOMCAT: machine-learning-based satellite-corrected global stratospheric ozone profile data set from a chemical transport model, Earth Syst. Sci. Data, 13, 5711–5729, https://doi.org/10.5194/essd-13-5711-2021, 2021.

Dhomse, S. S. and Chipperfield, M. P.: Using machine learning to construct TOMCAT model and occultation measurement-based stratospheri
c methane (TCOM-CH4) and nitrous oxide (TCOM-N2O) profile data sets, Earth Syst. Sci. Data, 15, 5105–5120, https://doi.org/10.5194/essd-15-5105-2023, 2023.

How to cite: Dhomse, S. and Chipperfield, M.: Machine Learning to Construct Daily, Gap-Free, Long-Term Stratospheric Trace Gases Data Sets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2394, https://doi.org/10.5194/egusphere-egu25-2394, 2025.

Analogue has been a widely-used concept in atmospheric science, particularly useful in weather forecasting and climate-related studies. The underlying idea is straightforward. An analogue is identified by determining its level of similarity to a reference weather or climate condition, traditionally, via computing a Euclidean distance. Recently, a deep-learning based framework, called ClimaDist, was proposed for climate analogue identification, found to outperform traditional Euclidean distance metrics. Despite the promising performance, similarly to many deep-learning models, it is challenging to estimate the uncertainty of the analogue searching process undertaken by ClimaDist. This hinders its applicability to real-world operations, especially for those requiring decision making.

To address this challenge, this study extends the capabilities of ClimaDist through incorporating a uncertainty quantification method, together with explainable AI (XAI) techniques. Specifically, the Evidential Deep Learning (EDL) approach is applied to the analogue searching process undertaken by the ClimaDist. This enables effective quantification of the uncertainty associated with data and model, respectively, while exploring their relationship with overall model performance. Two distinct scenarios are applied to these two models using data that were seen and unseen during the training processing. 

An experiment has been designed to verify the proposed approach using ERA5 data over a square domain centred at the Nettebach (Germany) covering the geographic range of 55°N to 47°N and 3°E to 11°E. Two ClimaDist models, one with the best validation performance and the other one best training performance, respectively, are used for comparison. These models are assessed based on the similarity of the found analogues and via under two distinct scenarios –with input data seen and unseen during the training process, respectively. Preliminary results suggest that the integration of uncertainty quantification enhances the interpretability and reliability of analogue identification, enabling improved downstream applications. Specifically, high model uncertainty can be highlighted by the proposed approach while fully unseen data is used as input. This not only provides valuable insight in knowing the capacity of the underlying model but also allows the optimization of resource usage.

How to cite: Chen, C., Chen, P.-C., Tseng, C.-Y., and Wang, L.-P.: Uncertainty quantification through the climate analogue identification process by ClimaDist, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3610, https://doi.org/10.5194/egusphere-egu25-3610, 2025.

Databases of high-resolution interpolated climate data are essential for analyzing the impacts of past climate events and for developing climate change adaptation strategies for managed and natural ecosystems.  To enable such efforts, we contribute an accessible, comprehensive database of interpolated climate data for Africa that includes monthly, annual, decadal, and 30-year normal climate data for the last 120 years (1901 to present) as well as multi-model CMIP6 climate change projections for the 21st century. The database includes variables relevant for ecological research and infrastructure planning, and comprises more than 25,000 climate grids that can be queried with a provided ClimateAF software package. In addition, 30 arcsecond (~1km) resolution gridded data, generated by the software, are available for download (https://tinyurl.com/ClimateAF). The climate grids were developed with a three-step approach, using thin-plate spline interpolations of weather station data as a first approximation, subsequent fine-tuning with deep neural networks to capture medium-scale local weather patterns, and lastly dynamic lapse-rate based downscaling to a user-selected resolution, or to scale-free point estimates with the ClimateAF software package. The study contributes a novel deep learning approach to model orographic precipitation, rain shadows, lake and coastal effects, including the influences of wind direction and strength. The climate estimates were optimized and cross-validated with a checkerboard approach to ensure that training data was spatially distanced from validation data. We conclude with a discussion of applications and limitations of this database.

How to cite: Namiiro, S., Hamann, A., Wang, T., Castellanos-Acuña, D., and Mahoney, C.: Climate data interpolation with deep neural networks: a comprehensive dataset of historical and future climate for Africa, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4595, https://doi.org/10.5194/egusphere-egu25-4595, 2025.

The use of machine learning (ML) for climate science has attracted considerable attention within the last few years. A number of recent studies have used ML to extract information from global climate data (e.g. regional downscaling), predict future states of the climate system and evaluate models against observations. In particular, Brunner and Sippel (2023) showed that low-resolution global climate models and observations can reliably be distinguished based on the global distribution of daily temperature, even after removing the mean model bias. ML is thus able to isolate fundamental differences between models and observations even in the presence of substantial internal variability. This raises the questions of whether ML can also distinguish between model and observational data on a regional scale, whether ML is as successful for km-scale models as for coarse-resolution models, and whether more complex bias correction methods reduce the success of ML.

To answer these questions, we use daily temperature fields over Austria, a topographically very complex domain. As training data, we use 200 different, randomly drawn days from each of the 13 ÖKS15 bias-corrected EURO-CORDEX models with an output resolution of 1km, resulting in 2600 samples labeled “model” which are matched by the same number of random days labeled “observation” from the SPARTACUS observation dataset. We use the binary classification approach to distinguish between the two classes of models versus observations. A logistic regression classifier is trained to determine the probability that a daily temperature field belongs to one of the two classes. In order to evaluate the ML algorithm subsequently, all days from the out-of-sample 10-year period 2005-2014 are used as test data.

The ML algorithm succeeds in correctly identifying the overwhelming majority of the test data for the setup used, resulting in an accuracy of 99%. The  results remain consistent even when a different sample of 2x2600 random training days is used. In contrast to more complex classifiers, such as a convolutional neural network (CNN), the learned coefficients from the logistic regression allow insights into the spatial patterns that are crucial for distinguishing between models and observations. While the performance of climate models is typically evaluated on climatological timescales, our results highlight that such classifiers can be used to identify patterns of structural model biases. Our method hence offers a computationally efficient approach for model evaluation, especially when handling km-scale climate model data on a regional domain.

References:
Brunner L. and Sippel S. (2023): Identifying climate models based on their daily output using machine learning, Environmental Data Science, https://doi.org/10.1017/eds.2023.23

How to cite: Meindl, M., Voigt, A., and Brunner, L.: Using machine learning to distinguish km-scale climate models and observations on a regional scale, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6462, https://doi.org/10.5194/egusphere-egu25-6462, 2025.

North Atlantic sea surface temperatures (NASST), particularly in the subpolar region, exhibit some of the highest predictability across global oceanic systems. However, the relative contributions of atmospheric versus oceanic influences on the long term NASST variability remains ambiguous. In this study, we utilize neural networks (NNs) to assess the significance of various atmospheric and oceanic predictors in forecasting the state of NASST within the CANARI Large Ensemble, which employs the Met Office CMIP6 physical climate model (HadGEM3-GC3.1) at a high-resolution atmospheric scale (N216, approximately 60 km at midlatitudes) and a 1/4° resolution for oceanic data. The ensemble comprises forty members, driven by CMIP6 historical data and SSP3-7.0 scenarios for the period from 1950 to 2099. First, we evaluate the ability of the NNs to anticipate the phases of long term (multidecadal variability) using observational datasets, thereby investigating the consistency of physical processes influencing NASST variability between modeled predictions and real-world observations. Second, the research delves into how the interplay between oceanic and atmospheric predictors, alongside external forcings and internal variability (atmospheric noise), impacts the machine learning-based predictions and we use explainable AI techniques to identify the sources of predictability and to pinpoint physical mechanisms and regions crucial for accurate NN forecasts.

How to cite: Colfescu, I.: Explainable neural nets for disentangling sources of predictability in the North Atlantic Sea Surface Temperature (NASST), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6680, https://doi.org/10.5194/egusphere-egu25-6680, 2025.

Complex multiscale dynamics of the atmosphere in extratropical latitudes includes various persistent atmospheric regimes with the residence time up to several weeks. Identification, simulation and prediction of such dynamics remains one of the challenging problems. In this work we use a stochastic recurrent neural network (RNN) with specific architecture to address this problem, appealing to RNN’s ability to handle memory effects well. The proposed RNN connects two types of variables: (i) a low-dimensional representation of the physical variables via Principal Component Analysis (PCA), and (ii) Kernel PCA variables which serve to better represent the target atmospheric regimes [1]. The stochastic component of the RNN has a simple form which allows us to analytically write Bayesian log-posterior and log-likelihood functions to train and cross-validate the model given the particular dataset.
Using the observed and climate-model-generated winter geopotential height data in the Northern Hemisphere, we show that the proposed stochastic model is able to reproduce/predict various dynamical properties and distributions of the target regimes in the kernel space, as well as to reconstruct kernel variables from a low-dimensional representation of the original spatio-temporal field.

References
1. Mukhin et al. (2022). Revealing recurrent regimes of mid-latitude atmospheric variability using novel machine learning method. Chaos: An Interdisciplinary Journal of Nonlinear Science, 32(11). 

How to cite: Gavrilov, A., Mukhin, D., Safonov, S., and Samoilov, R.: Stochastic recurrent neural network for modeling atmospheric regimes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7404, https://doi.org/10.5194/egusphere-egu25-7404, 2025.

In hydrology, the use of machine learning (ML) has gained traction due to its ability to provide alternative or complementary approaches to traditional process-based modelling. These models identify numerical patterns in time series data without needing to solve conservation equations. This flexibility enables hydrological calculations in areas where data sources are incomplete or non-existent.
Studies benchmarking ML models (SVM, RNN, CNN) against process-based models have shown that ML models deliver promising results with lower computational cost and less information about the physical processes they are modelling. Consequently, they can effectively utilize spatially discretized physical data on a large scale.
This study designs a Long Short-Term Memory (LSTM) neural network to learn sequential relationships between atmospheric, climatic and geographic features and daily streamflow data from 39 headwater gauging stations in the northern Ebro river basin. LSTM models include an internal state that can store information and learn long-term dependencies, enabling them to model sequential data effectively. However, the numerical patterns identified by LSTM models do not inherently respect universal physical laws, such as the conservation of mass.
To address the limitation, the Mass-Conserving LSTM (MC-LSTM) model has been employed and compared with the standard LSTM model. The MC-LSTM model introduces a modified cell structure that adheres to conservation laws by extending the learning bias to model the redistribution of mass.
This analysis highlights not only the high accuracy of LSTM models in predictive hydrologic modelling but also the critical importance of integrating physics-based features to enable ML models to effectively capture the hydrological dynamics of the basin.
Acknowledgments: This work is funded by the European Research Council (ERC) through the Horizon Europe 2021 Starting Grant program under REA grant agreement number 101039181-SED@HEAD.

How to cite: González Planet, I. and Juez, C.: Streamflow forecasting in the Ebro river basin using Machine Learning (ML) and a physical mass constraint, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8105, https://doi.org/10.5194/egusphere-egu25-8105, 2025.

Accurate climate projections are critical for understanding climate change and to design adaptation and mitigation strategies. Weighting schemes that aggregate a range of climate model projections are widely used to provide more reliable estimates of future climate conditions. Recently, causal discovery has been successfully introduced in the weighting schemes to constrain uncertainties in climate model projections based on the performance and interdependence of climate models. However, the previous methodologies typically (and strongly) only utilize a single metric, the F1 score of performance and similarity between each climate model and observational data,  to compare the different models' causal structures. Here, we introduce alternative and more sophisticated causal weighting schemes inspired by the theory of kernel methods and Gaussian processes to compare causal graphs directly in suitable reproducing kernel Hilbert spaces. In addition, we propose alternative causal weighting schemes that rely on interventions, graph-based distances, and counterfactual evaluations. We will evaluate the causal weighting strategies in various synthetic and CMIP6 model datasets. 

How to cite: Camps-Valls, G., Debeire, K., Varando, G., Runge, J., and Eyring, V.: Causal Weighting for Climate Projections, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8363, https://doi.org/10.5194/egusphere-egu25-8363, 2025.

Atmospheric ozone is a crucial absorber of solar radiation and an important greenhouse gas. However, explicitly representing ozone in climate models is computationally expensive. A recent study introduced a simple linear machine learning-based ozone parameterization scheme (mloz) for daily ozone prediction based on temperature. Here we develop and implement the mloz in the UK Earth System Model (UKESM) for long-term idealized climate simulations. It produces stable ozone predictions over 50 years with a computational cost of less than 0.5% of the total runtime. The scheme accurately predicts ozone distribution, with climatology field errors of less than 10% in the stratosphere. It also realistically represents ozone variabilities, including seasonal and Quasi-Biennial Oscillation-related variabilities, despite a slight underestimation of amplitudes over the stratospheric polar regions. Additionally, we further demonstrated its generalizability by successfully transferring the mloz trained on UKESM to the ICOsahedral Nonhydrostatic model (ICON). Over 30 years of climate sensitivity tests indicate that it can effectively represent the response of ozone to the sudden quadrupling of CO₂, significantly outperforming the simplified linearized ozone photochemistry scheme (Linoz) in the troposphere. This implies that the mloz can be transferred to other climate models without a full chemistry module to enable an efficient explicit ozone simulation.

How to cite: Ma, Y., Abraham, L., Versick, S., Ruhnke, R., Braesicke, P., and Nowack, P.: A Highly Efficient Machine Learning-based Ozone Parameterization for Climate Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9358, https://doi.org/10.5194/egusphere-egu25-9358, 2025.

Assessing precipitation impacts due to anthropogenic climate change relies on accurate and high-resolution numerical Earth system model (ESM) simulations. However, such simulations are computationally too expensive, and their discretized formulation can introduce systematic errors. These can, for example, lead to an underestimation of spatial intermittency and extreme events.
Generative machine learning has been shown to skillfully downscale and correct precipitation fields from numerical simulations [1].
However, these approaches require separate training for each Earth system model, making corrections of large ESM ensembles computationally costly.
Here, we follow a diffusion-based approach [2] by training an unconditional generative consistency model [3] on high-resolution ERA5 precipitation data. Once trained, a single generative model can be used to efficiently downscale arbitrary ESM simulations in an uncertainty-aware and scale-adaptive manner. Using three different climate models, GFDL-ESM4 [4], POEM [5], and SpeedyWeather [6], we evaluate the performance and generalizability of our approach.

[1] Harris, L., McRae, A.T., Chantry, M., Dueben, P.D. and Palmer, T.N., 2022. A generative deep learning approach to stochastic downscaling of precipitation forecasts. Journal of Advances in Modeling Earth Systems, 14(10), e2022MS003120.
[2] Hess, P., Aich, M., Pan, B., and Boers, N., 2024. Fast, Scale-Adaptive, and Uncertainty-Aware Downscaling of Earth System Model Fields with Generative Machine Learning. arXiv preprint arXiv:2403.02774.
[3] Song, Y., Dhariwal, P., Chen, M., and Sutskever, I. 2023.  Consistency Models. In International Conference on Machine Learning (pp. 32211-32252).
[4] Dunne, J.P., Horowitz, L.W., Adcroft, A.J., Ginoux, P., Held, I.M., John, J.G., Krasting, J.P., Malyshev, S., Naik, V., Paulot, F. and Shevliakova, E., 2020. The GFDL Earth System Model version 4.1 (GFDL‐ESM 4.1): Overall coupled model description and simulation characteristics. Journal of Advances in Modeling Earth Systems, 12(11), e2019MS002015.
[5] Drüke, M., von Bloh, W., Petri, S., Sakschewski, B., Schaphoff, S., Forkel, M., Huiskamp, W., Feulner, G. and Thonicke, K., 2021. CM2Mc-LPJmL v1.0: biophysical coupling of a process-based dynamic vegetation model with managed land to a general circulation model. Geoscientific Model Development 14, 4117–4141.
[6] Klöwer, M., Gelbrecht, M., Hotta, D., Willmert, J., Silvestri, S., Wagner, G.L., White, A., Hatfield, S., Kimpson, T., Constantinou, N.C. and Hill, C., 2024. SpeedyWeather.jl: Reinventing atmospheric general circulation models towards interactivity and extensibility. Journal of Open Source Software, 9(98), 6323.

How to cite: Hess, P., Aich, M., Pan, B., and Boers, N.: Downscaling precipitation simulations from Earth system models with generative machine learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9742, https://doi.org/10.5194/egusphere-egu25-9742, 2025.

Subseasonal-to-seasonal (S2S) forecasts are crucial for decision-making and early warning systems in extreme weather. However, the chaotic nature of atmospheric dynamics limits the predictive skill of climate models on S2S timescales. Teleconnections can provide windows of improved predictability, but leveraging these external drivers to enhance S2S forecast skill remains challenging. This study introduces a spatio-temporal neural network (STNN) designed to predict weekly North Atlantic European (NAE) weather regimes at lead times of one to six weeks during boreal winter. The STNN integrates a stacked vision transformer (ViT) encoder and a long short-term memory (LSTM) decoder to capture short- and medium-range variability. By incorporating spatio-temporal data on the stratospheric polar vortex, tropical outgoing longwave radiation, and 1D NAE regime time series, the network can access patterns linked to teleconnections of key drivers of European winter weather. Its modular design enables the application of mechanistic interpretability, providing novel neuron-level insights into the prediction behavior. The improved predictive skill beyond lead week three and enhanced accuracy for specific regimes suggest novel learned patterns of external drivers. Using Activation Maximization (AM), we analyze these learned representations, and by incorporating gradient-based explanations of correct predictions, we infer additional insights into prevalent teleconnections. 

How to cite: Bommer, P. L., Kretschmer, M., Spurler, F., Bykov, K., Boehnke, P., and Hoehne, M. M.-C.: Combining spatio-temporal neural networks with mechanistic interpretability to investigate teleconnections in S2S forecasts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9753, https://doi.org/10.5194/egusphere-egu25-9753, 2025.

Local climate information is crucial for impact assessment and decision-making, yet coarse global climate simulations cannot capture small-scale phenomena. Current statistical downscaling methods infer these phenomena as temporally decoupled spatial patches. However, to preserve physical properties, estimating spatio-temporally coherent high-resolution weather dynamics for multiple variables across long time horizons is crucial. We present a novel generative approach that uses a score-based diffusion model trained on high-resolution reanalysis data to capture the statistical properties of local weather dynamics. After training, we condition on coarse climate model data to generate weather patterns consistent with the aggregate information. As this inference task is inherently uncertain, we leverage the probabilistic nature of diffusion models and sample multiple trajectories. We evaluate our approach with high-resolution reanalysis information before applying it to the climate model downscaling task. We then demonstrate that the model generates spatially and temporally coherent weather dynamics that align with global climate output.

How to cite: Schmidt, J., Schmidt, L., Strnad, F., Ludwig, N., and Hennig, P.: Spatiotemporally Coherent Probabilistic Generation of Weather from Climate, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11083, https://doi.org/10.5194/egusphere-egu25-11083, 2025.

Recent years have seen rapid advancements in large-scale data-driven models for weather forecasting. Several of these models can now compete with, and in some respects outperform, physics-based numerical models for medium-range forecasting. They offer significant computational savings and potential forecasting accuracy improvements approximately equivalent to a decade of progress in traditional methods. This progress has prompted announcements from weather institutes across the world about plans to integrate AI-driven models into their operational workflows in the near future.

As data-driven models become integral to operational forecasting, critical questions about fairness and equity remain. Studies reveal substantial variations in forecast quality across regions, particularly for extreme weather. Unlike physical models, the disparities in data-driven models often stem from passive design decisions, such as inductive biases and weighting schemes, which may be reassessed and changed, if needed. Moreover, ensuring equitable access to these models, along with the means to effectively utilise and improve them, is essential so that both high- and low-income countries can share in their benefits.

This work explores fairness in data-driven weather forecasting, with a focus on outcome-based perspectives. We begin by defining fairness from both process and outcome viewpoints. We then analyse the performance of current data-driven models across different regions and socio-economic groups globally. Our findings reveal significant disparities that may exacerbate pre-existing socio-economic and climate-related vulnerabilities. To address these challenges, we advocate for a deliberate focus on fairness and equity in data-driven model development, emphasising the importance of active design choices to promote equitable outcomes.

How to cite: Olivetti, L. and Messori, G.: Whose weather is it? A fairness perspective on data-driven weather forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11668, https://doi.org/10.5194/egusphere-egu25-11668, 2025.

Machine learning (ML) parameterisations for climate models are emerging as a promising approach for capturing subgrid-scale processes, which are not explicitly resolved in climate models due to limitations on resolution. These ML parameterisations are typically trained on datasets generated by high resolution climate models or existing parameterisations (“offline”), but evaluated based on their performance when coupled into an existing climate model (“online”). Quantifying uncertainties associated with ML parameterisations is crucial for gaining insights into the reliability of hybrid ML-climate models.

I will discuss uncertainties associated with an ML parameterisation for atmospheric GWs, focusing on the parametric uncertainties which originate during the training process. I will show how these can propagate when coupled online, becoming a significant source of uncertainty in climate model circulation that we must consider carefully when building ML parameterisations.  

How to cite: Mansfield, L. and Sheshadri, A.: Uncertainty Quantification of Machine Learning Parameterisations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12129, https://doi.org/10.5194/egusphere-egu25-12129, 2025.

High-resolution climate projections are vital for understanding the local impacts of climate change in fields like agriculture, hydrology, energy production, and disaster risk management. However, Earth System Model (ESM) output often lacks the spatial detail needed to capture regional to local-scale variability, while showing large biases when compared to observational data. Statistical downscaling (SD) is commonly used to address such issues by refining the coarse spatial resolution of ESM output. While the current ISIMIP3 (Inter-Sectoral Impact Model Intercomparison Project, third round) SD algorithm is robust and computationally efficient, it struggles with increasing differences between source and target resolutions. To address these limitations, we applied deep-learning-based SD methods to create a globally consistent, high-resolution dataset for near-surface climate variables.

Using the perfect prognosis approach, we combined ERA5 as large-scale atmospheric predictors with ERA5-Land as high-resolution predictands (target resolution of ~10 km) to create accurate transfer functions (TFs) that align with ISIMIP's requirements, such as trend preservation and inter-variable consistency. These TFs are subsequently applied to ESM output to generate downscaled climate forcings. The resulting framework is both scalable and computationally efficient, making it suitable for multi-model applications. The results were compared with similar methodologies and its improvements were demonstrated in a cross-validation framework, particularly in capturing local-scale features.

Our approach offers a robust tool for generating high-resolution climate data, providing valuable insights to researchers and decision-makers working on climate impact assessments and adaptation planning. This work contributes to the next iteration of ISIMIP and to OptimESM, targeting the CMIP7-based modeling framework. The derived high-resolution projections are designed to complement CMIP7 datasets, enabling the creation of downscaled ensembles that conform with ISIMIP's objectives and support a wide range of impact modeling applications.

How to cite: Quesada-Chacón, D., Sauer, I., Mengel, M., and Frieler, K.: A deep learning approach to statistical downscaling and its potential to increase the resolution of the impact model simulations within ISIMIP, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12175, https://doi.org/10.5194/egusphere-egu25-12175, 2025.

Estimation of the Reference Evapotranspiration (ET₀) is critical in water resources management under climate change, especially for agricultural water management in arid and semi-arid regions. Thus, estimating baseline ET₀ poses significant challenges, particularly in inadequate climatological monitoring regions. In this study, a hybrid modeling approach based on the incorporation of empirical models, Particle Swarm Optimization (PSO), and XGBoost algorithm (Empirical-PSO-XGBoost) was developed and evaluated to forecast ET₀ under limited climate variables. The results showed the Empirical-PSO-XGBoost outperformed the purely calibrated empirical and Temperature-PSO-XGBoost models for estimating monthly (daily) ET₀ with NSE reaching 0.99 (0.86) and 0.98 (0.67) for the calibration and validation phases, respectively. Besides, up to 63 CMIP6 projections were coupled with Empirical-PSO-XGBoost for forecasting the long-term ET₀ under SSP245 and SSP585 climate change scenarios. Thus, the simulation showed a significant increase in ET₀ and seasonal patterns compared to the baseline ET₀ where the change in range of [+5, +10] % is associated with probability values of 0.65 and 0.78 for SSP245 and SSP585, respectively. Overall, the developed framework is useful for implementing adaptation strategies to mitigate climate change effects on water resource allocation and agricultural management. It provides the ET₀ associated with Exceedance probability for each month which is useful for assessing the water availability-related-risk in scheduling irrigation and sowing date of crops.

How to cite: Elbilali, A., Hadri, A., Taleb, A., EL Khalki, E. M., Tanarhte, M., and Kharrou, M. H.: A Novel Modeling Framework based on Empirical models, PSO, XGBoost, and multiple GCMs for the projection of Long-Term Reference Evapotranspiration, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12646, https://doi.org/10.5194/egusphere-egu25-12646, 2025.

Climate variability and change significantly influence crop production, presenting challenges that extend from understanding the basic crop growth principles to evaluating the effects of extreme weather events on crop development. Addressing this requires effective agro-management strategies guided by tailored climate services. However, a critical gap exists between scientific insights and their practical application. This study introduces and evaluates an AI-driven methodology designed to simulate crop growth and predict grain maize yields across Europe. Specifically, nested Recurrent Neural Networks (RNNs) are tested as a computationally efficient surrogate model for the process-based ECroPS model developed by the European Commission’s Joint Research Centre. Traditional mechanistic crop models, like ECroPS, require numerous meteorological inputs and significant computational resources, limiting scalability for applications such as large-scale climate simulations or ensemble modeling that explore variables like climate projections and CO₂ effects. In contrast, the surrogate AI model relies on just three weather inputs—daily minimum and maximum temperatures and daily precipitation—trained using ECMWF-ERA5 reanalysis data. This streamlined approach demonstrates the potential to bridge the gap between resource-intensive crop modeling and scalable, data-driven solutions for climate impact assessments.

How to cite: Vlachopoulos, O., Luther, N., Ceglar, A., Toreti, A., and Xoplaki, E.: Surrogate impact modelling for crop yield assessment with nested RNNs, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13061, https://doi.org/10.5194/egusphere-egu25-13061, 2025.

How to cite: Laanti, T., Ezhova, E., Lintunen, A., Noe, S., Kulmala, M., Miles, V., and Heljanko, K.: XAI finds signs of clouds in the Net Ecosystem Exchange of boreal forest , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13234, https://doi.org/10.5194/egusphere-egu25-13234, 2025.

Making projections of possible future climates with models is essential to improve our understanding of the causes and implications of anthropogenic climate change. While Earth system models are currently the most complete description of the Earth system, these models are computationally expensive. Simpler models (emulators) are therefore useful to explore the large space of possible future climate scenarios and to generate large ensembles. One class of emulators are simple climate models (SCMs), which model the Earth system with simplified physics. A second class of emulators are statistical models, which learn relationships directly from correlations in climate model data. In this preliminary work, we seek to combine the benefits of the physical grounding of SCMs with those of purely statistical emulators, using tools from causal representation learning. The resulting causal climate emulator may allow exploration of the effect of various interventions on the Earth system, including the effect of changing forcings.

The goal of causal representation learning (CRL) is to simultaneously learn low-dimensional latent representations from high-dimensional data, and a causal graph between these latent representations. In the context of climate model data, we aim to infer latent variables representing regions with shared climate variability from fine-grid climate model data, and causal teleconnections between these regions, representing climate dynamics. We build on recent previous work by Boussard et al., which illustrated how a CRL method, Causal Discovery with Single-parent Decoding (CDSD), may be used for this task. CDSD is a continuous optimization method to learn a distribution over latent variables such that every grid-point observation is driven by a single latent variable, and a causal graph between these latents is also learned. 

We illustrate that on surface fields of monthly pre-industrial climate model data, CDSD learns physically-reasonable latent variables but learning a robust causal graph between the latent variables remains a challenge. We evaluate our models on synthetic data that approximate the spatiotemporal structures that we observe in climate model data. By autoregressively rolling out the model we can then generate an ensemble of future climate trajectories with the learned generative model. We develop a Bayesian filter to maintain a constant spatial spectrum throughout our autoregressive rollout, and show that it leads to stable climate prediction. Finally, we explore approaches for including the effect of forcings such as greenhouse gasses in the model.

How to cite: Boussard, J., Hickman, S., Trajkovic, I., Kaltenborn, J., Gurwicz, Y., Nowack, P., and Rolnick, D.: Causal climate emulation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13307, https://doi.org/10.5194/egusphere-egu25-13307, 2025.

Earth system models (ESMs) are widely used to study climate changes resulting from both anthropogenic and natural perturbations. Over the past years, significant advances have been made through the development of new numerical schemes, refined physical parameterizations, and the use of increasingly powerful computers. Despite these advances, tuning ESMs to accurately reproduce historical data remains largely a manual process, and persistent errors and biases continue to challenge their accuracy. Reducing uncertainties in long-term climate projections and accurately estimating the spread of climate simulations continue to be critical challenges.

Recent advances in machine learning have motivated the development of learning-based methods for the calibration of ESMs. One emerging area of research is the design of hybrid modeling approaches, which combine a physical core with a machine learning model. Training these hybrid models end-to-end (or online) has the potential to unify various challenges in ESMs development, ranging from building subgrid scale parameterizations, to bias correction and parameter tuning.

Training hybrid models online requires working with an optimization problem that depends on the numerical integration of the system. Solving this optimization problem using gradient-based approaches requires the system to be differentiable, or to have access to the adjoint of the numerical model, which is not the case for most of the large-scale physical models. Beyond the need for differentiability, developing hybrid models requires interfacing a physical core that is implemented in low-abstraction languages that are running on CPUs, with AI-based models that are developed using high-abstraction, rapidly evolving languages that run on GPUs. While this interface is not a problem at inference time, doing this interface at calibration time, which is necessary when doing online learning, is not trivial as it would require an iterative communication between components that are implemented on different architectures.

In this work, we aim to investigate online learning and hybrid models to develop new computing paradigms, tools, and calibration methods for designing numerical models that are closely aligned with observations. We study the potential of online learning for deriving efficient and scalable solutions to the above-mentioned problems for applications that include both short-term forecasting and long-term simulations, which require stability considerations of the resulting hybrid systems. We explore learning configurations that include both fully differentiable and black-box physical cores. The latter configuration aims at evaluating the extent to which differentiable programming frameworks can upscale modeling capabilities in terms of accuracy, computational efficiency, and adaptability to represent diverse physical processes.

How to cite: Ouala, S., Meunier, E., Fablet, R., and Le Sommer, J.: Leveraging Differentiable Programming and Online Learning for the design of Hybrid Numerical Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13717, https://doi.org/10.5194/egusphere-egu25-13717, 2025.

Clouds influence Earth’s climate by reflecting sunlight and trapping heat, but their role in climate change remains uncertain, causing major unpredictability in models. Global 3D cloud data can improve predictions.

Observations from NASA’s CloudSat mission have advanced our understanding of cloud structures but are limited by long revisit times and narrow coverage. Imaging instruments offer broader, faster coverage but lack vertical information.

In [1] a deep learning approach addressed this challenge by combining MSG/SEVIRI satellite imagery with CloudSat profiles to extrapolate vertical cloud structures beyond observed tracks. Using geospatially-aware Masked Autoencoders, models were pre-trained on a year of MSG data (2010) and fine-tuned with CloudSat tracks as ground truth. This self-supervised training improved reconstruction, outperforming previous methods and simpler architectures [2].

In this work, we explore to what degree including information of the temporal dynamics of clouds can further improve the quality of the 3D cloud reconstruction. Instead of using a single image  as input we use a temporal sequence of MSG/SEVIRI images, spanning a period of several hours before and after the target cloud vertical profile. We use a combination of the geospatial encodings used in [1] and the temporal encoding used in [3] to embed these spatiotemporal MSG/SEVIRI cubes in rich, general purpose latent space. We then use a finetuning model as in [1] to map the embeddings into 3D radar reflectivity maps. 

We perform a sensitivity analysis to explore how the quality of the reconstruction varies as a function of the amount of temporal information included. We also explore the relative strengths of different pre-training strategies with respect to the quality of the 3D reflectivity reconstruction and cloud type segmentations. With this, we provide insights on self-supervised learning for atmospheric applications.

References

• Stella Girtsou et al. “3D Cloud reconstruction through geospatially-aware Masked Autoencoders” 2024. arXiv: 2501.02035 [cs.CV]. URL: https://arxiv.org/abs/2501.02035.
• Sarah Brüning et al. “Artificial intelligence (AI)-derived 3D cloud tomography from geostationary 2D satellite data”. en. In: Atmos. Meas. Tech. 17.3 (Feb. 2024), pp. 961–978.
• Yezhen Cong et al. SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery. 2023. arXiv: 2207.08051 [cs.CV]. URL: https://arxiv.org/abs/2207.08051.

How to cite: Diaz, E., Bintsi, K.-M., Castiglioni, G., Eisinger, M., Freischem, L., Girtsou, S., Johnson, E., Jones, W., Jungbluth, A., and Massant, J.: Reconstructing 3D vertical cloud profiles using cloud dynamics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13734, https://doi.org/10.5194/egusphere-egu25-13734, 2025.

Reducing climate model biases is crucial for decreasing uncertainties in future climate projections. Despite recent efforts, improvements between the latest generations of Earth System Models (ESMs) have been modest, primarily due to the continued reliance on subgrid-scale parametrizations. These parametrizations are necessary because the model resolutions in CMIP6 are too coarse to explicitly simulate too small-scale processes such as ocean mesoscale eddies and deep atmospheric convection, which significantly influence regional and global climate patterns. Recent advances in computational power have enabled higher-resolution models, allowing for some of these processes to be simulated explicitly, reducing the need for parametrization. Here, we combine a convolutional neural network (CNN) classifier and explainable AI (XAI) to investigate the role of increased resolution in simulating winter surface temperature fields. The CNN is used to classify ESMs with varying resolutions based on snapshots of their surface temperature fields, while the XAI approach explains which regions and features the CNN relies on to make these distinctions, providing deeper insights into ESM performance. Results indicate that models with similar ocean grids are more frequently confused by the CNN than those from similar modeling centers, emphasizing the crucial role of ocean resolution, particularly the presence of mesoscale eddies, in shaping climate simulations. Although the analysis is restricted to surface air temperature, the XAI approach offers a more nuanced understanding of model differences compared to traditional bias analyses. This methodology can be extended to other climate variables and ESM features, offering a powerful tool for enhancing model intercomparison and evaluating ESM performance.

How to cite: Michel, S., Strommen, K., and Christensen, H.: Unravelling the role of increased model resolution on surface temperature fields using explainable AI, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15256, https://doi.org/10.5194/egusphere-egu25-15256, 2025.

Climate change is profoundly affecting ecosystems and societies. Impacts on hydrological regimes, water resources, and urban heatwaves are particularly important, emphasizing the need for a detailed understanding of these changes at local scales to inform effective adaptation strategies. Achieving this requires reliable, high-resolution projections of future climate conditions. However, current climate models operate at coarse spatial resolutions, limiting their ability to capture small-scale processes and extreme weather events. To bridge this gap, robust downscaling techniques are essential for refining the outputs of global and regional climate models.

We propose a multivariate super-resolution (SR) approach to downscale temperature and precipitation data in Switzerland to improve the representation of localized patterns, particularly in Alpine regions, while simultaneously capturing the interdependencies between temperature and precipitation, which are crucial for hydrological applications. We leverage advanced machine learning techniques, including Generative Adversarial Networks (GANs) and Diffusion models, to overcome the limitations of classical methods in capturing inter-variable dependencies. These models provide an ensemble framework, providing multiple possible realizations, to account for downscaling uncertainties, resulting in more robust and reliable outputs for impact modeling and decision-making. We test different loss functions, like a regional CRPS, to allow for variability in the generated meteorological fields.

We compare the performance of GANs and Diffusion models along with the differences between univariate and multivariate settings. Our approach includes applying a multivariate bias correction prior to downscaling. The downscaled results are compared to a setting based on univariate bias correction. Additionally, we present the pipeline, which integrates bias correction and downscaling and is intended to be open source.

How to cite: Horton, P., Samarin, M., Otero, N., Allen, S., and Volpi, M.: Multivariate climate downscaling using deep learning models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15363, https://doi.org/10.5194/egusphere-egu25-15363, 2025.

Understanding the potential impacts of climate change on global vegetation dynamics is crucial for effective environmental management and biodiversity conservation. This study employs a machine learning-based framework to analyze historical NDVI data and project future vegetation growth under different climate scenarios. Utilizing the GIMMS NDVI dataset (1981–2000) for model training and CMIP6 climate projections (2021–2100) for scenario analysis, the study evaluates changes in vegetation growth across four Shared Socioeconomic Pathways (SSPs). Results indicate a significant near-term increase in global mean NDVI (2021–2040) under all scenarios, followed by divergent trends. While SSP126 and SSP245 sustain modest increases, SSP370 and SSP585 show sharp declines in NDVI over the long term, driven by adverse temperature effects. Regional analyses reveal contrasting patterns: NDVI values in Africa, South America, and Oceania decline under most scenarios, while North America, Europe, and Asia exhibit potential increases, except under high-emission scenarios like SSP585. These findings underscore the importance of targeted interventions to mitigate climate impacts and highlight the role of machine learning in predicting vegetation responses to environmental changes. The study provides actionable insights for policymakers, emphasizing the need for sustainable land management practices and greenhouse gas reduction strategies to preserve global ecosystems.

How to cite: Nguyen, A. K. and Chen, W.: Machine Learning Projections of Climate Change Impacts on Global Vegetation Dynamics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15622, https://doi.org/10.5194/egusphere-egu25-15622, 2025.

Earth’s seasonality profoundly influences nearly every aspect of life on our planet. It plays a key role in driving vegetation cycles and shaping wildlife behavior. Seasonality also impacts human life significantly, affecting health, mood, social dynamics, and cultural patterns. Despite its importance, seasonality is still traditionally defined by astronomical seasons—equal-length divisions applied uniformly across the Earth. Although this division is simple and intuitive, it overlooks crucial seasonal patterns influenced by atmospheric weather. In this study, we propose a data-driven approach to redefining seasons using objective clustering. We develop an algorithm that segments various meteorological factors into meaningful seasonal clusters. Building on this algorithm, we objectively define seasons for each region globally and analyze the effects of Climate Change on these clusters. We find that seasonality is driven by different meteorological factors in different regions on Earth. Additionally, we observe that Climate Change has significantly altered the duration and onset of Earth’s seasons.

How to cite: Shmuel, A., Magaritz-Ronen, L., Raveh-Rubin, S., and Milo, R.: Revisiting Earth’s Seasonality using Machine Learning Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16234, https://doi.org/10.5194/egusphere-egu25-16234, 2025.

Machine learning (ML), and deep learning (DL) in particular, hold the potential to solve long-standing challenges in understanding and modeling the Earth system. Earth system model (ESM) development is reluctant to implement DL algorithms because they are considered intransparent, meaning it is unclear how these models extrapolate to unseen conditions, e.g., under a changing climate. Still, machine learning is often used to extrapolate into the future,  which can lead to misleading results.
We demonstrate these limitations and the dangers of performing naive extrapolation by using a set of deep neural networks to emulate simulated data of gross primary production (GPP). We use a process-based model (PBM) that simulates photosynthetic CO2 uptake as a product of radiation (PAR), stress from daily meteorology (f_Tmin , f_VPD , f_SM ), vegetation state (fPAR), and CO2 (ε_max(CO2 )). It is given by

GPP = ε_max (CO2 ) · PAR · fPAR · f_Tmin · f_VPD · f_SM + ε.                                                           (1)

The PBM contains many of the typical challenges when using ML for Earth’s system science. It accounts for stochastic noise (ε), is capable of exhibiting multi-year memory, and the predictors are highly correlated on multiple time scales. Further, this model exhibits interesting extrapolation behavior as some of the factors  (f_Tmin , f_VPD , f_SM , f_PAR) saturate in extreme meteorological conditions while others (PAR, ε_max ) do not. We feed the PBM with predictors obtained from historical and future climate simulations of a comprehensive Earth system model. The training dataset contains the predictors and predictions of the PBM for various locations in a similar climate zone but different continents and for the historical time frame (1850-present) together with a spurious predictor, namely, surface wind speed. To obtain a set of independent models, each of the co-authors separately implements a custom architecture, without knowing which predictor is which. This results in four different models, namely a linear model, a multi-layer perceptron, a long-short term memory (LSTM), and an attention-based model.
We find that all models show strong prediction performance in cross-validation (Normalized Nash–Sutcliffe Efficiency (NNSE) > 0.9), decent performance when extrapolating to sites on different continents (NNSE > 0.7), but three out of four models show virtually no skill when predicting to a changed climate (NNSE < 0.6). Additionally, most models emit gradients in the same order of magnitude as the PBM when ignoring values where  some factors saturate. This indicates that the networks did not learn the saturation behavior from the data. Further, the model that extrapolates best is the LSTM, a model that has a built-in maximum output and, hence, has to saturate.
In conclusion, strong spatial generalization and cross-validation performance do not guarantee decent extrapolation for neural networks even in relatively simple, stable systems. These findings highlight the importance of selecting architectures in line with the expected extrapolation behavior when predicting Earth’s system processes under climate change conditions.

How to cite: Reimers, C., ElGhawi, R., Kraft, B., and Winkler, A. J.: Limitations of Machine Learning Models in Extrapolating to a Changing Climate, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16360, https://doi.org/10.5194/egusphere-egu25-16360, 2025.

The performance of a deep convolutional neural network in predicting near-surface air temperature (T2m) and total precipitation (P) over Europe is assessed, comparing its results with the Copernicus European Regional Reanalysis (CERRA) and the regional dynamical model HCLIM. The ML-model accurately captures broad seasonal temperature and precipitation patterns, with minor biases in summer and more pronounced warm biases in winter. While the model effectively reproduces the probability density functions (PDFs) of daily temperature and precipitation, it underestimates extreme cold events and the high precipitation extremes in some regions. Climate indices, including cold extremes (TM2PCTL), warm extremes (TM98PCTL), consecutive dry days (CDD), and consecutive wet days (CWD), highlight that the ML model aligns closely with CERRA. However, it slightly underestimates CDD and overestimates CWD, particularly in mountainous and Mediterranean regions. Analyses of spatio-temporal variability demonstrate high correlations with CERRA for temperature, exceeding 0.99 for spatial patterns and 0.95 for temporal correlations, while correlations for precipitation are lower, with underestimated temporal variability. The ML model generally outperforms HCLIM, particularly in aligning with observed data, although challenges remain in capturing extremes and reducing biases in certain regions. These results further highlight the potential of the ML model for regional climate downscaling and impact studies, while emphasizing the need for further refinement to enhance its representation of extreme events and improve spatial accuracy.

How to cite: Fuentes-Franco, R., Ivanov, M., Koenigk, T., Krus, K., Aldama Campino, A., and Wang, F.: Assessment of a Pan-European high-resolution downscaling through Deep Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17278, https://doi.org/10.5194/egusphere-egu25-17278, 2025.

Despite the remarkable strides made by AI-driven models in modern precipitation forecasting, these black-box models cannot inherently deepen the comprehension of underlying mechanisms. To address this limitation, we propose an AI-driven knowledge discovery framework known as genetic algorithm-geographic weighted regression. Through this framework, we have constructed an iterative optimization of knowledge generation and utilization. On the one hand, new explicit equations are discovered to describe the intricate relationship between precipitation patterns and terrain characteristics. Experiments have shown that the discovered equations demonstrate remarkable accuracy when applied to precipitation data, outperforming conventional empirical models. Notably, our research reveals that the parameters within these equations are dynamic, adapting to evolving climate patterns. On the other hand, these previously undisclosed equations have contributed new knowledge about terrain-precipitation relationships, which can be embedded into the AI model for better interpretability and climate projection accuracy. Specifically, the unveiled equations can enable fine-scale downscaling for precipitation predictions using low-resolution future climate data. This capability offers invaluable insights into the anticipated changes in precipitation patterns across diverse terrains under future climate scenarios, which enhances our ability to address the challenges posed by contemporary climate science.

How to cite: Xu, H., Chen, Y., Zeng, Z., Li, N., Li, J., and Zhang, D.: Exploring Terrain-Precipitation Relationships with Interpretable AI for Advancing Future Climate Projections, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17628, https://doi.org/10.5194/egusphere-egu25-17628, 2025.

In recent years, machine learning has emerged as a promising alternative to numerical weather prediction models, offering the potential for cost-effective and accurate forecasts. However, a significant limitation of current machine learning methods for weather forecasting is the lack of principled and efficient uncertainty quantification—a key element given the complexity of the Earth's climate system and the challenges in modeling its processes and feedback mechanisms. Inadequate uncertainty quantification and reporting undermines trust in and the practical use of current weather forecasting methods (Eyring et al., 2024).

Uncertainty quantification methods for weather forecasting typically use prediction intervals and can be categorized into Bayesian and frequentist approaches. Bayesian methods, while theoretically appealing, often involve restrictive assumptions and do not scale well to the complexity of spatio-temporal data. Frequentist approaches, such as ensemble-based methods, are widely used in weather forecasting and include techniques like perturbing initial states with noise (Bi et al., 2023; Scher et al., 2021), varying neural network parameters (Graubner et al., 2022), or training generative models (Price et al., 2023). However, most frequentist methods provide only asymptotically valid prediction intervals, which may not suffice in all weather forecasting applications.

Conformal prediction (CP) is a promising uncertainty quantification framework that delivers valid and efficient prediction intervals for any learning algorithm, without requiring assumptions about the underlying data distribution (Vovk et al., 2005). Despite its growing popularity in the machine learning and statistics communities, traditional CP methods are not tailored to spatio-temporal data in weather forecasting. This is due to challenges arising from spatial and temporal dependencies—such as spatial autocorrelation and temporal dynamics—that violate the exchangeability assumption underlying standard CP methods. Several recent studies attempted to address these challenges by introducing new CP algorithms specifically designed for various types of non-exchangeability (Oliveira et al., 2024). However, these adaptations face several limitations, including high computational complexity, asymptotic guarantees, and/or the need for recalibration of prediction intervals.

In this presentation, we will evaluate CP methods in the context of weather forecasting and discuss several limitations. In addition, we will highlight recent advances and discuss potential future directions that could address challenges underlying the use of CP in weather forecasting.

References:

Eyring, V., et al. Pushing the Frontiers in Climate Modelling and Analysis with Machine Learning. Nature Climate Change, 2024.

Leutbecher, M., et al. Ensemble Forecasting. JCP, 2008.

Bi, K., et al. Accurate Medium-range Global Weather Forecasting with 3D Neural Networks. Nature, 2023.

Scher, S., et al. Ensemble Methods for Neural Network-based Weather Forecasts. JAMES, 2021.

Graubner, A., et al. Calibration of Large Neural Weather Models. NeurIPS, 2022.

Price, I., et al. Probabilistic Weather Forecasting with Machine Learning. Nature, 2025.

Vovk, V., et al. Algorithmic learning in a random world. New York: Springer, 2005.

Oliveira, R.I., et al. Split Conformal Prediction and Non-exchangeable Data. JMLR, 2024.

How to cite: Mortier, T., Decancq, C., Sale, Y., Javanmardi, A., Waegeman, W., Hüllermeier, E., and Miralles, D. G.: Valid Prediction Intervals for Weather Forecasting with Conformal Prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17783, https://doi.org/10.5194/egusphere-egu25-17783, 2025.

Earth System Models (ESMs) are fundamental to understanding and projecting climate change. While they have demonstrated continuous improvements over the last decades, systematic errors and large uncertainties in their projections remain. A large contribution to these uncertainties stems from the representation of unresolved processes such as clouds and convection that occur at scales smaller than the model grid spacing. This impacts the models’ ability to accurately project global and regional climate change, climate variability, and extremes. High-resolution models with horizontal grid spacing of a few kilometers or less alleviate many biases of coarse-resolution models, but at high computational costs. Yet short simulations from high-resolution models can be used to inform machine learning (ML)-based parameterizations that are then incorporated into hybrid (physics+ML) ESMs. This new generation of hybrid models promises to reduce systematic errors and enhance projection capabilities compared to current state-of-the-art ESMs [1, 2]. In an effort to design a comprehensive hybrid ESM, the ICOsahedral Non-hydrostatic (ICON) model is equipped with a variety of physics-aware ML parameterizations, including moist convection, cloud cover and radiation. This talk will present an overview of the modelling activities undertaken within this framework, with a special focus on the developed ML-based cloud cover parameterization. This parameterization takes the form of an interpretable non-linear equation discovered through a combination of ML techniques including symbolic regression and sequential feature selection [3]. We demonstrate that, with this new parameterization, ICON runs stably over several decades and reduces global biases in cloud cover and radiation metrics. In addition, the new equation is controlled by only 10 free parameters that we automatically calibrate to achieve more accurate climate projections. This approach of discovering a low-dimensional data-driven equation for a parameterization with subsequent tuning of the hybrid model can be used in any host ESM provided suitable training data.

References:

[1] Eyring, V., Collins, W.D., Gentine, P. et al., Pushing the frontiers in climate modeling and analysis with machine learning, Nat. Climate Change, doi:10.1038/s41558-024-02095-y, 2024.

[2] Eyring, V., Gentine, P., Camps-Valls, G., Lawrence, D.M., and Reichstein, M., AI-empowered Next-generation Multiscale Climate Modeling for Mitigation and Adaptation, Nat. Geosci., doi:10.1038/s41561-024-01527-w, 2024.

[3] Grundner, A., Beucler, T., Gentine, P. and Eyring, V., Data-driven equation discovery of a cloud cover parameterization, J. Adv. Model. Earth Sys., doi:10.1029/2023MS003763, 2024.

How to cite: Savre, J., Schwabe, M., Grundner, A., Hefner, K., Heuer, H., Klamt, J., Pastori, L., Schlund, M., Gentine, P., and Eyring, V.: Towards a prototype hybrid ICON-ML model with physics-aware machine learning parameterizations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17999, https://doi.org/10.5194/egusphere-egu25-17999, 2025.

Aerosols affect the Earth’s energy budget by both scattering and absorbing solar radiation. Measuring parameters that separately quantify the two components, such as the aerosol absorption optical depth (AAOD), is key to better understanding the aerosol direct climate effect. As most satellite instruments can only retrieve the total aerosol extinction signal, the most reliable source of global AAOD observations is the ground-based AERONET sensor network. AERONET comprises hundreds of stations worldwide; however, their spatial distribution is uneven and coverage remains sparse in many relevant regions. To effectively reduce our uncertainty related to absorbing aerosols and efficiently expand the network, new stations should be placed in locations that maximise measurement informativeness. In this study, we address the problem of optimal sensor placement using convolutional neural processes (ConvNPs). ConvNPs are meta-learning models that use convolutional neural networks to learn maps from heterogeneous input datasets to a context-dependent Gaussian predictive model. We train ConvNPs using reanalysis data to learn to model daily global AAOD from sparse point observations given at station locations and additional gridded auxiliary data. The model’s probabilistic predictions are then harnessed in an active learning framework to sequentially propose new observation locations that optimally reduce model uncertainty and improve the network's informativeness. Our subsequent analysis considers further practical factors that might trade off with informativeness in the selection of new station locations, such as cloudiness and remoteness. The resulting proposed placements identify locations that would optimally enhance ground-based AAOD observation and can inform and focus future network expansion efforts.

How to cite: Pelucchi, P., Coca-Castro, A., Andersson, T. R., Vicent Servera, J., and Camps-Valls, G.: Optimal Sensor Placement for Aerosol Absorption Optical Depth with Convolutional Neural Processes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18735, https://doi.org/10.5194/egusphere-egu25-18735, 2025.

How to cite: Atkinson, J., O'Gorman, P., Berner, J., and Weinzierl, M.: Coupling a new convection parameterisation trained using high-resolution simulations to the Community Atmospheric Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19177, https://doi.org/10.5194/egusphere-egu25-19177, 2025.

Neural operators are transforming computationally intensive scientific disciplines such as weather forecasting and climate modeling, accelerating simulations by several orders of magnitude. However, they often fail to respect fundamental physical principles, such as conservation laws, during long autoregressive rollouts. We introduce an efficient correction layer that enforces global conservation constraints in neural operators. For initial conditions approximately satisfying the constraints, we prove that conservation can be guaranteed while only moderately increasing the total runtime. In a number of fluid dynamics experiments, our method produces physically realistic simulations while maintaining the computational advantages of neural operators. Our results enable the development of reliable and efficient climate model emulators by ensuring that crucial physical balance equations, such as mass and energy, are preserved during extended simulations.

How to cite: White, A., Duruisseaux, V., Bonev, B., Azizzadenesheli, K., Anandkumar, A., and Boers, N.: Enforcing Conservation Laws in Neural Operators for Earth System Modeling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19180, https://doi.org/10.5194/egusphere-egu25-19180, 2025.

Generative Deep Learning architectures, such as Diffusion models, offer an alternative to traditional physical modeling and regression models due to their ability to produce stochastic ensembles with a single run. Even though these models are capable of downscaling coarse data, they are often trained in contained regions, which can lead to severe spatial overfitting as the model learns location-specific patterns rather than generalizable physical relationships. In practice, the usability of the models is constrained to the area where they were originally trained, and their predictive capabilities degrade significantly when applied to regions outside the training domain, even if these regions share similar characteristics.
This study presents a one-step and two-step diffusion model capable of downscaling 2-meter temperature from ERA5 to higher-resolution grids in large areas, such as the Contiguous United States or Europe, without spatially overfitting. We use CONUS404, a reanalysis dataset created using simulations of the Weather Research and Forecasting (WRF) model over the Contiguous United States, as our target data and ERA5 and constants involved in the creation of CONUS404, such as altitude and land use, as our input. The model has been trained over the whole area using 10 years of 3-hourly data, and two years have been used for testing. To study the spatial generalization capabilities of the model, we reserve an area of the study region solely for testing and compute evaluation metrics separately for this area to ensure meaningful results. We compare the results of training in large and small areas and the number of years. In addition, we discuss the usefulness of ensemble prediction and the effect that the number of ensemble members has on the performance of the downscaling. Future steps include applying this methodology for downscaling EURO-CORDEX to EMO1 and multivariate downscaling.

How to cite: Benitez Benavides, M., Rodríguez, M., Margalef, T., Panadero, J., and Dutta, O.: Stochastic diffusion model for large-scale temperature downscaling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19397, https://doi.org/10.5194/egusphere-egu25-19397, 2025.

We consider a data-driven framework for climate prediction tasks in which the dynamics are learnt in a low-dimensional latent space. We rely on dimensionality reduction techniques --- linear principal component analysis and nonlinear autoencoders and their variants --- to then learn dynamical evolution  in the corresponding latent space using disparate methodologies --- linear inverse modeling, dictionary-based sparse regression, reservoir computing, neural differential equations, attention-based transformers, etc. In this setting, we seek to better understand the interplay between the spatial and temporal representations of variability and how they affect prediction skill.

Balu Nadiga, Los Alamos National Laboratory and Kaushik Srinivasan, University of California Los Angeles

How to cite: Nadiga, B. and Srinivasan, K.: Climate Prediction Based on Latent Space Dynamics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21270, https://doi.org/10.5194/egusphere-egu25-21270, 2025.

Understanding internal wave is essential, as they exert a profound influence on a multitude of oceanic processes, including mixing and the transfer of energy across a vast range of spatial scales. The phase of internal waves can undergo a rapid alteration during propagation, resulting in the formation of broad spectral peaks. In this study, we introduce a stochastic model designed to parametrise the spectral properties of coastal internal waves. This model employs a Lorentzian function to characterise the broad internal tide peaks and a Matern function for the energy continuum. The efficacy of our model is validated using long-term in-situ mooring temperature data from the Australian Northwest Shelf (NWS) and Timor Sea. By optimising the model parameters using debiased Whittle likelihood in the frequency domain, our approach is able to reproduce the spectrum of internal wave incoherent peaks and the continuum of energy down to the buoyancy frequency. The fitted parameters allow for a comparison of internal wave properties between sites, depths, and seasons. The decorrelation timescale, indicative of the extent of the phase shift, exhibited a median value between 3 and 5 days and demonstrated minimal variation across sites and depths. The depth variation for the energy continuum amplitude and the amplitude of the semidiurnal peak exhibited an internal wave mode-1-like structure, particularly at the deeper mooring sites. The greatest amplitudes were observed within the surface mixed layer and thermocline. The slope parameter of the continuum exhibited a median value slightly less than the content slope in Garret-Munk spectral model and demonstrated seasonal variation, with a more rapid decay of energy in the summer compared to winter. The parameters obtained through our method can be further utilised to construct more realistic internal tide boundary conditions using Gaussian processes, thereby enabling more sophisticated modelling of internal waves in coastal regions. 

How to cite: Zheng, Y., Rayson, M., Jones, N., and Astfalck, L.: A Machine Learning Parametrisation for the Internal Gravity Wave Spectrum , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-637, https://doi.org/10.5194/egusphere-egu25-637, 2025.

Herein we apply Nonlinear Autoregressive models with exogenous Inputs (NARX) to estimate the interannual variability of satellite-derived chlorophyll-a (CHL) at a global scale, as function of sea-surface height (SSH) from a satellite product provided by Copernicus. A previous analysis shows that SSH is one of the top drivers of CHL in key regions of the tropical and south Atlantic, which is herein corroborated at a global scale, showing a significant CHL-SSH correlation in most of the world ocean between 60°S and 60°N (where the most continuous data series are available). This correlation, generally low for a linear estimation, opens the possibility to CHL reconstruction using higher-performance non-linear techniques like NARX. Herein the NARX model was generated with 10 neurons in the hidden layer, trained with a Levenberg-Marquardt algorithm, and applied to the CHL and SSH monthly composites from Oct 1997 to Sep 2024. A noise level of 0.57 for the model correlations was defined as the 95^th percentile of 10,000 NARX-modeled random series. This noise level is exceeded by 97% of the CHL-anomaly series modeled for the 1997-2024 period. The NARX-model successfully reproduces the CHL interannual variability: 59% of the modeled CHL present correlations > 0.90. Then, the NARX-model can be potentially used to predict CHL beyond the training period. In this study’s next stage, the predictability of CHL will be evaluated using SSH for a post-training period, and an ultimate goal for the NARX-model will be a predictability assessment using numerical-model predictions. Thus, the proposed method opens the possibility for reconstruction and prediction not only for CHL but also for other related biogeochemical variables.

How to cite: Rivas, D., Fransner, F., and Keenlyside, N.: Estimation of global satellite-derived chlorophyll-a as function of sea-surface height using shallow neural networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-766, https://doi.org/10.5194/egusphere-egu25-766, 2025.

Understanding ocean salinity is crucial for tracking changes in the Earth's water cycle and climate. However, collecting accurate salinity data has been challenging due to limited observations, especially in certain regions. This study focuses on the development of a method to create 2-dimensional maps of ocean salinity and its trends on pressure surfaces from sparse observations. An unsupervised classification technique called Gaussian Mixture Modeling (GMM) is used to identify coherent regions where temperature and salinity are tightly related at constant pressure. By grouping similar ocean regions using GMM, we are able to predict missing salinity data and fill gaps in historical salinity records from 1970 to 2014. The results show that this approach effectively estimates past salinity data. In the South Atlantic, at a pressure of 539 dbar, the root mean square error of salinity and of the linear trend of salinity are 0.040 g kg⁻¹ and 2.1 10⁻³g kg⁻¹ yr⁻¹. The method could help fill in missing salinity observations and thus improve our understanding of the intensification of the global water cycle in response to climate change.

How to cite: Piedagnel, E., Sohail, T., and Zika, J.: Mapping ocean salinity data using Gaussian Mixture Modeling., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1385, https://doi.org/10.5194/egusphere-egu25-1385, 2025.

The El Niño-Southern Oscillation (ENSO) is a global significant signal in marine science and exerts substantial climatic and socioeconomic impacts worldwide. However, the long-term prediction of ENSO remains a challenge because of its diversity, irregularity and asymmetry. Here, we develop a spatiotemporal fusion transformer network (STFTN), which designed a parallel encoder structure to effectively extract spatiotemporal information from sea surface temperature anomaly and Niño3.4 index simultaneously, thereby enhancing the precision of Niño3.4 index forecasts. STFTN leverages the attention mechanism within its parallel encoder structure to extract global characteristics and establish remote dependencies on targets. With this structure, STFTN displays better prediction accuracy in different lead months. Furthermore, the activation map used in STFTN visualizes the contribution of the predictors to the output which helps to comprehend the factors contributing to ENSO events. The results highlight the potential of our model of ENSO forecasts and comprehension. 

How to cite: Zhao, A. and Du, Z.: ENSO Forecasts with Spatiotemporal Fusion Transformer Network, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2550, https://doi.org/10.5194/egusphere-egu25-2550, 2025.

ENSO exerts profound impacts on global climate change through ocean-atmosphere interactions and serves as a critical factor in global climate prediction. However, its prediction remains challenging due to the complex spatiotemporal interactions and evolution processes, as well as the varying degrees of correlation and teleconnection across different geographical regions. To address this issue, this study proposes an advanced ENSO forecasting framework based on regional predictions and model ensemble. The framework leverages a graph self-attention mechanism (GAT) to learn and capture the spatiotemporal dependency signals of ENSO, which are then incorporated as physical constraints into a spatiotemporal graph convolutional neural network (STGCN) for regional predictions. Furthermore, machine learning algorithms, including XGBoost and SVR are employed to integrate the predictions from different regions. Experimental results based on reanalysis data demonstrate the effectiveness and robustness of the proposed framework, achieving a correlation skill exceeding 0.8 within a 12-month lead prediction period, and significantly improving the computational efficiency by filtering key signals.

How to cite: Xiao, H., Song, Z., and Wang, L.: Regional Ensemble ENSO Prediction Based on Graph Neural Networks with Self-Attention, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2635, https://doi.org/10.5194/egusphere-egu25-2635, 2025.

The airborne radar altimeter can be extrapolated to a variety of parameters, including sea surface height, sea surface wind speed, significant wave height, and the topography of land, sea ice and ice cap. However, the airborne radar altimeter observation data contains signal error terms such as airborne platform jitter and ocean waves, which will lead to a large bias in the observation data. Here, we propose a method based on the combination of bandpass filtering and adaptive feature AI analysis to achieve the inversion of high-resolution sea level anomaly (SLA) data from airborne radar altimeter aliased signals.

For the airborne altimeter along-track data, statistical analyses were first performed. After that, the along-track data are filtered to remove the influence of ocean waves signals and flight platform oscillations, and the secondary interpolation is fitted based on the interval of the airborne altimeter data. According to the sampling interval of the altimeter data, the mean sea surface (MSS) and tide data under the along-track are processed to obtain the corresponding SLA data. The same interpolation method is used to process AVISO and SWOT L3 data. Finally, through the deep learning framework, the adaptive feature AI analysis is constructed to invert the SLA data, optimise the model and achieve accurate SLA prediction. The experimental results show that the RMSE of the SLA of the airborne altimeter inversion data with the along-track SWOT L3 and AVISO data are 1.12cm and 0.44cm, respectively, and the airborne altimeter data can acquire more small-scale change signals. This study verifies the working mechanism of the new system payload and lays a solid data and algorithm foundation for the development of subsequent satellite payloads.

How to cite: Ren, M., Yu, F., Zhang, X., Tang, J., and Chen, G.: Research on sea level inversion method from airborne radar altimeter , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2984, https://doi.org/10.5194/egusphere-egu25-2984, 2025.

The irregular and incomplete coverage of in-situ ocean temperature profile observations is a major problem for various scientific applications in ocean and climate research and operational fields. However, high-resolution gridded datasets are needed to support applications. Here, we explore a physics-informed machine learning approach based on partial convolutions with multi-branch U-Net neural network structure to reconstruct the subsurface temperature profile fields with 0.1°×0.1° weekly resolution in Western Pacific Ocean. The input data include in-situ temperature profile observations, high-resolution satellite remote-sensing products (including sea surface height, sea surface temperature, sea surface salinity, etc.), and a coarse-resolution (1°× 1°) gridded subsurface temperature product (IAPv4). We show that the new reconstruction retained the large-scale features represented by the 1°× 1° temperature gridded data but added mesoscale features (because of the inputs of high-resolution satellite data). The application of physical constraints for subsurface vertical structure improves the reconstruction near thermocline. The root mean square error (RMSE) can be reduced by ~12% in the target region in average with greater improvements in the upper layer (0-700m). Further analysis shows the small-scale information is performed well also in the sparse observation coverage area with some typical mesoscale vortex features can be identified, and the features in the strait and offshore regions can be effectively improved compared with coarse resolution 1°× 1° temperature gridded data. The successful application of machine learning in this study provides confidence for the accurate reconstruction of high-resolution ocean and climate data, which can improve and complement the existing data assimilation and objective analysis methods for reconstructing multi-scale ocean information in complex regions.

How to cite: Wei, W., Cheng, L., and Tian, T.: Physics-Informed Machine Learning Reconstruction of High Resolution Ocean Subsurface Temperature Profiles From In-Situ and Satellite Observations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3468, https://doi.org/10.5194/egusphere-egu25-3468, 2025.

Marine ecosystems are a vital component of the global carbon cycle. Our understanding of the cycle within the ocean relies on a combination of numerical models and satellite observations, which are combined through data assimilation (DA) methods. Here we developed a global ensemble DA system for marine ecosystem prediction using the NEMO-MEDUSA coupled ocean-biogeochemistry model and the Parallel Data Assimilation Framework. Unlike deterministic DA systems, the ensemble approach provides flow-dependent uncertainty estimates, improving the reliability of global marine ecosystem forecasts.

How to cite: Chen, Y. and Partridge, D.: Phytoplankton carbon assimilation in a global ensemble marine ecosystem data assimilation system, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3630, https://doi.org/10.5194/egusphere-egu25-3630, 2025.

Climate change risks are often assessed using climate impact indicators (CIIs) determined for various socio-economic scenarios. Ideally, for every scenario an impact model, e.g. an ecological model or a hydrological model, processes outputs of a climate model to produce CIIs. However sometimes, even if outputs of a climate model are available for all scenarios, computation costs of the impact model can limit the number of scenarios with available CIIs. 

To fill this gap, we propose to infer CIIs for unseen scenarios, i.e. scenarios not processed by the impact model, with an interpretable equation. This equation is discovered using symbolic regression on a scenario processed by the impact model. Specifically, we discover an equation that predicts CIIs based on climate impact drivers (CIDs), where CIDs are variables of the climate model averaged monthly and spatially.

In our application, the impact model is a biogeochemical model of the Mediterranean Sea driven by the same regional climate model for two scenarios: RCP4.5 and RCP8.5.  Our CII is the annual mean Net Primary Production (NPP) summed over an offshore area in the Gulf of Lion (located in the North-western Mediterranean basin), where NPP is the total rate of organic carbon production by photosynthesis of marine phytoplankton minus their respiration.

Preliminary results show that the discovered equation reproduces well the trend and the interannual variability of NPP for the testing scenario RCP4.5, unseen during the training. Indeed, the scenario RCP8.5 is preferred for training as it spans a wider range of climatological contexts. If our preliminary results are confirmed, we could extend our approach to a large ensemble of climate models, in order to characterize the uncertainty of CIIs.

How to cite: Le Roux, E., Tandeo, P., Granero Belinchon, C., Baklouti, M., Le Sommer, J., Sevault, F., Somot, S., Doury, A., and Al Najar, M.: Equation discovery for climate impact: symbolic regression to emulate climate impact indicators for unseen scenarios, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3836, https://doi.org/10.5194/egusphere-egu25-3836, 2025.

Mesoscale ocean eddies are dynamic structures controlling a significant proportion of water exchanges between the surface and the deep ocean, and therefore of heat, carbon and nutrient transfers. The eddy dynamics, i.e. changes in height, velocity and energy, are classically computed through complex ocean equations such as the quasi-geostrophic balance. However, those computations are time-consuming and slow down decision-making in operational situations. Some recent studies have managed to define eddy dynamics with simple properties - centroid position, amplitude, radius, current velocity, and horizontal displacement - and to predict their future evolution with machine learning models (Wang et al., 2020). We aim to implement a simple machine learning model to predict eddy properties that can reconstruct eddy dynamics and to include it in operational tools.

In this study, we simplified eddy structures, converting their 2D/3D gridded physical space into a parametric space, characterized by the eddy properties obtained with the AMEDA algorithm (Le Vu et al., 2017). Thus we considered eddies as 2D ellipse structures with additional properties - centroid position, amplitude, semi-axis of ellipse, rotation angle, maximal current velocity, and horizontal displacements. Explainable simple ML models were trained to learn the evolution of those parameters between two consecutive time steps. Here we selected two approaches of the least square regression model: the global linear regression on the whole training dataset and the local linear regression based on the nearest neighbors observations. Performances of each model are evaluated with the RMSE metric and compared to identify which model gives the most satisfactory results for eddy prediction. 

Our analysis shows better performances with the local linear regression. However, the choice of more adapted models or a better selection of eddy properties would enhance the prediction of eddies. The next steps to the inclusion of the model in operational tools will be the consideration of eddy interactions - splitting and merging -, the uncertainty quantification and the data assimilation of eddy dynamics with an object-oriented approach.

References

Wang, X., Wang, H., Liu, D., Wang, W., 2020. The Prediction of Oceanic Mesoscale Eddy Properties and Propagation Trajectories Based on Machine Learning. Water 12, 2521. https://doi.org/10.3390/w12092521

Le Vu, B., Stegner, A., Arsouze, T., 2018. Angular Momentum Eddy Detection and Tracking Algorithm (AMEDA) and Its Application to Coastal Eddy Formation. Journal of Atmospheric and Oceanic Technology 35, 739–762. https://doi.org/10.1175/JTECH-D-17-0010.1

How to cite: Dealbera, S., Tandeo, P., Granero-Belinchon, C., Raynaud, S., and Boussidi, B.: Object-oriented mesoscale eddy prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4043, https://doi.org/10.5194/egusphere-egu25-4043, 2025.

Ocean General Circulation Models (hereafter OGCM) are critical to the study of past and present climate, and the production of future projections. Unfortunately, they require large amounts of computations at simulation time. On the other hand, deep learning emulators trained on reanalyses are starting to deliver accurate short-term predictions, using comparatively small computational resources, yet they struggle to deliver long term predictions, are not interpretable and do no't factor in the uncertainty in physical parameters. As a result, they cannot be used by climate scientists to understand mechanisms of climate variability, such as tipping points, nor adjustment processes to greenhouse gas emissions.

Aiming to take the best from each world and establish a close interaction between emulators and OGCM, we investigate whether an emulator can be used to provide state variables to an ocean model, which would then handle the temporal integration using physics equations. Namely, we trained a diffusion model on a large dataset of ocean variables produced by NEMO, analysed the newly generated states, propose metrics to assess their physical consistency, and use them as initial conditions of simulations to assess their compatibility (physical and numerical) with NEMO.

Overall, we want to determine whether unconstrained generative models are able to produce realistic solutions, and to assess the tolerance of OGCM to externally generated ocean states, what we consider as a first step towards building an hybrid OGCM.

How to cite: Meunier, E., Lguensat, R., Gachon, G., Kamm, D., and Deshayes, J.: Can diffusion model generate ocean states compatible with OGCM ? , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4189, https://doi.org/10.5194/egusphere-egu25-4189, 2025.

The Atlantic Meridional Overturning Circulation (AMOC) is a critical component of the Earth system and one of the most prominent tipping element. In a warming climate, the AMOC is at risk of collapse due to increased freshwater input in the North Atlantic. Such an extreme event could lead to severe consequences for the global climate, with strong socio-economics impacts. Such a tipping event has been demonstrated to occur in conceptual, intermediate complexity and, recently, in the Community Earth System Model (CESM). Therefore, Reliable early warning signals are required for detecting whether the AMOC is approaching a tipping point. To estimate the distance of the AMOC to tipping, we propose a novel methodology, based on a Convolutional Neural Network (CNN) which uses sea surface salinity and temperature across the Atlantic as input. First, we validate our approach within the model of intermediate complexity Climber-X, demonstrating its ability to generalize to different forcing rates and in the presence of noise. We also explore the use of alternative climate variables such as the full-depth salinity profile at 35°S. Second, we assess the generalization capability of our methodology to a model of higher complexity. To this end, we use the CNN trained on Climber-X and successfully apply it to the AMOC collapse recently simulated in the CESM model. To demonstrate the physical consistency of the CNN model and increase its interpretability, we identify the most relevant regions to estimate the distance of the AMOC to tipping via the Layer-wise Relevance propagation technique.

How to cite: Guardamagna, F., Sinet, S., and Dijkstra, H.: Estimating the distance to the AMOC tipping point using convolutional neural networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4190, https://doi.org/10.5194/egusphere-egu25-4190, 2025.

The ocean's biogeochemistry is crucial for understanding the global ocean carbon cycle. Within the climate ocean model Nemo, the PISCES module (Pelagic Interactions Scheme for Carbon and Ecosystem Studies), is based on the numerical calculation of 24 different biological, physical and chemical variables which contribute to a complex bio-geo-chemical relationship to be able to estimate the net source and sinks of primary carbon production. In this work, we want to present the first steps toward using the Deep Neural Networks as a multi-variate problem trained on the model output to predict the next sequences and replace the module with an emulator solely based on machine learning (ML). 

How to cite: Rahpoe, N. and Bernardello, R.: A Deep Learn Emulator for Ocean Biogeochemical Modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4191, https://doi.org/10.5194/egusphere-egu25-4191, 2025.

The 4D variational assimilation (4DVar) framework is widely used in classical numerical weather prediction and geophysical data assimilation. However, a crucial assumption in 4DVar is that the model state that is close to the true state corresponds to the minimizer of the 4DVar cost function. Using a single-layer quasi-geostrophic (QG) model, we study scenarios where this assumption breaks down, particularly in the presence of model errors and suboptimal initialization. By introducing controlled perturbations in the initial conditions—we design experiments to investigate the sensitivity of 4DVar solutions. We find that minimizing the 4DVar score does not always correlate with achieving lower accuracy, suggesting the presence of local minima in the optimization process. 

4DVarNet, an end-to-end neural network based on variational data assimilation formulation, is trained in a supervised manner to solve the data assimilation task. This study aims to understand the advantage of trainable solvers that solve the same optimization problem using supervised learning, generating more accurate solutions efficiently. Through this case study based on observing system simulation experiments for sea surface geophysical fields, we show that supervised learning can overcome the minimization challenges of 4DVar when faced with observations that are irregular and highly sparse which are critical to address problems in ocean reconstruction. The advantage of learning allows 4DVarNet to discover hidden representations that are suitable for solving specific data assimilation tasks with better accuracy.

How to cite: Roy, S. K. and Fablet, R.: Performance Gains and Advantages of 4DVarNet in End-to-End Learning for Data Assimilation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4300, https://doi.org/10.5194/egusphere-egu25-4300, 2025.

Accurate forecasting of ocean dynamics is essential for understanding the distribution of heat, salinity, and nutrients in the ocean. While data-driven machine learning models offer promising solutions for ocean forecasting and emulating ocean models, they often lack physical consistency (i.e., adherence to the physical laws of fluid dynamics) and explainability. In this study, we introduce a deep neural network architecture leveraging Fourier Neural Operators (FNO) for efficient forecasting of ocean surface dynamics: sea level, temperature, and salinity. FNOs excel in learning resolution-invariant solutions of partial differential equations (PDEs), offering a scalable alternative to traditional physics-based models. Operating in Fourier space enables differentiation to be treated as multiplication, which is the basis of spectral methods used for solving PDEs, including the Navier-Stokes equations that govern hydrodynamic models. Therefore, it is intuitive that by directly parameterizing the integral kernel in Fourier space, the model can learn PDE solutions more efficiently. FNOs also enable training on low-resolution data and evaluation on high-resolution data, which helps minimize the growth of autoregressive errors.

Our model is trained on the Baltic Sea Physics Analysis and Forecast dataset to predict sea surface parameters, including sea level, temperature, and salinity. The Baltic Sea is a non-tidal, semi-enclosed sea with a complex coastline, shallow sea, significant salinity gradients, and permanent stratification, which makes it a unique and challenging testbed for ocean modelling. Input variables include the initial state, atmospheric forcing, and bathymetry, and the model is trained to predict ocean surface dynamics (sea level, temperature, and salinity) and learn the mapping from time t to t+1. In the inference step, the model is initialized with the initial sea surface inputs from an out-of-sample testing dataset and iteratively generates forecasts for τ time steps. Evaluation of the model demonstrates competitive forecasting skill compared to physical models, while significantly reducing computational costs. This study highlights the potential of FNOs to advance knowledge-driven machine learning models for ocean forecasting. These models, as cost-effective alternatives to high-resolution physical ocean models, can pave the way for more efficient, scalable approaches to understanding and predicting ocean dynamics.

How to cite: Jahanmard, V., Ellmann, A., and Delpeche-Ellmann, N.: Fourier Neural Operators for Emulating Ocean Models: Towards a Knowledge-Driven Machine Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4354, https://doi.org/10.5194/egusphere-egu25-4354, 2025.

The eddy-permitting NEMO model (ORCA025) is known to exhibit sub-par Southern Ocean circulation features, such as a too weak Antarctic Circumpolar Current transport and cool and warm biases on the Antarctic shelf. The ORCA025 model sits in the numerical grey zone, which is where the horizontal grid resolution can only resolve mesoscale processes over part of the domain. In other parts of the domain, the eddies need to be parameterised, such as high-latitude regions. This difficulty in representing eddies has in-part contributed to the poor Southern Ocean circulation, leading to great uncertainty in key climate metrics such as carbon and heat transport, and the Antarctic ice mass balance. The key question is, how do we parameterise mesoscale eddies where they are most needed, without being detrimental to the resolved flow. Scale- and flow-aware parameterisations have been implemented in NEMO and have led to improvements in some flow characteristics. However, an alternative approach is to leverage data, physics, and machine learning to develop an improved eddy parameterisation.

As part of the project, AI4PEX, we aim to develop a data- and physics-driven mesoscale eddy parameterisation that better captures the dynamical feedback between mesoscale eddies and the large-scale ocean circulation, reducing model uncertainty. In our work, we will attempt to improve an eddy parameterisation that is available in NEMO, GEOMETRIC. To do this we will use a Neural Network trained on high resolution data from realistic global models ORCA12/ORCA36. To reduce the black-box nature of the Neural Network, we will design a loss function that is informed by the physics of mesoscale eddies. Initial investigation of the eddy parameterisation will take place offline in an idealised configuration.

How to cite: Wilder, T., Kuhlbrodt, T., and Swaminathan, R.: Developing a data and physics driven machine learning mesoscale eddy parameterisation for NEMO, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4485, https://doi.org/10.5194/egusphere-egu25-4485, 2025.

The spin-up of the ocean component is a critical step in coupled global climate simulations, allowing the model to achieve a physically consistent equilibrium by stabilising key variables such as temperature, salinity, and ocean currents. Without an adequate spin-up, residual drifts can undermine the accuracy and reliability of long-term climate projections. This study explores data-driven strategies to accelerate the spin-up, reducing computational costs while preserving the fidelity of simulated climate states. Using a low-resolution configuration of the EC-Earth4 Earth System Model (ESM), we tested few deterministic approaches to optimise the spin-up phase. A key method relies on iterative adjustments of the oceanic state by projecting multi-decadal trends in temperature and salinity. Empirical Orthogonal Function (EOF) analysis was employed to filter internal variability and generate new initial conditions that minimise numerical instabilities. Additionally, vertical stability was ensured to reduce energy imbalances and maintain physical consistency. Overall, our approach can significantly enhance the efficiency of spin-up processes in coupled climate models by at least a factor of two. These findings pave the way for the development of more sustainable and sophisticated strategies (e.g. exploiting machine learning and AI techniques) in climate modelling. Such advancements will be particularly helpful for high-resolution simulations, where achieving computational efficiency is critical.

How to cite: Sozza, A., Davini, P., and Corti, S.: Data-driven approaches for accelerating ocean spin-up in coupled climate simulations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4573, https://doi.org/10.5194/egusphere-egu25-4573, 2025.

In the ocean, submesoscale physical phenomena O(100m) to O(1km) have been reported to play a key role in ocean oxygen ventilation, nutrient supply to the surface ocean, and carbon export, as well as the transfer of energy to larger scales [1]. However, due to limitations in computational resources, current ocean general circulation models are frequently run at resolutions on the order of O(10km) to O(100km) and cannot directly resolve submesoscale turbulence (i.e., subgrid-scale phenomena). Therefore, parameterization schemes are required to simulate these subgrid-scale phenomena.

Recent advances in machine learning have triggered the active exploration of data-driven approaches to parameterization for subgrid-scale phenomena that utilize data from observations and simulations. In previous studies [2, 3], the neural network is trained directly using the same variables as the neural network's output, such as viscosity and diffusivity coefficients. However, this approach does not guarantee that the inferred model parameters accurately represent the state of subgrid-scale phenomena they aim to reproduce. We propose a novel parameterization method for estimating diffusivity and viscosity parameters to parameterize subgrid-scale phenomena and have implemented this method in MITgcm, an ocean simulator [4, 5]. This method trains a neural network using the state variables (i.e., velocity fields, potential temperature, and salinity) derived from the simulation results at a resolution that can directly resolve subgrid-scale phenomena. Therefore, unlike previous studies, the diffusivity and viscosity parameters inferred by the trained network can reproduce the global state of subgrid-scale phenomena.

The ocean simulator MITgcm is implemented in Fortran, which does not have a built-in package to compute gradients within the neural network, in contrast to deep learning libraries (e.g., PyTorch) like Python. In our previous work [4, 5], we used a quasi-newton optimization method, which does not require computation of these gradients. However, the optimization performance of this method was limited. In this study, we use adjoint code within MITgcm to compute gradients for optimizing neural networks and examine the effect of different optimizers on training performance.

Acknowledgments
This work used computational resources of supercomputer Fugaku provided by the RIKEN Center for Computational Science through the HPCI System Research Project (Project ID: hp240394).

References
[1] M. Lévy et. al (2024), The impact of fine-scale currents on biogeochemical cycles in a changing ocean, Annual Review of Marine Science, 16(1), 191–215.
[2] Y. Han et. al (2020), A moist physics parameterization based on deep learning, Journal of Advances in Modeling Earth Systems, 12(9), e2020MS002076.
[3] Y. Zhu et. al (2022), Physics-informed deep-learning parameterization of ocean vertical mixing improves climate simulations, National Science Review, 9(8), nwac044.
[4] R. Irie et. al (2024), Parameterizing ocean vertical mixing using deep learning trained from high-resolution simulations, EGU General Assembly 2024, EGU24-2297.
[5] R. Irie et. al (2024), Optimizing a deep-learning model for parameterizing submesoscale phenomena in an ocean simulator, Workshop on Scientific Machine Learning and Its Industrial Applications.

How to cite: Irie, R., Stewart, H., Hisada, M., and Yaguchi, T.: Performance evaluation and optimization of a deep learning parameterization method trained from submesoscale-permitting ocean simulations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4606, https://doi.org/10.5194/egusphere-egu25-4606, 2025.

Gas hydrate is an important future alternative marine energy resource to fossil fuels, with the advantages of high energy, large reserves, wide distribution, and shallow burial. Accurate identification of gas hydrate reservoirs and estimation of hydrate saturation are the prerequisites for the development and utilization of gas hydrate resources. This research focuses on the difficult issues of hydrate identification, combined with the multidisciplinary technology of ocean-geology-artificial intelligence (AI). The effective hydrate formation identification technology method is studied and put forward based on the geophysical attributes. The method has been verified in the Dongsha area of the northern South China Sea. This study uses machine learning algorithms to analyze whether the sediment contains gas hydrates. Several commonly used machine learning algorithms are selected, such as random forest, Bagging, AdaBoost, and K-Nearest Neighbor (KNN). These algorithms are used to analyze the data of the P-wave velocity and density with high sensitivity to the change of hydrate. The parameters of different algorithm models are optimized through training, and the identification and classification effects of different algorithm models are compared. Finally, the results show that these algorithms could well distinguish whether there is hydrate in the sediment, among those, the KNN algorithm has a good application. The results show method based on machine learning can improve the identification accuracy of gas hydrate. The identification method of this research provides strong technical support for the subsequent exploration and development of hydrates.

How to cite: Tian, D. and Yang, S.: Identification of gas hydrate based on machine learning in the northern South China Sea , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5478, https://doi.org/10.5194/egusphere-egu25-5478, 2025.

Nighttime ship detection plays a vital role in understanding oceanic patterns and human activities in marine environments. As an observational approach in ocean science, it enables researchers to monitor vessel distribution patterns, analyze maritime traffic flows, and collect valuable data about human interactions with marine ecosystems. While the VIIRS Day-Night Band (DNB) sensor enables nighttime vessel detection from space, conventional detection methods primarily rely on threshold-based techniques, which show limitations in handling complex environmental factors such as cloud coverage and varying atmospheric conditions. To overcome these challenges, this study presents an automated ship detection approach that combines VIIRS DNB imagery with AutoML techniques. Our AutoML framework automatically optimizes model parameters and features to adapt to various environmental conditions, providing more robust detection capabilities compared to traditional threshold-based methods. The methodology incorporates AIS data for model training and validation to enhance detection accuracy. Our experimental results demonstrate improved detection performance across diverse maritime environments and weather conditions, effectively addressing the limitations of conventional threshold-based approaches. This research contributes to advancing pattern recognition in oceanic observations by providing an automated approach for identifying vessel activities in nighttime satellite imagery.

How to cite: Seong, N., Jung, O., Jung, Y., and Song, S.-H.: Nighttime Ship Detection Using VIIRS DNB Data: An AutoML Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5626, https://doi.org/10.5194/egusphere-egu25-5626, 2025.

Since the late 1800s, and especially in the last century, the natural biogeochemical cycle of lead (Pb) in the ocean has been severely perturbed by anthropogenic emissions generated by the use of leaded gasoline, waste incineration, coal combustion and non-ferrous metal smelting. Lead and its isotopes are powerful tools to study the pathways of Pb pollution from land to sea and, simultaneously, investigate biogeochemical processes in the ocean. For these reasons, the study of Pb concentrations and isotope compositions of seawater is a core part of the international marine geochemistry programme GEOTRACES. However, the scarcity and sparsity of in situ measurements of Pb concentrations and isotope compositions do not allow for a comprehensive understanding of Pb pollution pathways and marine biogeochemical cycling on a global scale.

We present here three machine learning models developed to map seawater Pb concentrations and isotope compositions leveraging the global GEOTRACES dataset together with historical data. The models are based on the non-linear regression algorithm XGBoost and use climatologies of oceanographic and atmospheric variables as features from which to predict Pb concentrations, ²⁰⁶Pb/²⁰⁷Pb, and ²⁰⁸Pb/²⁰⁷Pb. Using Shapley Additive Values (SHAP), we found that seawater temperature, atmospheric dust and black carbon, and salinity are the most important features for mapping Pb concentrations. Dissolved oxygen concentration, salinity, temperature, and atmospheric dust are the most important features for mapping ²⁰⁶Pb/²⁰⁷Pb, while atmospheric black carbon and dust, seawater temperature, and surface chlorophyll-a for ²⁰⁸Pb/²⁰⁷Pb. The output of our models shows that (i) the highest levels of pollution are found in the surface Indian Ocean, (ii) pollution from previous decades is sinking in the North Atlantic and Pacific Ocean, and (iii) waters characterised by a highly anthropogenic Pb isotope fingerprint are spreading from the Southern Ocean throughout the Southern Hemisphere at intermediate depths. The analysis of the uncertainty associated with the mapped distribution of Pb concentrations, ²⁰⁶Pb/²⁰⁷Pb, and ²⁰⁸Pb/²⁰⁷Pb suggests that the Southern Ocean is the key area to prioritise in future sampling campaigns.

How to cite: Olivelli, A., Arcucci, R., Rehkämper, M., and van de Flierdt, T.: Mapping the global distribution of lead and its isotopes in seawater with explainable machine learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6069, https://doi.org/10.5194/egusphere-egu25-6069, 2025.

Understanding the dynamics of phytoplankton communities in response to physical
environmental changes is essential for evaluating the impact of climate change on marine
ecosystems. Satellite observations provide a rich dataset spanning over two decades,
capturing physical sea surface parameters such as temperature, salinity, and sea surface
height, alongside biological insights such as ocean color. Ocean color data, in particular, is
processed to estimate sea surface chlorophyll-a concentrations — a widely recognized proxy
for phytoplankton biomass. Recent advancements in ocean color observation have further
enabled the characterization of phytoplankton community structure in terms of functional
groups or size classes.
However, linking satellite-derived physical parameters to biological indicators remains
challenging due to spatial and temporal variability.
Can physical data reliably predict patterns in ocean color, such as chlorophyll-a
concentrations and phytoplankton community structures, and potentially assess their
variations? This study addresses this question through a deep-learning approach, utilizing
an attention-based autoencoder model to learn relationships between physical variables and
ocean color data, including chlorophyll-a concentrations and phytoplankton size classes at
weekly and 1° spatial resolution.
Our trained deep-learning model effectively captures patterns and correlations between
physical parameters, chlorophyll concentrations, and phytoplankton size classes. It enables
detailed exploration of how physical factors influence biological variability across different
temporal scales. Utilizing a phytoplankton database spanning 1997–2023, this approach
demonstrates promising results in replicating chlorophyll concentrations, inferring
phytoplankton size classes, and shedding light on the potential links between physical and
biological data.
This study highlights the potential of machine learning for ecological research, contributing to
more accurate trend analyses. Understanding phytoplankton variability is critical for marine
ecosystem management, given their role in global carbon cycling. This methodology
underscores the value of deep-learning to anticipate phytoplankton dynamics under
changing environmental conditions.

How to cite: Ollier, L., ElHourany, R., and Levy, M.: Deep learning algorithm to uncover links between satellite-derived physical drivers and biological fields., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6755, https://doi.org/10.5194/egusphere-egu25-6755, 2025.

We discuss the use of systematic ‘a posteriori’ calibration in the development of complicated (but theory-based) parameterizations. With ‘a posteriori’ calibration, model error is assessed using the results of forward simulations, thereby incorporating numerical error, numerical stability, model-specific implementation details,  and alleviating the need for explicit data for all parameterized model components. We show how calibration illuminates the parameterization development trade-off between reductions in model bias, producing better predictions, and increased parametric complexity, the latter which can decrease a model’s ability to extrapolate, increase both the data requirements and computational expense of the calibration. We illustrate the importance of a posteriori calibration by describing the iterative development of CATKE, a new parameterization we develop within CliMA for the fluxes associated with small- or "micro-scale" ocean turbulent mixing on scales between 1 and 100 meters. For calibration we use Ensemble Kalman Inversion to minimize the error between a set of large eddy simulations (="the truth") and predictions of the parameterization and this way find optimal values for the free parameters. Without systematic calibration we cannot make informed choices about parameterization development because we cannot distinguish between structural error and error due to non-optimal parameter values.

How to cite: Constantinou, N., Wagner, G., Hillier, A., Silvestri, S., Souza, A., Burns, K., Hill, C., Campin, J.-M., Marshall, J., and Ferrari, R.: Calibration-driven parameterization development, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7180, https://doi.org/10.5194/egusphere-egu25-7180, 2025.

Submesoscale eddies are oceanic structures that occur on horizontal scales from 0.1-10 km, vertical scales from 0.01-1 km, and last from hours to several days. They are characterised by a Rossby number of Ro = ζ/f ~ O(1), where surface vertical vorticity ζ is similar to Coriolis frequency f. Submesoscale eddies are important in setting the stratification in the ocean surface mixed layer, mediating air-sea exchanges, and transporting energy between large and small scale motions. However, the study of submesoscale eddies on a global scale has been hindered by a shortage of global, long-term datasets. To fill this gap, we train and apply an unsupervised machine learning method adapted from the Profile Classification Model (PCM) to density profiles collected by Argo floats over global ocean from 2000-2021, producing the first global observational climatology of submesoscale activity. The adapted PCM identifies regions with high submesoscale activity using solely the density profiles and without any additional information on the velocity, location, or horizontal density gradients. The climatology shows that submesoscale activity peaks in spring in both hemispheres and lags behind the maxima of mixed layer depth by one month, suggesting that submesoscale eddies play important role in re-stratifying the mixed layer. Hotspots of submesoscale activity can be found in the Norwegian Sea and the Drake Passage in spring. This observational reconstruction of submesoscale activity enables the study of submesoscale distribution, seasonality, and inter-annual variation on a global scale.

How to cite: Yao, L. and Taylor, J. R.: Global Climatology of Submesoscale Activity Using Machine Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7671, https://doi.org/10.5194/egusphere-egu25-7671, 2025.

Accurate ocean forecasting models are crucial for both scientific research and practical application, such as understanding ocean dynamics and efficient ship route planning. While traditional numerical ocean models have proven effective, they require substantial computational resources due to the complexity of solving partial differential equations. In recent years, data-driven weather forecasting models have demonstrated their ability to provide accurate predictions at lower computational costs compared to conventional numerical weather prediction models while their application to ocean forecasting remains limited. 

This study explores a data-driven ocean forecasting model for 10-day global forecasting, employing a multi-scale graph neural network (GNN) to capture the multi-scale features of ocean variables while incorporating graph structures that account for land masks. To reflect the effects of atmospheric forcing, surface atmospheric variables are combined with ocean variables and used as GNN’s node input features. The model was initially trained on paired reanalysis data samples with a 1-day interval to minimize the mean squared error. Subsequently, it was fine-tuned using auto-regressive rollouts across multiple time steps. The forecasting process involves autoregressive steps, where the predicted ocean variables from the previous step and weather forecasting variables provided by an operational center are used as inputs for the next step.

Preliminary experiments comparing the proposed model with persistent forecasts showed the skillfulness of the proposed model. Sensitivity experiments were conducted to evaluate the impact of atmospheric forcing by replacing weather forecasting data with climatological data. The evaluation was conducted over a one-year period across the global ocean employing reanalysis data as references. The results showed that using weather forecasting data improved the accuracy of surface ocean variable predictions compared to using climatology.  Specifically, the RMSE was reduced by 6.6%, 6.2%, and 1.0% for 3-day-ahead, 5-day-ahead, and 10-day-ahead forecasts, respectively, representing the median improvement across the period and variables.  The improvements varied across variables; for instance, salinity showed a consistent improvement of almost 1% across all lead times, whereas northward velocity showed greater improvements at shorter lead times, such as an improvement of 22% at 3-day-ahead forecasts.

The results indicate that it is crucial for data-driven ocean models to incorporate atmospheric forcing, similar to numerical ocean models. These findings suggest that the multi-scale GNN-based ocean forecasting model that integrates atmospheric forcing offers a potential approach for 10-day global ocean forecasting.

How to cite: Hirabayashi, Y., Matsuoka, D., and Kimura, K.: Data-driven Ocean Forecasting Models with Multi-Scale Graph Neural Networks for 10-day Global Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8030, https://doi.org/10.5194/egusphere-egu25-8030, 2025.

The improvement of remote sensing systems together with the emergence of new model fitting algorithms based on sophisticated methods, such as machine-learning techniques, have allowed the determination of the partial pressure of carbon dioxide (pCO_2,sw) in the Canary Islands waters based on mathematical modeling. Among all the fitted models, the most powerful one seems to be the bootstrap aggregation (bagging), giving an RMSE < 6 µatm (R² > 0.95), although the multilinear regression (MLR), neural network (NN) and categorical boosting (CatBoost) also have a good predictive performance, with RMSE ranging from 9 to 13 µatm for 360 < pCO_2,sw < 481 µatm. Using the most reliable model that uses sea surface temperature (SST), Chlorophyll a (Chla), and mixed layer depth (MLD), it was determined that during the period comprised between 2019 and 2024, the Canary basin behaved as a slight net sink of atmospheric CO₂, with an average daily flux of -1.45 ± 0.08 mmol m^-2 d^-1, resulting in the sequestration of -2.59 ± 0.15 Tg CO₂ yr^-1.

How to cite: González-Dávila, M., Sánchez-Mendoza, I., González-Santana, D., Curbelo-Hernández, D., Estupiñan, D., Suarez de Tangil, M., González, A. G., and Santana-Casiano, J. M.: Modeling pCO2,sw in the Canary Islands region based on satellite measurements, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8349, https://doi.org/10.5194/egusphere-egu25-8349, 2025.

Shipwrecks have long fascinated people with their stories of mysteries and hidden treasures. UNESCO estimates that more than three million shipwrecks lie undiscovered in the world’s oceans and lakes, yet less than 10% of these have been precisely located. Beyond their historical and archaeological significance, shipwrecks can pose significant environmental threats. Instead of treasures, they often conceal harmful substances like fuels and corroded heavy metals, which, if released, can harm surrounding ecosystems and nearby communities.

This study introduces an innovative artificial intelligence (AI) approach, leveraging convolutional neural networks (CNNs) and open-access remote sensing data, to detect and map shipwrecks in remote coral reefs. The method is designed to identify wrecks based on the environmental footprint they leave, referred to as "Black Reefs", even in cases where the shipwreck itself has completely degraded.

One of the primary challenges was the limited availability of known black reef locations, which restricted the training dataset. To address this, a supervised fully convolutional neural network architecture, called SimpleNet, was employed. This architecture is specifically suited for scenarios with small labelled datasets. From a shortlist of eight suitable reefs (e.g., Kenn, Nikumaroro, Kingman, Kanton, and Rose), five were used for generating training and evaluation data, while the remaining were excluded due to low-resolution imagery or cloud interference.

Image tiles of 256 x 256 x 3 bands were extracted from the training reefs, resulting in approximately 1,600 labelled images. For evaluation, small sections of Kenn and Rose reefs were used to train the model, while other portions served as test datasets. Training was conducted using the IRIDIS supercomputer at the University of Southampton, utilizing 12 CPUs, one node with 264 GB of memory, and MATLAB 9.6 (2019b). The training process took approximately two hours.

The results demonstrate that even with limited training data, the SimpleNet architecture, featuring just eight fully convolutional layers, can efficiently identify and classify black reefs, indicating the presence of shipwrecks. Moreover, the algorithm provides a tool for monitoring reef discoloration and assessing ecological impacts over time through time-series imagery.

This study underscores the potential of AI-driven methods to enhance shipwreck detection and environmental monitoring, offering an efficient, cost-effective solution for tackling the challenges posed by limited ground data and inaccessible regions.

How to cite: Karamitrou, A., Sturt, F., and Bogiatzis, P.: Hidden Wrecks and Black Reefs: Harnessing AI to Unveil Maritime Mysteries and Environmental Risks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9298, https://doi.org/10.5194/egusphere-egu25-9298, 2025.

How to cite: Laloue, A., Anadon, C., Treboutte, A., Ballarotta, M., Pujol, M.-I., and Fablet, R.: Level 4 global topography mapping with 4DVarNet, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9778, https://doi.org/10.5194/egusphere-egu25-9778, 2025.

Understanding the response to climate change of the Venice Lagoon is fundamental for the conservation and sustainable management of a vulnerable environment, with important ecological and socio-economic consequences. Deterministic dynamic models that can reproduce the behavior of the lagoon have a very high computational cost, that limits substantially their applicability, particularly considering the multiple and multidecadal simulations required to analyses climate change. This study explores the use of artificial neural networks (ANNs) to model the relationships between climate drivers and key parameters (temperature and salinity) of the Venice lagoon to understand their different dynamics within the lagoon environment. We carry on a sensitivity study on the various drivers utilized and examine the simultaneous presence of different response patterns within the lagoon. The analysis is based on the combination in situ observations of the lagoon water temperature and salinity with large-scale data from the Copernicus Marine Services’ reanalysis  to estimate how the main physical parameters of the lagoons are driven by key climatic drivers. The sensitivity analysis was conducted by excluding from the ANN or randomizing single drivers to assess their importance for describing the variability of the lagoon environment. This analysis allow to identify three clusters, defining three areas of the lagoon, whose differences that can be physically interpreted. The riverine cluster (central/northern lagoon) is influenced by the presence of small tributaries and, consequently, by local precipitation; The marine cluster is located in the part of the lagoon near the sea outlets, where salinity and temperature values are strongly influenced by marine salinity and temperature; The mixed cluster  (in the south lagoon) where both the marine and riverine regimes overlap with comparable effects on salinity and temperature.

Financial support from ICSC – Centro Nazionale di Ricerca in High Performance Computing, Big Data and Quantum Computing, funded by European Union – NextGenerationEU. Project code CN_00000033, CUP C83C22000560007 and  from NBFC – National Biodiversity Future Center, funded by European Union – NextGenerationEU. Project code CN_00000033, CUP F87G22000290001

How to cite: Bozzeda, F., Sigovini, M., and Lionello, P.: Using artificial intelligence for exploring the climatic drivers of the Venice Lagoon environmental variability and response to climate change., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9911, https://doi.org/10.5194/egusphere-egu25-9911, 2025.

Measuring sea surface currents (SSC) directly is challenging. Instead, SSC are often inferred from indirect measurements like altimetry. However, altimetry-based methods only provide large-scale (>100 km) geostrophically-balanced velocity estimates of SSC. Here, we present a statistical inversion model to predict fine-scale SSC using remotely sensed sea surface temperature (SST) data. Our approach employs Gaussian Process (GP) regression, where the GP is informed by a two-dimensional tracer transport equation. This method yields a predictive distribution of SSC, from which we can generate an ensemble of surface currents to derive both predictions and prediction uncertainties. Our approach incorporates prior knowledge of the SSC length scales and variances that appear in the covariance function of the GP, which are then estimated from the SST data. The framework naturally handles noisy and incomplete SST data (e.g., due to cloud cover), without the need for pre-filtering.  We validate the inversion model through an observing system simulation experiment (OSSE), which demonstrates that GP-based statistical inversion outperforms existing methods, especially when the measurement signal-to-noise ratio is low.  When applied to Himawari-9 satellite SST data over the Australian North-West Shelf, our method successfully resolves SSC down to the sub-mesoscale. We anticipate our framework being used to improve understanding of fine-scale ocean dynamics, and to facilitate the coherent propagation of uncertainty into downstream applications such as ocean particle tracking.

How to cite: de Kreij, R., Zammit Mangion, A., Rayson, M., Jones, N., and Zulberti, A.: Statistical inversion of surface tracers to infer fine-scale near-surface ocean currents, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10550, https://doi.org/10.5194/egusphere-egu25-10550, 2025.

Ocean salinity is a fundamental variable that determines seawater density and, therefore, stratification and oceanic dynamics. To understand how salinity is affected and how it contributes to ocean processes, its variability must be studied, particularly through in situ observations. Unfortunately, while temperature observations were limited during the 20th century, salinity observations were even sparser, as some instruments were designed to measure temperature only. The development of the Argo observing system since 2002 has improved sampling and reduced the disparity between both variables, enabling better assessment of salinity variability over the past 20 years at interannual to decadal scales. In this study, we estimate salinity covariability with temperature from the Argo period to reconstruct monthly subsurface salinity fields, in the tropical Pacific between 1930 and 2001, leveraging temperature observations. The analysis is performed using the data-driven RedAnDA method, which combines Data Assimilation, Analog Prediction, and Reduced-space Interpolation, first validated using synthetic data. We reconstruct the 20th century interannual variability of salinity associated with ENSO events both at the surface and in the subsurface. Notably, thanks to the coupling with temperature, the representation of stratification and its modulation by vertical salinity gradients is enhanced. This new method and product provide for the first time the possibility to extend the hydrological time series consistently in the past, offering potential new insights into mechanisms generating decadal variability in the Pacific.

How to cite: Oulhen, E., Kolodziejczyk, N., Tandeo, P., Blanke, B., and Sévellec, F.: Reconstructing historical salinity fields over the 20th century using a data-driven method and Argo data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10637, https://doi.org/10.5194/egusphere-egu25-10637, 2025.

Through global warming, ocean deoxygenation is considered as a major concern since it consequently reduces the quality and the quantity of suitable habitats for marine life. Eutrophication plays a major role in its depletion which enhances respiration at different depths. Many species such as fishes, benthic worms or even plankton are affected by this phenomenon.

This study aims to get a better understanding of benthic worm species on the continental shelf of the Black Sea which is well known for high frequency oxic stresses. Our main objective is to map species through their biological traits (i.e. body length, burial depth, reproductive frequency…)  in order to assess their vulnerability towards environmental variations that occur at this location. 

Unfortunately, in the oceanographic field, one of the major issues is the sparsity of in-situ observations, especially when it comes to benthic biology. Therefore, we have decided to use a multivariate approach allowing us to use related datasets with significantly better spatial and temporal coverage. This multivariate approach is implemented using deep learning in order to get complete maps of traits on our domain. An adapted convolutional neural network allowing to capture non-linearities is used to reconstruct the traits repartitions. 

Thus, as an input for the neural network, we consider our traits dataset and environmental variables which are likely to enhance their reconstruction; Surface currents, particulate organic carbon, oxygen concentration and bathymetry are considered. A chosen period from 2008 to 2017 is selected. Traits datasets are located by stations (238) and were constructed through fuzzy coding and rescaled by their biomass. 

The neural network architecture is composed of an encoder and a decoder where the encoder considers a gappy and non-gridded dataset. The encoder uses a series of convolutional layers followed by max pooling layers which reduce the size of the dataset. The decoder does essentially the reverse operation by considering convolutional and interpolation layers. 

In order to avoid overfitting, the model has skip connections which ensure to keep information from the input dataset. For additional information please refer to Barth et al 2022. The model gives the reconstructed trait repartition and the standard error of the reconstruction.

This study will be helpful in the understanding of benthic traits repartition and will aim to link their patterns to environmental factors. This will help to get a deeper understanding of the ecological role and functions of this poorly known ecosystem. This work is carried in the frame of NECCTON European project.

How to cite: Dechenne, A., Chevalier, S., Gregoire, M., Alvera-Azcarate, A., and Barth, A.: Deriving benthic traits through deep learning methods, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10653, https://doi.org/10.5194/egusphere-egu25-10653, 2025.

Satellite observations provide a global or near-global coverage of the World Ocean. They are however affected by clouds (among others), which severely reduce their spatial coverage. Different methods have been proposed in the literature to reconstruct missing data in satellite observations. For many applications of satellite observations, it has been increasingly important to accurately reflect the underlying uncertainty of the reconstructed observations. In this study, we investigate the use of a denoising diffusion model to reconstruct missing observations. Such methods can naturally provide an ensemble of reconstructions where each member is spatially coherent with the scales of variability and with the available data. Rather than providing a single reconstruction, an ensemble of possible reconstructions can be computed, and the ensemble spread reflects the underlying uncertainty. We show how this method can be trained from a collection of satellite data without requiring a prior interpolation of missing data and without resorting to data from a numerical model. The reconstruction method is tested with chlorophyll a concentration from the Ocean and Land Colour Instrument (OLCI) sensor (aboard the satellites Sentinel-3A and Sentinel-3B) on a small area of the Black Sea and compared with the neural network DINCAE (Data-INterpolating Convolutional Auto-Encoder).  The quality of the reconstruction is assessed using independent test data. 

The spatial scales of the reconstructed data are assessed via a variogram, and the accuracy and statistical validity of the reconstructed ensemble are quantified using the continuous ranked probability score and its decomposition into reliability, resolution, and uncertainty.

The diffusion method compared favorably against the U-Net DINCAE. The RMSE of the reconstructed data using the denoising diffusion model was smaller than the corresponding reconstruction of DINCAE. The main advantage of the diffusion model is, however, the ability to reproduce an ensemble of possible reconstructed conditions on the available data. Each of these reconstructions contains small-scale information comparable to the scales of variability in the original data, avoiding a common problem where the results of U-Net and autoencoders produce images that are too smooth, as the information on small scales can typically not be recovered under clouds with a certain extent. The overall conclusion is robust when applying this technique to other areas of the Black Sea.

The ensembles of reconstructed data generated by the diffusion model can be used, for example, in the detection of gradients and fronts in the satellite images or in the estimation of the error in derived quantities, where information on how the error is correlated in space is also needed.

How to cite: Barth, A., Brajard, J., Alvera-Azcárate, A., Mohamed, B., Troupin, C., and Beckers, J.-M.: Reconstruction of missing satellite data using a Probabilistic Denoising Diffusion Model applied to chlorophyll a concentration, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11139, https://doi.org/10.5194/egusphere-egu25-11139, 2025.

Motivation The increasing amount of plastic debris in the oceans calls for quick action to prevent irreversibly damaging our world’s largest ecosystem. To this end, tracking plastic debris and understanding its dynamics could facilitate collection campaigns and help monitor the evolution of the threat. To achieve this goal, accurate models are necessary to predict the dynamics of floating objects at the ocean surface, which are subject to currents and winds. Physical models and remote sensing data estimate these influencing forces. However, using them directly in process-based models still leads to a significant gap between the true dynamics and the predicted trajectory. Hence, we aim to minimize this gap by resorting to data-driven machine-learning methods.

Data We can identify two different scenarios where the dynamics of floating objects differ: trajectories close to coastal regions and trajectories in the open ocean. As a consequence, we focus on two different datasets: the first aims to predict dynamics in coastal regions for 24 hours. The second focuses on open-ocean dynamics, where we try to predict trajectories for multiple days. As target variables, we use data from the Global Drifter program, which contains several thousand GPS-tracked free-floating buoys. The contextual information about the ocean surface current is extracted from Copernicus Marine and HYCOM. Wind data is taken from ERA5.

Approach We develop a denoising diffusion model that generates multiple trajectories based on surface current and wind, as provided by physical models. In contrast to the unstructured i.i.d. Gaussian noise in standard denoising diffusion, we use a more suitable process: Brownian motion noise, which has a small variance close to the start of the trajectories and increases with time. The denoiser model is autoregressive and based on a multilayer recurrent neural network that iteratively learns to remove the noise from random realizations of this Brownian motion.

Results We found that the model not only outperforms physical models on the coastal dataset but also provides a posterior distribution of the predicted trajectories, thus offering a measure of uncertainty without additional overhead.

How to cite: Donner, C., Goshtasbpour, S., Dalsasso, E., Volpi, M., Russwurm, M., and Tuia, D.: Autoregressive denoising diffusion for predicting trajectories of floating objects in oceans, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11223, https://doi.org/10.5194/egusphere-egu25-11223, 2025.

Biogeochemical models are computational approximations of systems of differential equations used to represent and predict biogeochemical constituents of the ocean. These might include carbon and nutrients cycles, and its interactions with biological components, such as different types of plankton. Unfortunately, these models tend to be constrained by their complex parametrization and computational cost, limiting their practical application and scalability.

Autoencoders are neural networks that are trained to learn a compressed representation of a dataset, typically with the goal of reconstructing the input to its original or a specified target dimension. But the bottleneck of this compression, or the latent space of the autoencoder, can offer interesting insights into the dominant features of the system.

Here we train different types of autoencoders to capture the main spatiotemporal dynamics from data modeled by the biogeochemical analysis BIO4 (based on NEMO-PISCES). This not only provides a basis for the development of computationally efficient emulators, but it can help us detect patterns and relationships that might not be immediately apparent in the high-dimensional output of the model. This offers interesting insights into how the model actually captures its constituting components. 

Such compressed representations can also be used for parameter sensitivity analysis, to develop data assimilation frameworks, and as tools for uncertainty quantification and outlier detection.

How to cite: Martinez Balbontin, G. and Ciavatta, S.: Peeking Into a Marine Biogeochemical Model with an Autoencoder , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11622, https://doi.org/10.5194/egusphere-egu25-11622, 2025.

Mixed layer eddies (MLE) are submesoscale structures, characterized by spatial and temporal scales of O(10 km) and O(1 day), generated by mixed layer instability under conditions of strong horizontal buoyancy gradient and weak stratification.  MLE produces mixed layer restratification, which has important implications for global ocean and climate dynamics. Existing parameterizations represent MLE effects with a streamfunction that depends on the horizontal buoyancy gradient, mixed layer depth, and the Coriolis parameter. Machine learning techniques have recently been proposed for improving existing MLE parameterizations. Bodner et al., (2024) proposed an approach for predicting submesoscale vertical buoyancy fluxes using a convolutional neural network (CNN), showing an improvement compared to previous parameterizations.

In this study,  we analyze the impact of a new MLE parameterization - based on Bodner et al. (2024) - in a global ocean model simulation performed with NEMO (eORCA25). The implementation of the CNN parameterization in NEMO is performed through EOPHIS (https://github.com/meom-group/eophis/). The CNN simulation (MLE-CNN) is compared with a simulation with a standard  parametrization and a simulation without MLE parameterization. With the CNN parameterization, maximum winter mixed layer depths are reduced by 10% with respect to the simulation without parameterization, which is comparable to the reduction obtained with the standard parameterization. The CNN parameterization differs from the standard parameterization in terms of  spatial variability.  For example, in the tropical region, the CNN produces a vertical heat flux across the mixed layer that can reach twice the magnitude of the standard parameterization. Mixed layer depth from simulations will be compared with observations. 

How to cite: Contreras, M., Barge, A., Le Sommer, J., Bodner, A., and Balwada, D.: Assessing the impact of the new mixed layer eddy parameterization based on machine learning in NEMO, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11727, https://doi.org/10.5194/egusphere-egu25-11727, 2025.

Accurate ocean forecasting is essential for a range of critical applications, from maritime safety to climate adaptation strategies. Given the inherent uncertainties in ocean dynamics, the ability to predict a range of probable ocean states is key to informed decision-making. Here, we present MerCast, a probabilistic ocean forecasting model designed to redefine global-scale prediction by quantifying uncertainty in ocean state estimates. Trained on decades of high-resolution reanalysis products, MerCast integrates diffusion models to generate ensembles of daily forecasts at 1/4-degree resolution, dynamically capturing local-global interactions while preserving fine-scale ocean features essential for accurate predictions.

MerCast's  performance is rigorously evaluated using an array of metrics tailored for stochastic forecasting systems, including ensemble spread, probabilistic error assessments, and metrics designed for process-oriented evaluations. Initial results highlight MerCast's skill in forecasting critical variables such as sea surface height, temperature, salinity, and ocean currents, with superior resilience to error accumulation over extended forecast horizons. This work establishes a foundational step toward integrating probabilistic methods in operational ocean forecasting, bridging the gap between efficiency, accuracy, and uncertainty quantification.

How to cite: El Aouni, A., Ruggiero, G., Gaudel, Q., Van Gennip, S., Drillet, Y., and Drevillon, M.: Probabilistic Global Ocean Forecasting Through Diffusion-Based Ensembles, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12581, https://doi.org/10.5194/egusphere-egu25-12581, 2025.

Big data has become increasingly important in marine geoscience, where in situ measurements are often limited, leaving large portions of the seafloor unsampled. To address this gap, we present a data-driven approach that leverages non-parametric machine learning algorithms—specifically, an ensemble of k-Nearest Neighbors (kNN) and Random Forest regressors—to predict a global geospatial prediction of median grain size (D50) at a 2-arc minute resolution. Our methodology incorporates parametric uncertainty quantification in the form of distance-to-nearest-neighbor metrics in feature space, thereby creating spatially explicit uncertainty maps that highlight regions where additional data collection would most effectively improve model predictions. This emphasis on parametric uncertainty serves as a roadmap for data-driven exploration, reducing the time, energy, and cost associated with collecting or curating a comprehensive dataset.

We train the model on ~40,000 publicly available, seafloor grain size measurements and iteratively optimize hyperparameters based on prediction error and out-of-sample validation. The final model is a global prediction of seafloor grain size with a correlation of ~0.65 between observed and predicted grain size values. We also apply a ranked noise grid analysis to select predictor variables that minimize the overall predictive error, ensuring the feature set is robust and agnostic to human bias.

Regions with sparse data coverage or atypical geological conditions manifest as areas of high uncertainty, underscoring the need for targeted sampling. By mapping this uncertainty, our framework facilitates strategic data acquisition efforts and reduces curation time and cost. We demonstrate the impact of sampling high uncertainty regions on not only improving predictions in the newly sampled geographical location but are also geologically similar (close in parameter space) around the globe. In doing so, it demonstrates how the synergy between machine learning approaches and systematic data-driven exploration can enhance the dependency of global seafloor property models. Our predicted grain size map provides a proxy for further regional and global studies that rely on grain size measurements, while more broadly highlighting the transformative potential of machine learning methods to refine our approach to data exploration and curation.

How to cite: Renzaglia, J., Lee, T., and Le, A.: Global Seafloor Grain-Size Prediction: A Data-Driven Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12879, https://doi.org/10.5194/egusphere-egu25-12879, 2025.

The effective spatial and temporal scales resolved by Earth System Models (ESMs) remain a key limitation in reducing uncertainties in climate projections. While increasing model resolution is computationally prohibitive, machine learning (ML)-based parameterizations offer a promising alternative. However, these approaches often face generalization challenges in ‘out-of-sample’ scenarios, leading to numerical instabilities when integrated into ESMs. In this study, we aim to tackle these challenges by developing a data-driven discretization neural network for multidirectional advection in ocean models. The canonical 1D advection problem is revisited by using neural networks to predict the coefficients of a three-node stencil trained on high-resolution solutions projected onto coarser spatial and temporal grids. Conventional discretizations generalize to all scalar fields, while the data-driven approach is, by construction, tied to the training data. First, it is shown that we can normalize inputs with min-max scaling to achieve generalization, while training on coarsened high-resolution data across multiple grid configurations reduces sensitivity to time steps and mesh resolution. We find that coarsening based on triangular test functions, instead of averaging, enables unique mapping of the fine-scale variations of high-resolution solutions, leading to monotonicity of the neural network. Hybrid ML discretizations that predict advective fluxes are investigated, with a focus on enforcing desirable numerical properties—such as monotonicity, accuracy, and stability. Finally, we aim to test the numerical and generalization properties of the new data-driven discretization on 2D geostrophic flows. These results provide guidance for the development of better end-to-end data-driven parameterizations and discretizations in ESMs.

How to cite: Nasser, A.-A. and Adcroft, A.: Generalizing machine-learned discretization for climate simulations: addressing ‘out-of-sample’ challenges for 2D data-driven advection discretization, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12958, https://doi.org/10.5194/egusphere-egu25-12958, 2025.

The dynamics of the ocean are dictated by processes that occur over a wide spectrum of scales. Of particular importance are the motions that occur at and around the Rossby radius of deformation, between approximately 10 and 1000 km, so called mesoscale eddies. Mesoscales exchange energy with large-scale ocean currents, thus influencing global ocean circulation. Mesoscale eddies extract potential energy (PE) from the large scale via baroclinic instability, and transfer kinetic energy (KE) upscale via the backscatter effect (inverse cascade). Accurately capturing the global ocean circulation, and the role of mesoscale eddies, is imperative in the development of reliable climate models. However, the resolution required to resolve mesoscales and their contribution to the global ocean energy cycle is far too computationally expensive, particularly for long climate integrations or large ensembles. Thus, the contributions of mesoscale eddies to the energy cycle must be parameterised in terms of the coarse-resolution flow variables of climate models. 

Most parameterisations of mesoscale eddies have focussed on resolving individual aspects of the energy cycle. Our approach aims to simultaneously address the downscale transfer of PE and the upscale transfer of KE by leveraging high-resolution simulations and machine learning. This endeavour relies on a theoretical framework that projects the buoyancy flux onto the momentum equations, resulting in an eddy forcing captured by the divergence of the Eliassen-Palm (EP) flux tensor. We develop our parameterisation using the idealised two-layer double-gyre (DG) configuration of MOM6 (ocean component of GFDL + NCAR model). High-resolution DG data is used to train an artificial neural network offline on the correlation between spatially-filtered (large scale) flow features with EP fluxes (subgrid-scale forcing). This parameterisation is shown to improve the representation of the eddy energy cycle in a DG configuration of MOM6. Our results are part of an ongoing effort towards a comprehensive parameterisation capable of capturing the entirety of the mesoscale eddy energy cycle. 

How to cite: Everard, K., Perezhogin, P., Balwada, D., and Zanna, L.: Leveraging machine learning to parameterise ocean mesoscale eddies, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13051, https://doi.org/10.5194/egusphere-egu25-13051, 2025.

As an ocean and climate modeller, I propose to expose a few venues of ocean modelling where Machine Learning (ML) is expected to break through persistent challenges. My prime target is the numerical representation of the global ocean, with distinguishable coarse spatial scale (25 to 100 km) and long duration (at least 100 years). Observations are not sufficient (too sparse in space, particularly at depth, and too short in time, spanning only the last few decades) to be used directly as the sole ground truth. Hence it is compulsory to consider perfect model set-ups, besides training on observed database. Current challenges in ocean modelling that ML could contribute to solving, are the following : equilibration of simulations, quantification of sensitivity to parameters, parameterizations of unresolved processes (due to reduced spatial resolution and/or complexity) and quantification of structural uncertainties. I will introduce a few ML-based solutions to these challenges based on recent bibliography and my own activities. Overall, we need to build capacity in bridging the gaps between these centennial global ocean simulations, useful for climate applications, process models at regional scale, global ocean hindcasts (simulations with data assimilation), large eddy simulations and models of the past, present and future climate. To reach this goal, I advocate combining various ML architectures, factoring in uncertainties of every pieces of this hierarchy.

How to cite: Deshayes, J.: Ocean models for climate applications : progress expected from Machine Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13540, https://doi.org/10.5194/egusphere-egu25-13540, 2025.

Satellite observations provide indispensable data that is assimilated into numerical ocean models to correct errors and biases. Traditionally, sea surface height (SSH) from satellite altimeter tracks, sea surface temperature (SST), and more recently, sea surface salinity (SSS), have been assimilated into these models. Temperature and salinity are part of the governing equations of ocean dynamics, and SSH is directly derived from the state of the resolved ocean, making these variables a first choice for data assimilation. However, satellite-derived Chlorophyll-a (Chl-a) data, which offer high-resolution information, is not typically assimilated. This is primarily because this variable is not solved by the physical models, and the biochemical models that simulate broader marine ecosystems, including phytoplankton dynamics and nutrient cycles which do estimate Chl-a, are computationally expensive and not used in operational models.

In this study, we utilize a ten-year free run of a biochemical ocean model of the Gulf of Mexico to simulate satellite observations, including altimeter tracks, SST,  SSS, and Chl-a. We trained and tested various machine learning architectures, including Convolutional Neural Networks (CNNs), Autoregressive Convolutional Neural Networks (AR-CNNs), and Vision Transformers, to learn the relationship between these variables and the SSH. The trained models were then used to estimate sea surface height from the simulated observations to estimate the current and future state of the sea surface height, leveraging the autoregressive properties of one of the tested architectures. Our results demonstrate that this approach outperforms the traditional interpolations in metrics like the RMSE. Finally, we applied the best-performing models to real satellite observations, highlighting the potential of improving SSH estimation quality.

How to cite: Velasco-Zavala, J., Zavala-Romero, O., Sheinbaum, J., Miranda, J., Hiron, L., Bozec, A., Bulusu, S., and Chassignet, E.: Enhancing sea surface height estimation using satellite-derived chlorophyll-a and temperature data via machine learning: a case study in the Gulf of Mexico, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13573, https://doi.org/10.5194/egusphere-egu25-13573, 2025.

OceanRep proposes a novel AI foundation model for ocean dynamics, a cornerstone for understanding and predicting climate change. Inspired by the success of AtmoRep, a deep learning model for atmospheric dynamics, OceanRep seeks to extend this framework to the ocean. In order to leverage transformer models and large-scale, multi-resolution oceanographic data (e.g., from ocean model FESOM2), the design is based on vision transformers, modified to handle four-dimensional data represented by space-time tokens, and with a U-net-type backbone to capture intricate interactions within the ocean system. For pre-training, BERT-style masking is used.

Preliminary results demonstrate OceanRep's ability to generate skillful week scale forecasts using data from a 1-degree resolution FESOM2 simulation. Ultimately, the project aims to create a robust model capable of simulating ocean and sea ice dynamics over decades. This will allow for extensive numerical experimentation and rapid generation of accurate "what-if'' scenarios. These capabilities hold immense value for climate adaptation strategies, policy development, and scientific exploration of the intricate dynamics governing the Earth system.

How to cite: Nowak, K., Koldunov, N., Jung, T., Danilov, S., Lessing, C., and Luise, I.: OceanRep: A Foundation Model for Ocean Dynamics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15061, https://doi.org/10.5194/egusphere-egu25-15061, 2025.

Over the past decades, global ocean oxygen inventories have declined by 0.5–3.3% relative to historical averages, with significant uncertainties in data-sparse regions such as the South Pacific and Indian Oceans. These gaps hinder accurate estimates of deoxygenation rates, potentially leading to underestimation of its magnitude. In this context, gridded oxygen products are essential for assessing global and regional trends and projecting the impacts of deoxygenation on marine ecosystems. However, traditional Optimal Interpolation (OI) methods are known to underestimate ocean oxygen loss, particularly in poorly observed areas.

To address these limitations, we propose a novel approach to build a gridded oxygen concentration product. Specifically, we develop a neural network emulator of oxygen concentration based on temperature and salinity measurements. This neural network is then used to generate emulated oxygen concentration data, which are combined with dissolved oxygen measurements to produce a new global gridded oxygen concentration product spanning 1965 to 2022. We evaluate our product against climatological estimates from the World Ocean Atlas and other gridded oxygen products. Future work will leverage this gridded product to study the regional evolution of ocean deoxygenation, particularly in Oxygen Minimum Zone (OMZ) regions.

How to cite: Lachkar, Z. and Ouala, S.: A Novel Global Gridded Ocean Oxygen Product Derived from Neural Network Emulators (1965–2022), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15255, https://doi.org/10.5194/egusphere-egu25-15255, 2025.

Accurate high-resolution forecasting of oceanic and atmospheric states remains a critical challenge. This study introduces an AI-based regional downscaling framework employing a U-Net deep learning architecture, trained on coarse-resolution simulations. By embedding physical constraints, the model effectively bridges scales, capturing fine-grained dynamics unresolved in traditional approaches.

The framework significantly enhances computational efficiency, reducing forecast times from hours to seconds per region while maintaining high accuracy. Its integration with data-parallel computing units enables scalable multi-region applications. Applied within a coupled ocean-atmosphere-wave-tide system, the model excels in reproducing extreme events and mesoscale dynamics.

This work highlights the potential of AI in offering scalable, precise solutions for forecasting, climate science, and disaster management.

How to cite: Huang, Y., Chen, R., Ji, L., and Zhang, S.: AI-Driven Regional Downscaling for High-Resolution Oceanic and Atmospheric Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16391, https://doi.org/10.5194/egusphere-egu25-16391, 2025.

Precise forecasting of sea surface currents is crucial for diverse applications, including navigation, pollution control, and ecosystem monitoring. Traditional high-resolution hydrodynamic models like NEMO generate detailed short-term forecasts but are computationally expensive and resource-intensive. To overcome these limitations, we present sciCUN: a deep learning framework designed for surface current inference using CNN-U-Net architecture.

In summary, sciCUN utilizes the zonal and meridional wind components, mean sea level pressure, air temperature, and dew point temperature from ECMWF Reanalysis v5 (ERA5) for the current day, along with the high-resolution zonal and meridional sea surface current velocity fields from the Copernicus Marine Service Baltic Sea Physics Reanalysis for the previous day, as input features. It then generates the high-resolution zonal and meridional sea surface current velocity fields for the current day.

As a case study, sciCUN was implemented in the Gulf of Riga domain. The model was trained to capture the influence of atmospheric forcing on preceding sea surface currents over a training period spanning 1993 to 2019. Its predictive performance was subsequently validated through a 4-year testing phase (2020–2023). Results showed that while prediction accuracy was slightly lower in coastal regions near river mouths and the Irbe Strait—areas where hydrodynamic models typically employ boundary conditions—sciCUN exhibited strong overall performance. The model achieved an average Euclidean distance of 2.30 cm/s between its predictions and reference data, with an average component-wise mean absolute error of 1.45 cm/s and correlation coefficient of 92.

How to cite: Barzandeh, A., Maljutenko, I., Rikka, S., and Raudsepp, U.: A Surrogate Model for Daily Sea Surface Current Fields Prediction Using CNN-UNET , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16890, https://doi.org/10.5194/egusphere-egu25-16890, 2025.

Differentiable programming has emerged as a powerful tool in geoscientific modelling, offering new possibilities for optimization and parameter calibration. However, this approach requires the underlying physical models to be differentiable in order to compute gradients and apply optimization algorithms. In practice, current-generation geoscientific models are generally not differentiable, which limits the use of variational approaches to calibrate their parameters. In the past few years, several strategies have been proposed to overcome this limitation.

Here, we explore the use of deep learning techniques for the calibration of vertical physics schemes of current-generation ocean models. We propose to build conditional emulators of single column ocean models to approximate the gradient of their solution with respect to their physical parameters. Our baseline is a single column ocean model, implemented in Jax, which provides a differentiable framework for the calibration of ocean vertical physics schemes. We leverage this framework to generate sets of simulations for the design of deep conditional emulators of the model, and assess their ability to approximate the gradient of the model in an inverse problem setting.

We focus on several idealized cases corresponding to different forcing conditions, starting from the Kato-Philips case. It describes the evolution of a water column with no heat flux and uniform wind friction velocity. We obtain various trajectories for uniformly sampled n-uplets defining the initial conditions, friction velocity, and physical parameters. With this dataset, we train and test different kinds of neural networks, exploring architectures and losses, to make the most of temporal and spatial dependencies.

Comparison with the fully differentiable baseline solution shows that deep conditional emulators are able to predict the system states both forward and backward, with different initial and forcing conditions, and can be used to calibrate ocean model  parameters. Our results therefore illustrate how deep emulators are a potential solution to take over the non-differentiability of existing geoscientific models, and  solve inverse problems for their calibration.

How to cite: Durif, A., Mouttapa, G., Le Sommer, J., and Fablet, R.: Deep Conditional Emulators for calibrating ocean vertical physics schemes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17009, https://doi.org/10.5194/egusphere-egu25-17009, 2025.

Phytoplankton is a central component of marine ecosystems. It contributes to biogeochemical cycles by absorbing carbon through photosynthesis at the ocean surface and transporting it deeper through sinking and subduction—hence contributing to the biological carbon pump. Plankton also represents the first link in marine food webs, supporting a wide range of marine life, from other plankters to the most productive fisheries on earth.

Satellites can help monitor phytoplankton over large-scales thanks to ocean color sensors. Current products provide daily, 4 km-resolution fields of chlorophyll-a concentration (Chla, the most widely used estimator of phytoplankton biomass) as well as its distribution in a few groups, hence estimating broad community composition. To produce these operational maps, the concentration of pigments measured by HPLC (High-Performance Liquid Chromatography) from in situ samples is regressed on reflectances at a few wavelengths matched to those samples in space and time. While incredibly useful, these models still display 30% error for Chla and at least as much when predicting community composition.

Numerous studies have shown the importance of considering mesoscale ocean structures, such as fronts and eddies, as they have a significant influence on the production and distribution of phytoplankton. These structures span tens to hundreds of kilometers and can be observed through ocean color but also infrared and radio wave satellite data.

In this work, we develop a deep learning model to predict the concentration of three phytoplankton size classes: pico-, nano-, and micro-phytoplankton. The in situ values are derived from over 7000 HPLC measurements spanning the globe, from 1997 to 2021. We use a Multi-Layer Perceptron to naturally combine reflectances with other satellite-derived variables that describe ocean physics (sea surface temperature, sea level anomalies, etc.) as input. The MLP is preceded by convolutional layers to summarise arrays of the input variables covering dozens of kilometers around the in situ observations. These two approaches are meant to capture the effect of mesoscale oceanic structures on the abundance and composition of phytoplankton.

This approach improves the estimation of phytoplankton communities on a global scale. It paves the way for in-depth studies on the influence of mesoscale structure in specific oceanic regions. Furthermore, it lays the groundwork for the future integration of the temporal dimension into the model, enabling a more comprehensive representation of ecological dynamics.

How to cite: Labourdette, E., Sauzède, R., Abbas-Turki, L., and Irisson, J. O.: Extending the inputs of deep learning models to capture the mesoscale context and better predict phytoplankton community composition, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17211, https://doi.org/10.5194/egusphere-egu25-17211, 2025.

Sea surface height (SSH) data derived from satellite altimetry are widely used in data assimilation to enhance the representations of ocean currents and subsurface temperature and salinity structure. However, accurately projecting SSH onto subsurface temperature and salinity structures presents significant challenges. Consistent adjustment to temperature and salinity profiles are required to conserve the potential vorticity, which depends on the vertical density gradient. Otherwise, SSH assimilation can produce adverse effects (Fu and Zhu, 2011). Several methods have been proposed to address this issue, including the CH96(Cooper and Haines, 1996) method used by Chang et al (2023), which constructs pseudo profile derived from altimetry data by preserving density structures. However, when tidal forcing is applied to an ocean model, the CH96 method becomes challenging to use due to the significant difficulty in removing tidal signals. To overcome these limitations, this study proposes a Transformer-based machine learning approach to reconstruct T/S (Temperature and Salinity) profiles from SSH. Transformers are well-suited for capturing complex correlations through attention mechanisms (Vaswani et al., 2017), making them ideal for learning T/S profiles influenced by diverse and intricate variables. Monthly GLORYS data from 2010 to 2020 was utilized to train a model for reconstructing T/S profiles. The data was structured into 1/2° grids, where learning was conducted grid-by-grid to capture spatiotemporal variability. For improved accuracy and better incorporation of surrounding grid influences, a combination of 4D-Var techniques and CNNs was employed. This approach learns patterns by grouping four neighboring grids into a quadrilateral for joint training, ensuring that the final profiles account for interactions across grids. During prediction, the surface information of a target point is distributed to its four neighboring low-resolution grids to generate profiles, which are then interpolated into a high-resolution 1/12° grid. The final profile is computed using inverse distance weighting (IDW) interpolation, prioritizing the influence of closer profiles for spatial consistency. Model performance was validated by comparing predicted profiles with low-resolution maps for 2021–2022 over the northwest Pacific region (10°S–45°N, 120°–170°E), achieving an RMSE of 0.55 for temperature and 0.12 for salinity. The model will be further validated against in-situ observational data. We plan to conduct experiments to investigate the impact of assimilation of the reconstructed profiles and compared against CH96-derived profiles to evaluate their accuracy and advantages.

How to cite: Lee, G.-M. and Kim, Y.-H.: Machine Learning-Based Reconstruction of T/S Profiles from Satellite-Derived SSH Using Transformer Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18226, https://doi.org/10.5194/egusphere-egu25-18226, 2025.

In the last decades, mesoscale air-sea interactions have received increasing interest from the scientific community. Mesoscale thermal (sea surface temperature influence, TFB) and mechanical (oceanic surface current influence, CFB) air-sea interactions have been shown to have a strong influence on the wind up to the troposphere and on ocean dynamics. However, from an oceanic perspective, running an atmospheric model is very expensive. To overcome this issue, we have developed a convolutional neural network (CNN) that aims to reproduce the mesoscale ocean-atmosphere interactions. Training was performed with simulated data from a realistic coupled ocean-atmosphere tropical channel simulation (45°S- 45°N) using NEMO for the ocean model, WRF for the atmosphere model, and the OASIS3-MCT coupler. As a first step, the CNN was trained over two energetic regions (the Agulhas Current and the Kuroshio) to predict mesoscale surface stress anomalies from large-scale atmospheric and mesoscale oceanic inputs. Validation over the Gulf Stream and other regions shows that the CNN successfully reproduces the surface stress anomalies associated with both TFB and CFB.  In a second step, to parameterize the mesoscale ocean-atmosphere interactions, we coupled the CNN to NEMO via an Eophis library (pyOASIS) and ran a simulation over the tropical channel configuration. In this talk, we will present our main results in terms of oceanic energetics and ocean-atmosphere energy transfer.

How to cite: Ernout, N., Renault, L., Simon, E., Benshila, R., Zhang, S., and Le Sommer, J.: Toward a New Parameterization of Fine-Scale Ocean-Atmosphere Interactions Based on a Machine Learning Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18713, https://doi.org/10.5194/egusphere-egu25-18713, 2025.

The combination of Machine Learning (ML) with geoscientific models has become an active area of research, but many technical challenges still remain because of the heterogeneous nature of programming languages, library environments and hardwares. Much efforts have been made over the recent years to propose different frameworks to perform online deployment of ML components within geoscientific models. One common drawback to all these solutions is the complexity of the required software environment. The latter often relies on versioned libraries and codes, both for the geoscientific and the ML models. Thus, ensuring the reproducibility of hybrid geoscientific model experiments is challenging, as it requires describing several tools and how to deploy them. This becomes even more problematic as the number of coupling solutions for hybrid modeling increases and may be unfamiliar to the members of the different modeling communities.

Here, we introduce Morays as an example of a community-based workflow for sharing reproducible hybrid ocean model experiments. Morays uses a GitHub organization to host hybrid experiments material that leverage the OASIS coupler (https://oasis.cerfacs.fr/en), which is widely used in European climate models. Our framework is based on a Python library (https://github.com/meom-group/eophis) that facilitates the use of OASIS for deploying hybrid modeling pipelines bridging FORTRAN solvers and ML models implemented in Python. The geoscientific model and ML scripts are executed separately and exchange data through the coupling API. 

In this presentation, we will showcase several successful deployments of hybrid ocean model experiments with the NEMO ocean/sea-ice modeling framework. These experiments implement ML-based parameterizations and model correction schemes for improving different aspects of model solution (vertical physics, eddy parameterization, surface fluxes). All the experiments are shared openly in a dedicated GitHub organization (https://github.com/morays-community), as individual repositories following a standard template. We will present the material available to the community (tutorials, test cases), explain how to contribute, and discuss the broader perspective of reproducible workflow for future hybrid geoscientific models.

How to cite: Barge, A., Meunier, E., Contreras, M., Kamm, D., and Le Sommer, J.: Morays-community: a framework to share reproducible hybrid Machine Learning and Ocean modeling experiments., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18745, https://doi.org/10.5194/egusphere-egu25-18745, 2025.

Phytoplankton play a key role in maintaining marine ecosystems and regulating global carbon dioxide concentrations through photosynthesis. Thus, it is crucial to assess and understand their temporal variations. However, fluctuations of phytoplankton biomass on multi-decadal and longer timescales remain uncertain, in contrast to seasonal and interannual ones, due to the lack of long-term observations on a global scale and the uncertainties related to the complex balance of processes that control their fate. As phytoplankton growth depends on the availability of nutrients in the sunlit upper ocean, which is closely linked to the stratification of the ocean, one can assume that at first order changes in phytoplankton is related to changes in ocean and atmosphere dynamics.

Over the last few years, several conventional data-driven deterministic approaches have been trained from physical observations (used as predictors) to reconstruct satellite ocean color time series (i.e., Chlorophyll-a concentration, Chl, which is used as a proxy of the phytoplankton biomass) and investigate their multi-decadal variability. Deterministic methods, such as encoder-decoder architecture U-Net, LSTM, FourCastNet, are robust but tend to fail in capturing probabilistic uncertainty because they produce deterministic outcomes. Additionally, these methods struggle with handling extreme and highly complex real-world scenarios. This study proposes a novel application of score-based generative diffusion models to address these challenges and present a comparative analysis against U-Net and FourCastNet. Probabilistic conditional diffusion model has been pretrained on simulation data and subsequently fine-tuned to learn the parameters using satellite observation data. This generative model learns the inherited uncertainty by generating ensembles of possible Chl mapping and analyzing the variability within the ensemble. The model can then be sampled efficiently to produce realistic Chl ensembles, conditioned on physical predictors and the baseline model U-Net. The ensembles from the diffusion model show greater reliability and accuracy, particularly in extreme event classification.

Our results demonstrate that when conditioned with a U-Net (meaning this input together with eight physical predictors), diffusion behaves better than the baseline method, especially when the number of samples is increased. It is visible from the spatial maps of standard deviation that as the sample size increases, the model's predictions stabilize and become more concentrated around the mean which leads to a reduction in the spread of outcomes.

How to cite: Lakra, M., Fablet, R., Drumetz, L., Martinez, E., Pauthenet, E., and Nga Nguyen, T. T.: Probabilistic Diffusion Models for Ocean Chlorophyll-a Prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18806, https://doi.org/10.5194/egusphere-egu25-18806, 2025.

Marine Primary Production (MPP) is a key component in understanding ocean ecosystems and their atmospheric carbon sequestration capacity. However, numerous challenges exist for obtaining MPP estimates. Algorithm variability is a significant issue, since various MPP models (chlorophyll-based or carbon-based algorithms) yield divergent results. Furthermore, the lack of observational data and periodic vertical profiles of the surface ocean hinder the ability to validate and refine such models.

This work focuses on improving MPP estimations by extending the state-of-the-art Carbon, Absorption, and Fluorescence Euphotic-resolving (CAFE) net primary production model with machine learning techniques to overcome current limitations. To improve the model's accessibility and versatility to be extended with data-driven methods, the original C code was rewritten in Python, resulting in a more user-friendly version named PyCAFE [https://github.com/jcrespinesteve/PYCAFE.git]. Using PyCAFE, simulations of MPP from 2003 to 2023 were conducted, producing a comprehensive dataset for training, validation, and testing. First, we train a random forest (RF) model using 500 random locations to emulate PyCAFE and to test global upscaling of MPP estimates. Our results show that the RF model has a strong capability for extrapolating MPP predictions with high accuracy [R²=0.96]. Second, we develop a hybrid model approach to simulate MPP: the HYPE-CAFE model (HYbrid marine Primary production Estimates based on the Carbon, Absorption, and Fluorescence Euphotic-resolving model). HYPE-CAFE combines the physical processes of the PyCAFE model with a neural network predicting the light-use efficiency (LUE), i.e., MPP is calculated as the product of absorbed photons and the predicted LUE. Preliminary results indicate that HYPE-CAFE provides an improvement over the predictions made with the CAFE model alone, especially in regions with variable environmental conditions. However, the lack of observational data limits the learning process. Therefore, in a next step we test a transfer learning approach to improve MPP predictions by HYPE-CAFE.

In conclusion, this project paves the way for the development of advanced hybrid modeling approaches, such as HYPE-CAFE, for global MPP estimation, and offers a transformative avenue for deepening our understanding of global ocean productivity, particularly in the context of climate change.

How to cite: Crespin, J., Benson, V., and Winkler, A. J.: HYPE-CAFE: Towards a Hybrid Model for Improved Marine Primary Production Estimates, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19488, https://doi.org/10.5194/egusphere-egu25-19488, 2025.

Extreme marine biological events, such as harmful algal blooms and mass mortalities, are increasingly driven by climate variability and anthropogenic pressures, profoundly impacting marine ecosystems. The Black Sea, with its distinct stratification, salinity gradients, and diverse phytoplankton functional groups, is particularly vulnerable to these changes. Understanding and forecasting the interactions between physical, chemical, and biological variables in this region is crucial for effective ecosystem management.

We present a neural network-based surrogate modeling framework to analyze and predict the dynamics of the Black Sea ecosystem. A 3D convolutional encoder-decoder network is trained on simulation data (1950–2014) produced be the University of Liège, including daily basin-scale values of temperature, salinity, nutrients, chlorophyll, and phytoplankton biomass. The model processes time series of spatial maps as input and predicts chlorophyll concentrations and the distributions of phytoplankton functional groups for the subsequent two weeks.

This approach efficiently captures complex interdependencies between variables, offering a computationally efficient alternative to traditional process-based models. By perturbing input variables, the model identifies key drivers of chlorophyll variability, enabling rapid scenario testing to explore the impacts of environmental changes on the ecosystem.

Our findings demonstrate the potential of neural network-based surrogate models to advance understanding of phytoplankton dynamics and support decision-making in marine ecosystem management.

How to cite: Smith, P. A. H., Chauhan, A., Christensen, A., St. John, M., Rodrigues, F., and Mariani, P.: Understanding Drivers of Phytoplankton Variability in the Black Sea Using Convolutional Neural Networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19563, https://doi.org/10.5194/egusphere-egu25-19563, 2025.

Understanding the seasonal dynamics of plankton in the Atlantic Ocean is the first step towards the proper assessment of marine ecosystem health and productivity. Ocean colour and surface chlorophyll (chl-a) distribution serve as proxies for phytoplankton biomass, providing insights into marine food web dynamics and biogeochemical cycles. This study examines the response of the total chlorophyll concentration to physical drivers observable by remote sensing in the Atlantic Ocean using a combination of multivariate Principal Component Analysis (PCA) and deep learning models. The results show that the Sea Surface Salinity (SSS), Absolute Dynamic Topography (ADT), and Sea Surface Temperature (SST) are found to be the predominant drivers of physical variability across the ocean, with distinct spatial patterns. The clustering of the principal components identifies regions characterised by distinct physical processes. Based on these clusters, we devised a Transformer Encoder model to predict chl-a concentrations in three distinct regions. The model outperformed climatological baselines, especially in the temperate and tropical regions, though accuracy varied seasonally, with higher accuracy in winter months and increased complexity in summer due to more dynamic oceanographic conditions. A SHAP-based sensitivity analysis showed that ADT and SSS dominate chl-a variability, particularly during summer months, while SST and wind stress also contribute significantly during transitional periods. The study highlights the necessity to account for both seasonal and regional differences in predictive modelling, and it underscores the importance of continuing to develop spatio-temporal models to improve forecasting accuracy for marine ecosystem management and conservation.

How to cite: Chauhan, A., Smith, P., Rodrigues, F., Christensen, A., Nardelli, B. B., John, M. St., and Mariani, P.: Deep Learning Models to Identify Seasonal Drivers of Chlorophyll Changes in the Atlantic Ocean, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19668, https://doi.org/10.5194/egusphere-egu25-19668, 2025.

Systematic biases pose a significant challenge in ocean general circulation models, where numerical approximations, unresolved physical processes, and parameterization choices can lead to state-dependent errors. Addressing these biases is crucial for improving forecasts of the Earth’s climate system, yet remains nontrivial—particularly given the sparse nature of ocean observations, which complicates bias detection and correction.

One promising route is to harness analysis increments within a Machine Learning (ML) framework to learn state-dependent systematic errors from archived data assimilation corrections. For instance, neural networks can be used to train a model with the ocean state as input and the Data Assimilation corrections as output. By training on these increments, the ML model learns how errors systematically depend on the local physical state.

In our work, we use outputs from ocean reanalysis data using variational data assimilation and the NEMO ocean model. The ML-based correction is embedded in NEMO’s tendency equations as an additional forcing term, allowing the model to evolve more realistically by accounting for state-dependent systematic errors in temperature and salinity.

However, the sparsity of ocean observations can lead to “punctual” analysis increments that contain not only model biases but also noise from intermittent measurement coverage, errors, and initial-condition uncertainties. To mitigate this issue, we apply a two dimensional low-pass filter to remove high-frequency fluctuations in both the ocean fields and the analysis increments, preserving larger-scale patterns.

We adopt a feed-forward neural network (NN) that processes vertical profiles. By focusing on the ocean’s vertical stratification and processes, the network is trained on these filtered analysis increments and learns the non-linear relationships linking NEMO’s state variables (temperature, salinity) to the corrections identified by the variational scheme. Through this level-specific, column-oriented NN, the model more effectively adjusts for systematic errors.

In this poster, we present preliminary results on offline validation of the trained NN—predicting analysis increments on independent test data beyond the training period—without yet applying these corrections in a fully integrated forecast. Our preliminary findings show how well the NN reproduces systematic biases at various depths and in different oceanic regions, even under sparse data conditions and complex multi-scale dynamics. This demonstration highlights the potential of combining analysis increments with ML to systematically reduce model errors in next-generation ocean prediction systems, setting the stage for future work that integrates these learned corrections into an online, real-time workflow.

How to cite: Nunziante, G., Storto, A., and Yang, C.: Systematic error correction in numerical ocean models with artificial neural networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19733, https://doi.org/10.5194/egusphere-egu25-19733, 2025.

This work is dedicated to analysing the simulations of the quasi-geostrophic double-gyre model from dynamical systems point of view to discover nonlinear low-order structures in this turbulent regime. The double-gyre is simulated by a stratified quasi-geostrophic model which is solved using high-resolution CABARET scheme [1]. The statistically stationary simulations of the double-gyre model are considered for 400 years after a 100-year spin-up period. Double-gyre simulations are coarse-grained (symbolized) based on the Taken’s embedding theorem [2] which is proved promising for identifying nonlinear patterns from the stochastic background in the turbulent flow signals. To analyse the coarse-grained time series, Permutation Entropy [3-6] is deployed to quantify repetitive mutual ordering between subsequent time series values using the deviations from uniformity in the distribution of occurrences for symbolic ordinal patterns. Based on permutation entropy analysis, the large-scale double-gyre circulation and its eastward jet demonstrate highly nonlinear behaviour while smaller-scale eddies spread throughout the domain behave linearly.  The results of this dynamical system analysis are also compared with data-driven and multi-scale reduced-order models previously developed for this ocean circulation [7,8].

References:

[1] Karabasov, S.A., Berloff, P. S. & Goloviznin, V. M. (2009). CABARET in the ocean gyres, Ocean Modelling, 30(2-3), 155–168.

[2] Takens, F. (2006, October). Detecting strange attractors in turbulence. In Dynamical Systems and Turbulence, Warwick 1980: proceedings of a symposium held at the University of Warwick 1979/80 (pp. 366-381). Berlin, Heidelberg: Springer Berlin Heidelberg.

[3] Bandt, C., & Pompe, B. (2002). Permutation entropy: A natural complexity measure for time series, Physical Review Letters, 88, 174102.

[4] Rosso, O. A., Larrondo, H. A., Martin, M. T., Plastino, A., & Fuentes, M. A. (2007), Distinguishing noise from chaos, Physical Review Letters, 99 (15), 1–5.

[5] Kobayashi, W., Gotoda, H., Kandani, S., Ohmichi, Y., & Matsuyama, S. (2019). Spatiotemporal dynamics of turbulent coaxial jet analyzed by symbolic information-theory quantifiers and complex-network approach, Chaos: An Interdisciplinary Journal of Nonlinear Science, 29 (12), 123110.

[6] Gryazev, V., Riabov, V., Markesteijn, A., Armani, U., Toropov, V., & Karabasov, S. A. (2024). A Dynamical System Method for Finding Flow Structures from Jet LES Data. In 30th AIAA/CEAS Aeroacoustics Conference (2024), 3087.

[7] Naghibi, E., Armani, U., Gryazev, V., Toropov, V., & Karabasov, S., (2024). Reconstruction of the North Atlantic Double-gyre Circulation with Genetic Programming, Springer Proceedings in Mathematics and Statistics, Proceeding of ATSF Conference 2024.

[8] Naghibi, S. E., Karabasov, S. A., Jalali, M. A., & Sadati, S. H. (2019). Fast spectral solutions of the double-gyre problem in a turbulent flow regime. Applied Mathematical Modelling, 66, 745-767.

How to cite: Naghibi, E., Gryazev, V., and Karabasov, S.: A Dynamical System Approach for Finding Nonlinear Flow Structures in Double-gyre Circulation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20725, https://doi.org/10.5194/egusphere-egu25-20725, 2025.

Climate change may influence energy demand, with shifts in energy needs not only altering the energy structure but also posing challenges to the sustainability and resilience of energy systems. These impacts could further complicate the feasibility of achieving decarbonization goals. Residential energy sector is a critical component of global energy consumption. As temperature fluctuates and weather variability intensifies, households will adapt energy use to maintain comfortable living conditions. Energy consumption may increase due to climate change, but the magnitude remains uncertain. Considering various income groups around the world, residents may react to climate change heterogeneously.

Traditionally, some models use Heating Degree Days (HDD) and Cooling Degree Days (CDD) to serve as index of temperature change, which are often calculated by formulas below, where i means gridded cell, j means region,  means daily temperature, and represents comfortable temperature,  means population. First, calculate gridded HDD/CDDs as the difference between daily temperature and comfortable temperature. Then aggregate the gridded daily HDD/CDDs to region.

  (1)

   (2)

                                      (3)

However, calculation for HDD/CDDs still have several aspects that could be further improved. First, most temperature data used are predicted on SRES, and HDD/CDDs are assumed to be constant, so HDD/CDDs need to be updated to better reflect future climate change. Second, previous calculation always neglects the impact of crucial factors such as GDP when aggregating gridded temperature difference to regional level, only considering population distributional effects. Third, the difference resulted from income and climate also should be considered, for rich residents can afford more energy consumption, and long-term climate also impact response of people when faced with climate change.

Considering potential shortcomings mentioned above, we update the global HDD/CDDs of 32 regions in Global Change Analysis Model (GCAM). First, we use daily temperature data predicted by four climate models under different Shared Socioeconomic Pathways (SSPs) combining Representative Concentration Pathways (RCPs) scenarios from the Coupled Model Intercomparison Project Phase 6 (CMIP6), thus bridging the gap between climate model and GCAM. For in GCAM climate input module, HDD/CDDs are calculated based on historical climate data and are lack of fine-scale calculation. Second, our calculation adopts two weighting methods considering influence of population and GDP distribution on residential energy demand respectively. Third, beyond global-scale calculation, we refine calculation for China to the provincial level.

Fig. 1 Research Framework

Based on the updated HDD/CDDs, we use GCAM to analyze how climate change impact residential energy demand, aiming to provide scientific support for formulating policies that address the challenges posed by climate change to energy system. Our analysis offers comprehensive insights into residential energy demand change under SSPs and RCPs scenarios, accounting for income heterogeneity. These findings are informative to design effective mitigation policies in the context of climate change.

How to cite: Zhu, M., Zhao, M., Zhu, R., Mei, F., and Ou, Y.: Effects of Climate Change on Residential Energy Structure, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-503, https://doi.org/10.5194/egusphere-egu25-503, 2025.

Digital city contributes to improving resource allocation efficiency and quality of life, with a key component being the division of functional areas. This division directly influences the optimal allocation of urban spatial resources and the efficient operation of various services. However, a mismatch exists between virtual and physical city functions. For instance, many office activities do not occur in physical office spaces. In this study, Guangzhou is taken as the research area to quantify this mismatch in office spaces, utilizing mobile signaling data and POI (Point of Interest) data, and analyzing the factors influencing the mismatch. Mobile signaling data, combined with office software usage records, reveals the precise locations of office activities from a virtual perspective. POI data provides detailed records of physical office locations, while also encompassing other potential locations where office activities may take place. This study reveals the spatial characteristics and influencing factors of this mismatch in Guangzhou, provides scientific support for the development of digital cities.

How to cite: Jiang, H.: Quantifying spatial mismatch between virtual and physical office spaces, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1372, https://doi.org/10.5194/egusphere-egu25-1372, 2025.

The Vulcan Project version 4.0 emissions data product has generated all fossil fuel CO2 emissions across the US landscape, every hour, from 2010-2022 down to the scale of neighborhoods. From this complex landscape, we have extracted FFCO2 emissions for every urban area, following multiple commonly used urban definitions. The information extracted includes both Scope 1 and Scope 2 emissions with a wide array of “functional” attributes such as sector, fuel, vehicle class, building class, road class, and industrial sub-sector. Here, we analyze ~4000 US cities in terms of their size scaling properties. In particular, urban scaling properties provide novel insight into emergent properties such as the relationship between urban metabolism and urban size properties. This relationship varies by region and is indicative of the relationship between urban form and economies of scale including implications for infrastructural development and urban sprawl.

How to cite: Gurney, K., Dass, P., Lobo, J., and Shutters, S.: Scaling Properties of Carbon Emissions in US Cities: Bigger is Better, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3332, https://doi.org/10.5194/egusphere-egu25-3332, 2025.

Urban geo-scenes (UGS) are an abstraction of the basic units of cities. Understanding and functional recognition of UGS is crucial to balancing and optimizing urban spatial layout, rationally allocating urban resources, and enhancing urban resilience and vitality. To construct UGS, urban geo-objects (UGO) e.g., derived from remote sensing must be combined with semantic information, which has seldom be done so far.  Consequently, this study designed a UGS recognition framework based on multimodal deep learning. First, we use very high-resolution satellite data to derive UGOs. Second, the self-built SE-DenseNet branch is used to mine deep physical visual features and social semantics from satellite image data and auxiliary data (POI, building footprints from UGOs). Finally, we build an urban fabric graph model to mine spatial semantics between UGOs.  In addition, a spatial semantic fusion module is introduced for the collaboration and interaction of multi-modal and multi-scale features. We evaluate the effectiveness of the proposed framework in the complex Beijing and Shenzhen regions of China. The overall accuracy is 91.35% and 90.24% respectively, which is higher than the state-of-the-art multimodal methods. In addition, our study also emphasizes the key role of spatial relationships and distribution patterns of UGO in UGS recognition, and the addition of POIs and building heights improves the recognition accuracy. The multimodal UGS recognition framework based on urban fabric can more effectively understand urban functions, thereby achieving urban planning and management.

How to cite: Bao, H., Zhou, L., and Lehnert, L.: Understanding and recognition of geo-scenes based on multimodal spatial semantics to monitor complex urban systems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4066, https://doi.org/10.5194/egusphere-egu25-4066, 2025.

The fast-evolving nature of urbanization and its complex patterns require precise and interpretable machine learning models to effectively predict urban growth. To address this challenge, this study introduces a novel framework combining Hybrid Intelligence and Explainable AI (XAI), specifically Shapley Additive Explanations (SHAP) to improve model performance, robustness, and transparency. Using a weighted ensemble technique, the proposed method systemically integrates linear, tree-based, and neural network models to propose a hybrid of Elastic Net, XGBoost, and Wide & Deep Neural Network (EN-XGB-WDN) frameworks for urban growth prediction. The methodology follows a multistep approach and includes the development of the hybrid model, its evaluation for binary classification, integration of SHAP-based feature analysis to identify key drivers of urban growth and improve model interpretability, retraining of the hybrid model to increase accuracy and reduce overfitting, and validation of the proposed framework using standard evaluation metrics including accuracy, precision, recall, F1 score, and AUC. The hybrid model achieves an overall accuracy of 87.34%, a weighted F1-score of 87.18%, and an AUC of 0.9442. The SHAP analysis revealed that Drive Time (DT), Distance from Roads (DfR), and Elevation are the most impactful features to understand the dynamics of urban growth. The findings revealed how variations in specific features, such as higher DT and lower DfR, significantly affect urban growth probabilities. The hybrid model also categorized urban growth probabilities into five classes: very low (40.62%), low (23.27%), moderate (15.38%), high (12.10%), and very high (8.63%), revealing spatial patterns of urban expansion. The framework combines hybrid ensemble methods with SHAP-based explanations to significantly enhance the predictive and explanatory power of urban growth models compared to the limitations of traditional approaches. This study highlights the efficiency of integrating hybrid machine learning and Explainable AI to understand and predict complex urbanization dynamics. The outcomes offer actionable insights for policymakers and urban planners, facilitating data-driven strategies for sustainable urban development. This research demonstrates the effectiveness of hybrid intelligence coupled with Explainable AI, offering a scalable and interpretable framework to better understand and predict urbanization patterns.

How to cite: Khan, D. and Khan, N.: Hybrid Intelligence and Explainable AI for Urban Growth Prediction Modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4759, https://doi.org/10.5194/egusphere-egu25-4759, 2025.

Urban geometry plays a key role in determining the urban climate through its complex shading and trapping effects on solar radiation. As a result, urban albedo is typically lower than rural albedo, suggesting a larger solar heat gain in urban areas. As cities grow larger and more heterogeneous in geometry, quantifying the impact of this variation on albedo at the city scale requires computationally efficient models that can also resolve the 3D geometry of real cities. To this end, we developed a simplified 3D urban radiation model and used it to examine the variations in albedo due to heterogeneous geometry in the city of Shanghai. The model reduces computational complexity from O(n²) to O(n) while maintaining an accuracy within 5% compared to traditional 3D models. The case study in Shanghai shows that albedo has a linear relationship with building height but varies nonlinearly with changes in building density. The lowest albedo occurs when the building density (λ_p) is around 0.2 and the building height-to-length (H/L) ratio is 6, while occurs at λ_p > 0.3 with H/L = 1. This suggests that optimizing building geometry could improve the urban climate and potentially being used to increase the utilization of solar energy.

How to cite: Zhou, H. and Wang, K.: Development of simplified 3D urban radiation model to examine the variations of albedo due to heterogenous geometry in the city of Shanghai, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4805, https://doi.org/10.5194/egusphere-egu25-4805, 2025.

Urban heat storage (Qs) is an essential component of urban surface energy balance. Urban with 3D structure has larger surface area than rural and urban would absorb and release more energy than rural. Qs is the main factor for urban heat island (UHI) at nighttime. The quantitative contribution of Qs to UHI is still unclear, due to the lack of a spatio-temporal continuous Qs dataset. In this study, firstly, we developed an urban surface thermal inertia model considering diurnal variation of surface temperature (LST) using hourly LST of Himawari-8. Secondly, the hourly Qs at 2-km resolution in three urban agglomerations in China was simulated by a half-order time derivative method which derived from combining the one-dimensional heat diffusion equation and Fourier’s law for heat conduction, using the urban thermal inertia model and hourly Himawari-8 LST. Thirdly, the relationship between Qs and air temperature (Ta) was studied at different time scales (day and nighttime, four seasons) and different LCZs (local climate zones). The Ta was derived from the interpolation of dense automatic weather stations with more than 10000 sites in China. Finally, some urban heat mitigation measures were provided based on the above analysis. Based on the in-situ observation, the accuracy of urban thermal inertial in this study was higher than other model, RMSE, MAE, R² were improved from 4.65 K, 3.58 K and 0.88 to 1.86 K, 1.53 K and 0.97. In addition, the simulated Qs were validated by the observed Qs (the minus of net radiation, sensible and latent heat flux from in-situ flux tower, and anthropogenic heat flux simulation) in Beijing, Shanghai and Guangzhou, R² could be up to 0.92. The results showed that, Qs was more consistent with Ta at nighttime than daytime, with R² of 0.96 and 0.1, respectively. That showed that Qs is the main factor for nighttime UHI in this study area. During nighttime, the high-rise building has higher Ta than low-rise building, due to higher Qs and release more energy than low-rise. In natural surfaces, water has larger Qs and higher Ta than dense trees. The loop (between hourly Qs and hourly Ta) shape were different at different LCZs, with different loop area and loop slope. Based on the loop area and slope, we found that high-rise building had higher UHI but varied quickly, however, low-rise UHI is lower but would last longer. The water surface in nighttime is also heat source and has a longer time UHI. Therefore, the high-rise building and water surface are not conductive to alleviating the nighttime UHI.

How to cite: Li, N., Guo, F., Dou, J., Ma, Y., and Miao, S.: Remote sensing-driven analysis of hourly urban heat storage and its effects on urban heat islands in China, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5462, https://doi.org/10.5194/egusphere-egu25-5462, 2025.

China has undergone rapid urbanization in terms of population and land use in recent years. However, there are notable lags in "people-oriented" dimensions of urbanization, including urban social services, environmental sustainability, and equity. Here, considering the complex interactions of sub-components of urbanizations, we examined 16 "people-oriented" urbanization indicators across four dimensions - economic, social, environmental, and equity dimensions - from 2005 to 2020. Using methods such as paired t-tests and the evenness measurement, we analyzed and identified the dynamic relationships between these 16 indicators with population/land urbanization at multiple scales, including national, regional, urban agglomeration, and different city sizes. We found that between 2005 and 2020, China's urbanization indicators showed an overall upward trend, with changes ranging from 1.09 to 53.95 times. Among "people-oriented" urbanization indicators, economic and social indicators lagged behind land and population urbanization, while environmental indicators took the lead. The evenness index among indicators showed a "U-shaped" change pattern. Particularly since the implementation of China's New-type Urbanization Plan in 2014, the evenness index among indicators gradually increased from 35.43 to 37.39 in 2020, representing a 6.9% improvement. Looking forward, it is necessary to strengthen investment in social service systems and implement placed-based coordination strategies to promote further development and balanced growth of "people-oriented" urbanization.

How to cite: Gu, T., Huang, Q., and Hou, Y.: Do people-oriented urbanization catch up with land and population urbanization in China？, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6392, https://doi.org/10.5194/egusphere-egu25-6392, 2025.

In recent decades, urban expansion across Europe has accelerated, driving the rapid growth of rural-urban interfaces. The increasingly complex and dynamic nature of territorial transitions calls for the timely development of classification systems designed to systematically organize areas along a spectrum, from distinctly urban to distinctly rural. Current classifications often rely on predefined criteria, such as population size and density, which may not fully capture the nuanced and evolving nature of transitions driven by the complex interplay of socioeconomic processes, demographic shifts, environmental factors, and dynamic geographic forces.

This research addresses existing gaps by employing modern data-driven approaches, including machine learning and clustering techniques, to develop adaptive typologies that integrate diverse demographic, socioeconomic, and environmental variables. Using Switzerland as a case study, the proposed methodology offers a dynamic and scalable framework for territorial classification, supporting the effective management of territorial transitions and landscape conservation in Alpine regions. The analysis leverages a multidimensional dataset derived from the 2020 official census, incorporating 18 variables that encompass demographic profiles, socio-economic, and the physical space characteristics.

We used Self-Organizing Map (SOM) combined with hierarchical clustering. SOM, a type of competitive learning neural network, reduces the complexity of high-dimensional data by mapping it onto a two-dimensional grid of neurons. Visual outputs, such as heatmaps, enhance the interpretation of trends and patterns, providing a clearer understanding of variables distributions and interrelationships. Afterward, the SOM output grid of neurons was aggregated into six distinct clusters, which were mapped onto the geographical space. This produced a visual representation of the spatial organization of territorial typologies along the rural-urban continuum in Switzerland at a detailed municipal level.

The data-driven clustering approach developed in this study proved effective in capturing the complex and diverse nature of Swiss territorial typologies. The key findings reveal a landscape marked by a complex rural-urban interface, extensive intermediate zones, and significant spatial fragmentation. These final six territorial typologies could be characterized as follows: urban centres, representing the main hubs at the highest level of the Swiss urban hierarchy; suburban areas, located near and well-connected to urban centres; two peri-urban areas, distinguished into aging-rural areas and rural-urban edge; rural-forest areas, situated at medium to high elevations, featuring a forested landscape and rural settings; unproductive areas, encompassing high-altitude regions and including critical Alpine glaciers.

How to cite: Tonini, M., Yu, J., and Hagen-Zanker, A.: Exploring Switzerland's Rural-Urban Continuum Through Unsupervised Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6870, https://doi.org/10.5194/egusphere-egu25-6870, 2025.

Adequate sunlight exposure is crucial for human wellbeing, yet its accessibility in cities is significantly compromised by both could cover and complex three-dimensional (3D) urban structure. Here we adopted an analytical framework that integrated natural day length variations, cloud cover effects, and 3D urban structure to quantify actual sunlight duration in urban areas. By using high-resolution satellite products, fine-scale canopy height data, and detailed 3D building footprints, we mapped the spatiotemporal patterns of sunlight availability and quantified the relative contributions of cloud cover and urban structures on the loss of sunlight for Chinese cities. Our analysis reveals pronounced spatial disparities and trends in urban sunlight resources in China, underscoring the urgent need for evidence-based urban planning strategies that optimize natural light accessibility for sustainable urban development.

How to cite: Wu, X. and Chen, B.: Quantifying urban sunlight accessibility across Chinese cities: Impacts from cloud cover and three-dimensional (3D) urban structure, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7638, https://doi.org/10.5194/egusphere-egu25-7638, 2025.

Over the past few decades, the rapid growth of cities has evolved into a significant social, demographic, and architectural phenomenon, highlighting the vital importance of urban planning in fostering sustainable development. In this context, machine learning has emerged as a game-changing discipline, utilizing advanced algorithms to reshape traditional approaches to urban data management and analysis.This study combines Geographic Information Systems (GIS), Deep Learning techniques, and verified data from the General Directorate of the Spanish Cadastre to perform a comprehensive analysis of the urban environment through façade images in Murcia, one of Spain’s most dynamic metropolitan areas.Leveraging the clustering analysis of the studied variables, an automated binary classification model for façade images was developed using the pretrained EfficientNetB0 architecture in Python. To enhance interpretability, heat maps were generated to visualize the regions the model focuses on during classification. These heat maps reveal the critical features of the facades that guide the model’s decisions, providing valuable insights into the key factor influencing the classification process.The results were integrated into ArcGIS PRO, using the cadastral reference of the properties as a key attribute for a detailed spatial analysis. This approach revealed two significant areas linked to the metropolitan growth of Murcia, laying a strong foundation for future urban studies in the region.

Funding: Twin-ER: Earthquake Risk Pilot Digital Twin. Grant PID2023-149468NB-I00, funded by MCIU/AEI/10.13039/501100011033 and FEDER/EU

How to cite: De La Cruz Luis, M. I., Martinez Cuevas, S., Garcia Aranda, C., Morillo Balsera, M. C., and Poveda Lorente, E.-M.: Advancing Urban Environment Studies in Murcia, Spain through an Automated Façade Image Classification Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7834, https://doi.org/10.5194/egusphere-egu25-7834, 2025.

Equitable access to daily necessities and services is crucial for enhancing human quality of life and is integral to achieving the United Nations’ Sustainable Development Goals. However, knowledge about global access to these essential resources remains limited and fragmented, primarily due to the absence of a comprehensive infrastructure inventory and scalable measures of accessibility. Here we compiled the most extensive global database of points of interest (POI) to represent six essential infrastructure categories—living, healthcare, education, entertainment, public transit, and work. We used refined 30-meter resolution friction surface data to map travel times to these critical infrastructures as a proxy for accessibility across the urban-rural continuum and assessed disparities across geographic, urbanization, and socio-economic contexts. Our results reveal that access to daily necessities and services is unevenly distributed in terms of total infrastructure, per capita availability, and travel time. Globally, only 38.7% (2.6 billion people) and 50.7 % (3.4 billion people) of the population resides within a 15-minute and 30-minute walking distance of essential daily necessities and services, respectively. These results highlight the urgent need to optimize strategies for planning, allocation, and management of critical infrastructure to promote inclusive and sustainable development.

How to cite: Chen, B., Wu, S., Nelson, A., and Gong, P.: Measuring global human access to essential daily necessities and services, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8416, https://doi.org/10.5194/egusphere-egu25-8416, 2025.

Evaluating human interaction with environmental health determinants in space and time is fundamental to estimate personal environmental exposures. The increasing demand in an exposure assessment of entire populations requires to combine environmental variables at high resolution on large spatial extent, e.g. at nationwide or continental scale, with the space-time activity pattern of each individual in a study population.

Modelling population health and citizens' exposures is a complex process involving multiple procedural steps. One major step is to generate spatio-temporal information on environmental factors, either considered as beneficial for human wellbeing, for example, accessibility to green space or blue space, or considered as having negative health impacts such as the existence of air pollution, noise or heat. To capture the spatial variability these datasets need to be generated at high resolution. To allow for studies comparing cities, regions or countries, a geographical extent of subnational or larger size is required. In addition, data can be temporal to cover diurnal or seasonal variation of an environmental variable. Another major step is to use the environmental factors to as input to models calculating exposures for entire study populations, ranging from a few hundred participants up to millions of citizen. Here, socio-economic variables, mobility, different travel modes, and other daily activities with accompanying location changes need to be considered to mimic the space-time paths of each participant of a study population. These tasks require sufficient flexibility in both constructing environmental models as well as executing those eventually on HPC systems to break computational barriers of common workstations.

We present a computational framework for implementing both procedural steps and show the development of two European scale raster maps on a 25m grid and their subsequent usage to estimate human exposures to greenness visibility and noise. The maps were created with LUE (https://lue.computationalgeography.org/), an open-source modelling framework providing a Python package with currently 115 general-purpose operations for the construction of spatio-temporal simulation models. We implemented two custom focal operations that make use of the LUE framework. The first focal operation calculates for each raster cell the visible green area within a particular buffer size (c.f. Labib 2021, https://doi.org/10.1016/j.scitotenv.2020.143050). The second focal operation aggregates traffic-related noise within a particular buffer size, considering attenuation due to geometric divergence, atmospheric absorption, ground effects and diffraction.

We calculated visible green within a radius of 800m and noise within 1500m radius using 768 CPUs on eight HPC cluster nodes, and then used Campo (https://campo.computationalgeography.org/) for activity-based exposure assessment. The obtained exposure estimates can show considerable differences for different typical human activity patterns, such as homemaker or commuter, as well as a high spatial variability.

How to cite: Schmitz, O., de Jong, K., Shen, Y., and Karssenberg, D.: Assessing human exposures to environmental risk factors at continental-scale: accounting for short range variation in environmental factors and human activity patterns, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8801, https://doi.org/10.5194/egusphere-egu25-8801, 2025.

Anthropogenic heat generated by building energy use contributes to the urban island and climate change. Quantifying high spatiotemporal resolution city scale building energy use (BEU) and anthropogenic heat emission (AHE) is necessary for understanding urban microclimate and sustainable development. However, the current shortage of such data is insufficient to support urban energy management and climate decision-making. We estimated BEU and AHE from buildings in Hong Kong using a GIS-based city-scale building energy model (GIS-CBEM) and investigated their spatiotemporal variations. First, all buildings were categorized into 11 types, and a prototype was developed for each type. These prototypes were then calibrated using annual building energy consumption data from surveys. We studied the energy use profile for each building prototypes under the Typical Meteorological Year (TMY) weather data. Then, we estimated hourly BEU and AHE for all buildings in Hong Kong at the individual building level. The study results unveiled the spatiotemporal variation of buildings in Hong Kong at high resolution and detected divergent structure of building end-use and fuel use for different building prototypes. We found that the total BEU of all buildings in Hong Kong peaked at 5.1 × 10⁹ kWh in August, with 36.7% from HAVC system, while the lowest BEU was found in February at 3.5 ×10⁹kWh, with 14.1% from HAVC system. Total AHE from all buildings reached a maximum of 8.1 × 10⁹ kWh in July and minimum of 4.1 × 10⁹ kWh in February. Our findings have critical significance in enhancing energy efficiency, reducing environmental impact, and promoting sustainable development.

How to cite: Liu, Q. and Zhou, Y.: Unveiling Spatial and Temporal Variations of Building Energy Use and Anthropogenic Heat Emissions in Hong Kong, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9314, https://doi.org/10.5194/egusphere-egu25-9314, 2025.

Traffic congestion continues to challenge urban development, yet most research emphasizes large-scale factors such as road layouts and land use, overlooking localized visual aspects encountered by drivers. This study employs geographically weighted random forest, a non-linear and spatially explicit method, to explore how localized visual features—such as vehicle density, building structures, greenery, and road conditions—impact traffic congestion in Chicago. By integrating transport network dynamics with visual streetscape characteristics, the geographically weighted random forest approach captures spatial heterogeneity and complex interactions more effectively than traditional models. Results demonstrate that incorporating these multi-scale features improves model fit, revealing that greenery mitigates congestion, while dense urban structures and vehicle clusters exacerbate delays. These results highlight the potential of integrating visual characteristics of streetscapes into urban strategies to address congestion more effectively.

How to cite: Xu, M. and Weng, Q.: Modeling the Spatial Dynamics of Traffic Congestion Through Street-Level Visual Features: Evidence from Street View Images in Chicago, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9394, https://doi.org/10.5194/egusphere-egu25-9394, 2025.

Exploring the effect of building morphology on Land Surface Temperature (LST) has received surging attention. In this process, a fundamental precondition is selecting an appropriate spatial statistical unit to calculate building morphological indices and corresponding LST. However, different units lead to divergent results, indicating they inevitably suffer from the Modifiable Areal Unit Problem (MAUP), which brings large uncertainties. This study places special emphasis on proposing a new spatial unit, the Homogenous Unit of Building Morphology (HUBM), to re-describe building morphology and re-analyze its effect on LST with less uncertainty. Results show: (1) building morphology portrayed by HUBM maintains more spatial characteristics and remains relatively stable across scales, which is more consistent with the realistic building environment. (2) The relationship identified by HUBM shows building morphology is not strongly correlated with LST in essence and is regarded as more authentic due to the more objective portrayal of building morphology, while this relationship may be overestimated by previous common units. (3) The effect of building morphology on LST explored by HUBM also remains relatively stable across different scales (R² fluctuation amplitude of 0.08, 0.12, and 0.08 in the spring, summer, and winter, respectively) compared to regular grids (R² fluctuation amplitude of 0.18, 0.2, and 0.2), effectively alleviating the uncertainty associated with the MAUP. These findings provide new insights into re-examining the authentic effect of building morphology on LST, assisting in addressing urban heat island effects and promoting sustainable urban development. Moreover, HUBM can be applicable to other urban issues for mitigating MAUP.

How to cite: Yang, L., Yang, X., and Li, S.:  Is 3D building morphology really related to land surface temperature? Insights from a new homogeneous unit, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10149, https://doi.org/10.5194/egusphere-egu25-10149, 2025.

With rapid urbanization, cities face critical sustainability challenges, including poverty, resource shortages, pollution, and climate impacts. The EU Cities Mission supports 112 cities in developing Climate City Contracts to achieve climate neutrality by 2030 [1], emphasizing strategic, cross-sectoral approaches and stakeholder collaboration. This study introduces a systematic and indicator-based assessment of urban resilience, utilizing EU-sourced environmental, OpenStreetMap, and a few nationally sourced data. The methodology incorporates 12 key indicators, mapped at high resolution for 83 Italian cities using open-source GIS software [3], ensuring full reproducibility and applicability to other European cities. The indicators are categorized into five classes:

(i) nature and biodiversity, including forest canopy coverage, native habitat areas, biodiversity, geodiversity [4], ecological corridors, and heat island effects [5];

(ii) natural hazards, including susceptibility to flooding, earthquakes, wildfires, and landslides [6];

(iii) air pollution, including concentration of PM_2.5 and NO₂;

(iv) transport, including availability of sustainable and affordable transport systems;

(v) social indicators, including population living in close proximity to green spaces or water sources, and public services.

This study evaluates the current state of Italian cities [7], identifies regional differences, and highlights the strengths and weaknesses of each city individually, based on results provided by the urban indicators.

The software developed for this study is flexible, as the input data exists for the whole of Europe and it is easily extensible with modular scripts, to include additional indicators. The scripts processes data to produce spatially distributed results (raster maps) for each indicator in each class listed above and then summarize each indicator with a numerical figure.

Preliminary findings suggest significant regional variation in factors contributing to climate resilience and citizen well-being [8]. Cities in Northern Italy exhibit larger green space coverage but also higher air pollution levels. In contrast, Central Italy stands out for its high species biodiversity and geodiversity. Moreover, results uncover regional spatial patterns, offering actionable insights for policymakers to design locally informed and effective strategies. The findings contribute to advancing sustainability goals, supporting urban transformations toward enhanced resilience and reduced environmental impact. A comprehensive set of urban indicators, including those derived in this study and summarized into a single numerical output for each category, allows ranking of cities and promoting the adoption of data-driven strategies for sustainable development.

References

[1] United Nations (2023) https://sdgs.un.org/goals/goal11

[2] Sarretta et al., Int. Arch. Ph. Rem. Sens. Spat. Inf. Sci. (2021) https://doi.org/10.5194/isprs-archives-XLVI-4-W2-2021-159-2021.

[3] Neteler et al., Env. Mod. Softw. 31 (2012) https://doi.org/10.1016/j.envsoft.2011.11.014

[4] Burnelli et al., Geomorphology 471 (2024) https://doi.org/10.1016/j.geomorph.2024.109532

[5] Morabito et al., Sci. Tot. Env. (2021) https://doi.org/10.1016/j.scitotenv.2020.142334

[6] Loche et al., Earth-Science Reviews 232 (2022) https://doi.org/10.1016/j.earscirev.2022.104125

[7] Alvioli, Land. Urb. Plan. 204 (2020) https://doi.org/10.1016/j.landurbplan.2020.103906

[8] Boeing et al., Lancet Global Health 10 (2022) https://doi.org/10.1016/S2214-109X(22)00072-9

How to cite: Tarasova, E. and Alvioli, M.: Measuring resilience of urban areas using public data and a reproducible approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11618, https://doi.org/10.5194/egusphere-egu25-11618, 2025.

Due to climate change, new challenges arise in drinking water infrastructure planning and in the re-assessment of well-established urban drinking water utilities. We observed temperatures exceeding 25 °C in drinking water supply pipes, which pose a health threat and water quality problem, as these temperatures are favorable for microbial growth.

We set out to predict temperatures in drinking water supply networks. The key step to achieve this goal is to monitor and model soil temperatures and soil moisture derived from meteorological forcing functions. With meteorological observations and soil material properties, we describe the heat transport and water flow from the ground surface into the subsurface and from there into the pipes and with the water in the pipes.

In order to achieve this goal, we solved the heat and water balances jointly at the atmosphere-subsurface interface, using the open-source numerical simulation framework DuMuX. We were able to do this because of the available meteorological observations (e.g., radiation balance, precipitation intensity) next to the newly installed pipes. These balances provide a novel interface condition for heat transport and water flow modelling. We coupled the heat transport through the drinking water pipe walls to the drinking water in the pipes and to the subsurface transport processes.

At a pilot site, we installed typical drinking water pipes (PE and cast iron), backfilled with known material (typical gravelly conditions below roads and naturally existing sandy clay), and applied land-cover (asphalt and natural vegetation). We were able to reproduce the joint measurements of temperatures and soil moisture under various conditions (well-draining gravel vs. less-draining clayey material; vegetation vs. asphalt).

In this presentation, we demonstrate results of the multi-year measurement campaign, the results of 1D and 2D subsurface heat transport models coupled to dynamic hydraulic conditions in the drinking water pipes, and an innovative surrogate-based Bayesian active learning-assisted model calibration methodology.

This work presents an important first step towards predicting temperatures in drinking water supply pipes and will be directly relevant for chemical and biological processes that occur in non-isothermal conditions (e.g., due to climate change), for example, in relation to contaminant remediation. Our results are of relevance for drinking water supply companies, shallow geothermal design, and urban planning.

How to cite: Haslauer, C., Kroeker, I., Nißler, E., Oladyshkin, S., Nowak, W., Class, H., and Osmancevic, E.:   Large Temperatures in Water Distribution Pipes as a Water Quality Threat: Measurements and Modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13131, https://doi.org/10.5194/egusphere-egu25-13131, 2025.

Urban climate modelling tools like the Surface Urban Energy and Water balance Scheme (SUEWS) are indispensable for investigating complex surface–atmosphere interactions and guiding urban adaptation strategies. However, these models often present substantial barriers to use: they require extensive technical know-how, involve intricate input datasets, and can be time-consuming to set up and interpret. Recent advancements in Large Language Models (LLMs) hold promise for bridging this gap by transforming complex domain-specific tasks—such as data validation, simulation setup, and error diagnosis—into user-friendly interactive experiences.

In this study, we propose a novel workflow that leverages LLM capabilities—such as generative text, code suggestion, and context-driven troubleshooting—to streamline SUEWS usage and improve accessibility for researchers and practitioners:

• Automated Model Configuration
  We explore the use of LLM-guided prompts to generate properly formatted SUEWS input files, such as specifying hourly meteorological forcing data (e.g., temperature, wind speed, and humidity) or land cover fractions required for accurate simulations. By conversing with the model about location, time range, and data availability, users can rapidly produce consistent and error-checked setup files, reducing manual edits that often lead to inconsistencies.

• Interactive Error Diagnosis
  LLMs can parse error logs and suggest potential solutions in real time. For example, if SUEWS outputs an error related to missing albedo values for a specific land cover type, the LLM can pinpoint the source of the issue and suggest default values or a method for calculation based on site-specific conditions. For example, if a runtime error indicates a mismatch in the date format of meteorological input data, the LLM can identify the exact line causing the error, recommend the correct format, and provide a command or script snippet to rectify the issue. Through iterative dialogue, the model clarifies the root causes of typical setup or runtime issues, explaining how to fix them without requiring the user to trawl through detailed documentation.

• Model Output Interpretation
  Interpreting large volumes of SUEWS output, such as energy balance components (net radiation, latent heat flux, and sensible heat flux) or water budget terms (runoff and evapotranspiration), can be daunting, especially for newcomers. LLMs can summarise key metrics—like energy flux partitioning and surface runoff patterns—and highlight discrepancies in data, thereby assisting in rapid analysis and scenario comparison.

Our findings indicate that an LLM-enabled approach substantially lowers the learning curve and operational overhead associated with SUEWS, while still maintaining scientific rigour. We piloted trial deployments in teaching and professional contexts, reporting improvements in both setup speed and user confidence. Future work includes refining the LLM’s domain-specific training to ensure physically consistent responses—such as maintaining energy balance across flux computations or ensuring water budget closure—and incorporating advanced visualisation plugins for immediate data interpretation.

By harnessing the dialogic strengths of LLMs, we aim to remove barriers to the complexity of urban climate modelling, ultimately broadening participation and fostering more informed decision-making in cities worldwide.

How to cite: Sun, T.: Remove Barriers to Accessible Urban Climate Modelling with Large Language Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13462, https://doi.org/10.5194/egusphere-egu25-13462, 2025.

The frequent occurrences of heat wave events and air pollution episodes have become pressing global concerns. Concurrent heat and ozone pollution events, in particular, have been widely documented across various regions and often result in more severe impacts compared to isolated stressors, leading to increased mortality and morbidity rates. However, our understanding of these compound events in urban environments, particularly their dynamics under different background climates and urban settings, remains limited. In this study, we systematically characterized the frequency, intensity, and duration of compound heat and ozone pollution events during warm seasons across all urban areas in the continental U.S. using long-term, high-resolution daily air pollution and air temperature datasets. Results suggest that urban heat waves, defined by daily maximum temperature, were more frequent, more intense, and longer lasting than their rural counterparts, primarily due to the urban heat island effect. In contrast, over half of the U.S. cities experienced fewer, less intense, and shorter ozone pollution episodes than surrounding rural environments. The spatially heterogeneous disparities in ozone pollution episodes among cities are mainly attributed to whether ozone production is limited by VOC or NOx, as revealed by time series analyses. Despite the overall decreasing trend of surface ozone concentrations during the last two decades, 89% of U.S. cities experienced more frequent compound heat and ozone pollution episodes than rural areas. Additionally, the cumulative heat and ozone intensities were higher in 91% and 88% of U.S. cities, respectively, than in their rural backgrounds. The duration of compound events tends to be shorter in urban areas. These findings highlight the dependence of such compound events on local and background conditions, emphasizing the need for locally tailored mitigation plans to reduce their impacts. This study also calls for detailed regional numerical simulations to elucidate the mechanisms driving these events.

How to cite: Wang, C., Hu, X.-M., and Leffel, J.: Compound heat and ozone pollution in urban areas, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13711, https://doi.org/10.5194/egusphere-egu25-13711, 2025.

Cities are highly interconnected networks of networks (referred to as the Urban Multiplex) that include the power grid and transportation networks, surface water and groundwater, sewerage and drinking water systems, inland navigation and dams – all intertwined with the natural environment and socioeconomic and public health sectors. While the Urban Multiplex is physically and functionally connected, the data produced within its individual sectors are not. This prevents us from fully understanding how the Urban Multiplex is connected, and how failures triggered by external stressors like floods cascade.  

Knowledge Networks are an AI technology that (i) integrates Urban Multiplex data, (ii) produces real-time flood forecasts across the continental U.S., (iii) serves as the foundation to evaluate the total impact of floods on cities, and (iv) supports queries at the nexus of water and energy. This talk will describe the development of the Urban Flooding and Water-Energy Nexus Open Knowledge Networks that aim to provide actionable answers to questions such as:

• Real-time flood mitigation and response: Will my neighborhood flood? Will I have access to water and power? Will this storm disrupt the power grid, drinking water treatment plant, or a bridge?

• Long-term design, planning and research: What is the total socioeconomic impact of this flood? Which critical urban infrastructure will likely fail in a future flood? Which failures will affect the most people or the most vulnerable people? Are there vulnerable communities downstream of this coal mine?

The interdisciplinary team behind this project has brought together academic researchers, industry, federal government, U.S. National labs and local stakeholders. It is funded by the U.S. National Science Foundation’s Convergence Accelerator Program that is structured to enable rapid advancement in highly complex problems of critical societal importance.

How to cite: Yeghiazarian, L. and the Knowledge Networks Team: Knowledge Networks help address urban flooding and water-energy challenges , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13868, https://doi.org/10.5194/egusphere-egu25-13868, 2025.

In recent decades, inequalities in economic, health, and education sectors have intensified spatial clustering of populations and resources, further reinforcing disparities within urban environment. Identifying these geographic boundaries is crucial for developing targeted policies to address inequality effectively. While traditional approaches to studying urban segregation rely primarily on socioeconomic indicators, this research introduces a novel methodology that combines subjective perceptions of the urban environment and objective characteristics of urban areas—such as land use and infrastructure—to identify distinct spatial clusters within Glasgow, a city with a varied socioeconomic landscape. Using MIT Place Pulse dataset of crowd-sourced streetscape perceptions, we developed a deep learning model to predict perception scores for new areas. These perception scores, along with image embeddings and land use information, enabled the geographic clustering of areas based on perceived and functional similarities. Our analysis reveals that perception-based boundaries often diverge from traditional census dissemination areas, suggesting that administrative boundaries may not fully capture the lived experiences of urban space. This research advances our understanding of urban inequality by demonstrating how perceived environmental qualities interact with physical infrastructure to shape distinct urban zones, providing policymakers with new tools for targeted intervention strategies.

How to cite: Pawar, S., Robertson, T., Marino, A., and Anderson, C.: Redefining Urban Clusters: Combining Subjective Perceptions and Objective Data to Map Inequality, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16365, https://doi.org/10.5194/egusphere-egu25-16365, 2025.

The rapid expansion of urban areas has led to significant environmental changes, most notably the Urban Heat Island (UHI) phenomenon, characterized by higher temperatures in urban areas compared to their rural counterparts. Addressing and mitigating UHIs is vital for public health, energy demand management, and enhancing urban livability, especially amidst global climate change. This study focuses on classifying Local Climate Zones (LCZs) in Taipei, Taiwan, using digital building data, satellite imagery, and urban morphological indices. LCZs offer a standardized framework to analyze urban morphology and its influence on local climates. By applying unsupervised clustering methods, we achieved a detailed classification of urban areas, enabling a data-driven exploration of their climatic and morphological characteristics.

To downscale and refine the analysis at the community level, Principal Component Analysis (PCA) was employed to reduce data dimensionality and extract key features such as building coverage, vegetation index, and sky view factor. K-means clustering was then used to categorize urban morphological types, resulting in distinct LCZs across Taipei. Our findings reveal significant differences in environmental variables among clusters. These results highlight how urban morphology, including building density and vegetation cover, impacts local climate conditions. The study also emphasizes the role of thermal comfort, underscoring the complex interplay between urban form and environmental factors.

This research demonstrates the effectiveness of unsupervised classification methods in identifying urban climate zones and provides a practical framework for urban planning and climate adaptation. By enabling targeted interventions, such as greening strategies or ventilation optimization, the study contributes to enhancing urban sustainability and resilience. The findings underscore the importance of interdisciplinary approaches to address the multifaceted challenges of urbanization and climate change.

How to cite: Chen, W.-J., Juang, J.-Y., and Chien, S.-S.: Self-Organizing Local Climate Zones by Using Integrated information in Urban Community – a case study in DaXue Village, Taipei, Taiwan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17835, https://doi.org/10.5194/egusphere-egu25-17835, 2025.

Urban thermal environment is known to be strongly affected by the composition of urban land cover, with densely built-up areas characterised with distinctly higher temperatures than densely vegetated ones. These observations come from the analysis of relatively coarse land surface temperature (LST) satellite data and conclusions are typically derived for city districts or other variedly defined mapping units. Whilst these analyses provide useful insights to excess heat mitigation at city scales, these do not describe the nuance of the thermal response of the heterogenous urban form at local levels. This study investigated the relationship between LST of variedly configured immediate neighbourhoods of single patches of different land cover types (buildings, paved, grass, trees) extending from 0 to 100m away to determine the shape and type of urban features including water that contribute to the formation of cold and hot urban spaces. The study area comprised three English towns: Milton Keynes, Bedford, and Luton, collectively comprising a wide range of urban forms that are representative for England and other European towns located in the temperate climate zone. The analysis was carried out for two summer days a month apart, capturing the different thermal responses as temperatures rise over summer.  The microscale of the analysis was enabled by downscaled LST obtained from Landsat 8 thermal bands acquired at 100m resolution down to 2m, supported by high resolution spectral indices derived from very high resolution hyperspectral aerial imagery. Patch-level landscape metrics were used to describe the shape of the different patches of urban land cover derived from land cover map at 2m resolution. K-means analysis was used to determine groups of land cover patches of a given type with common thermal and spatial properties. Random forest regression algorithm was used to identify the important descriptors of LST for these groups and ANOVA analysis to determine statistically significant effects for various spatial configuration metrics. The findings suggested that the coldest patches of buildings, grass and paved were associated with highly aggregated patches of trees in the immediate neighbourhood, with PLADJ greater than 73 to 85% and COHESION greater than 93 to 97%, and buildings requiring somewhat lower aggregation levels than grass or paved. Hottest patches of these land cover classes were associated with PLADJ smaller than 63–69% and COHESION smaller than 83–87%, with elevation and distance to water being the most important factors, whose importance increased as the summer progressed. Overall, this study provided further insights into the spatial characteristics of patches of common land cover types in urban areas that contribute to the formation of particularly hot or cold urban spaces, which can facilitate the design of climate resilient cities.

How to cite: Zawadzka, J., Harris, J., and Corstanje, R.: The importance of spatial configuration of urban form in local temperature regulation investigated from very high resolution LST and land cover data and landscape metrics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19759, https://doi.org/10.5194/egusphere-egu25-19759, 2025.

The building stock accounts for 34% of global energy demand and 37% of CO₂ emissions related to energy and industrial processes. Additionally, the current increase in urbanization rates poses significant environmental challenges. Policy makers are becoming increasingly aware of these impacts, developing strategies aimed at improving energy efficiency and obtaining decarbonization of the built environment. Achieving these goals requires modeling actual building stock energy consumption patterns, future energy developing trends as well as the impact of energy retrofitting measures on CO₂ emissions. Urban Building Energy Models (UBEMs) and bottom-up engineering models have proven to be valuable tools. However, these models  require detailed and accurate building attributes related to physical properties (building geometry, height, building type, thermal transmittances, etc.), local climate (air temperature, humidity, solar radiation, etc.) and data related to occupants' energy behavior (occupants’ schedule, heating and cooling energy demand, efficiency of the system etc.). Among others, building construction year is one of the most relevant parameters since it is a key proxy for essential characteristics such as morphology, facade design, building materials, and energy efficiency. However, obtaining building construction year is particularly challenging as it is rarely available in public databases and, when available, the data are often incomplete or inconsistent. In this regard, remote sensing techniques can play a crucial role in the study and monitoring of the building stock. In particular, satellite images represent an excellent tool for the estimation of building age at local or regional scale given their extensive temporal and spatial coverage, as well as and the continuous updates of collections. The study focuses on the city of Parma, for which seven images covering the year range between 1985 and 2011 were selected. After a literature review, five built-up area extraction indices suitable for TM sensor were selected: Normalized Difference Built-up Area Index (NDBI), New Built-up Index (NBI), Band Ratio for Built-up Area (BRBA), Normalized Built-up Area Index (NBAI), and Vegetation Index Built-up Index (VIBI). In addition, Normalized Difference Vegetation Index (NDVI) was also considered, leading to a total of six indices. To improve the ability of these indexes to discriminate urban surfaces from areas with similar spectral signature (bare soil, sand, rock, etc.) annual greenest pixel composite images were generated using Google Earth Engine. Indexes performance was then compared on each image evaluating Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves, as well as performance metrics such as F1-score and Area Under the Curve (AUC). The results indicate that the NDVI is the best- Finally, temporal series were derived from the classification of images from different years, enabling the assessment of urbanization growth over time and, consequently, the estimation of building ages.

This study was carried out within the RETURN Extended Partnership and received funding from the European Union Next-GenerationEU (National Recovery and Resilience Plan – NRRP, Mission 4, Component 2, Investment 1.3 – D.D. 1243 2/8/2022, PE0000005) – SPOKE TS 1

How to cite: Vigato, T., Dalle Vedove, L., Dalla Vecchia, C., Zandonella Callegher, C., and Zilio, S.: Comparing the effectiveness of Landsat-derived spectral indices for building age prediction in urban energy modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20244, https://doi.org/10.5194/egusphere-egu25-20244, 2025.

Seismic data forms the backbone of what we understand in the subsurface, and seismic data interpretation is still usually done by hand. Automatic seismic interpretation with deep learning is very promising, but there the problem is a lack of labelled training data. In this study, we use forward stratigraphic modelling and show how forward modelling can be advantageously used in deep learning.

Specifically, we focus on shelf-edge trajectories as the geological representations of lateral and vertical shifts in sediments’ position through time. They provide continuous tracks of changes in relative sea-level as well as sediment stacking patterns and depositional geometries. Mapping these trajectories and measuring their changing angles help in quantifying the sequence stratigraphic analysis and predicting ancient depositional environments.

Here, we evaluate the ability of deep learning models, trained on synthetic seismic data, to identify clinoforms and their rollover points for shelf-edge trajectories mapping. The synthetic training dataset generated using geological processed-based forward modelling represents different depositional slope scenarios. Controlling the different parameters that govern shelf-edges and shelf-edge trajectories (such as bathymetry, sediment supply, eustatic sea-level changes and subsidence) gave us a better chance to mimic realistic and diverse depositional setting, which helps in generalizing the deep learning model. In addition, the ground truth (labels) for the created synthetic seismic data is automatically generated by the forward model, without the need of manual labelling seismic data.

Higher accuracy score on both validation and testing datasets demonstrates the power and effectiveness of using synthetic as training dataset. This study shows that synthetic data can play a major role in bridging the gap between traditional seismic interpretation and automating the process using machine learning. It also shows that forward modelling is a powerful technique to combine with data modelling, such as machine learning.

How to cite: AlGharbi, W., Bell, R., and John, C.: Forward Stratigraphic Modelling to Generate Synthetic Seismic Training Dataset for Deep Learning: A Case Study to Predict Shelf-Edge Trajectories, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-288, https://doi.org/10.5194/egusphere-egu25-288, 2025.

This study aims to develop a CNN-LSTM hybrid network model integrated with a coupled self-attention mechanism, based on deep learning techniques, to simulate flood processes in the Inner Harbor area of Macau. With global climate change and accelerated urbanization, Macau, a low-lying coastal city, frequently experiences urban flooding due to typhoons and heavy rainfall. While traditional hydrological and hydrodynamic models can accurately predict flooding processes, their computational intensity and lack of real-time responsiveness make them unsuitable for emergency disaster warnings. To address these limitations, this paper proposes a convolutional long short-term memory (ConvLSTM) model enhanced with a coupled self-attention mechanism. The model leverages an encoder-decoder structure to predict the evolution of flood processes under 4–10 hours of heavy rainfall in the Inner Harbor area of Macau.

The model integrates CNN components for extracting spatial features, LSTM components for capturing temporal features, and a coupled self-attention mechanism to dynamically reweight spatial-temporal representations, improving the model's sensitivity to key flood patterns. The encoder encodes input sequences into fixed-length vectors, while the decoder translates these vectors into target sequences. The self-attention mechanism ensures the model focuses on critical spatial and temporal regions, further enhancing prediction accuracy and robustness.

The training and testing datasets were constructed from simulation data generated by hydrological-hydrodynamic models and static geographical information data, following preprocessing and normalization. Evaluation metrics, including mean squared error (MSE), Nash-Sutcliffe efficiency coefficient (NSE), and relative error, were used to assess model performance. Results demonstrate that the proposed hybrid model, augmented by the coupled self-attention mechanism, effectively simulates maximum water depth distribution and flood evolution processes, achieving high consistency with hydrodynamic simulation data while providing improved predictive performance.

How to cite: Wangqi, L.: Flood Process Simulation in Macau's Inner Harbor Area Based on CNN-LSTM, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2197, https://doi.org/10.5194/egusphere-egu25-2197, 2025.

AI enhanced environmental modelling workflows: Towards Automated Scientific Exploration in Hydrology

Authors: Darri Eythorsson, Kasra Keshavarz, Cyril Thébault, Mohamed Ismaiel Ahmed, Raymond Spiteri, Alain Pietroniro and Martyn Clark

Modern hydrological modeling has evolved into a complex scientific endeavour requiring sophisticated workflows that span multiple scales, processes, and computational paradigms. While existing workflow solutions address specific technical challenges, the field lacks comprehensive frameworks that can support end-to-end modeling while maintaining reproducibility and scalability. This works introduces two complementary frameworks that aim to address these fundamental challenges: CONFLUENCE (Community Optimization Nexus For Large-domain Understanding of Environmental Networks and Computational Exploration) and INDRA (the Intelligent Network for Dynamic River Analysis).

CONFLUENCE implements a modular architecture that enforces workflow reproducibility through a unified configuration system while maintaining the flexibility needed to support diverse modeling applications. The framework provides comprehensive solutions for four critical workflow components: (1) flexible geospatial domain definition and discretization, (2) model-agnostic data acquisition and preprocessing, (3) extensible model setup and parameterization capabilities, and (4) comprehensive evaluation and optimization tools. This systematic approach enables efficient, reproducible, and transparent hydrological modeling across scales.

INDRA augments this foundation by implementing a network of specialized AI expert agents that support various components of the hydrological modeling workflow. Through structured dialogue between domain experts (including AI specialists in hydrology, hydrogeology, meteorology, data science, and geospatial analysis), INDRA provides context-aware guidance while maintaining complete provenance of modeling decisions and their justification. This AI-assisted approach helps address three critical challenges: (1) the growing complexity of modelling decisions, (2) the need for reproducible workflows and detailed documentation, and (3) the technical barriers limiting broader adoption of advanced modeling practices.

The integration of these frameworks aims to explore how automation and AI assistance can enhance rather than disrupt traditional modeling practices. By maintaining clear documentation of decisions and their justifications, these systems help build trust in model results while creating opportunities for recursive learning from previous modeling experiments. Our case studies, spanning scales from individual catchments to continental domains, showcase the frameworks' capabilities while highlighting their potential to transform how researchers’ interface with complex environmental modeling workflows.

This work aims to advance both operational and research oriented hydrological modeling practices, offering a foundation for reproducible, scalable, and interoperable modeling while maintaining scientific rigor and flexibility. The framework’s open-source nature and modular design create opportunities for community-driven development and extension, potentially accelerating scientific discovery in hydrological sciences.

How to cite: Eythorsson, D., Keshavarz, K., Thébault, C., Ahmed, M., Spiteri, R., Pietroniro, A., and Clark, M.: AI enhanced environmental modelling workflows: Towards Automated Scientific Exploration in Hydrology, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4565, https://doi.org/10.5194/egusphere-egu25-4565, 2025.

This study explores integrating reinforcement learning (RL) with idealised climate models to address key parameterisation challenges in climate science. Current climate models rely on complex mathematical parameterisations to represent sub-grid scale processes, which can introduce substantial uncertainties. RL offers capabilities to enhance these parameterisation schemes, including direct interaction, handling sparse or delayed feedback, continuous online learning, and long-term optimisation. We evaluate the performance of eight RL algorithms on two idealised environments: one for temperature bias correction, another for radiative-convective equilibrium (RCE) imitating real-world computational constraints. Results show different RL approaches excel in different climate scenarios with exploration algorithms performing better in bias correction, while exploitation algorithms proving more effective for RCE. These findings support the potential of RL-based parameterisation schemes to be integrated into global climate models, improving accuracy and efficiency in capturing complex climate dynamics. Overall, this work represents an important first step towards leveraging RL to enhance climate model accuracy, critical for improving climate understanding and predictions. Code accessible at https://github.com/p3jitnath/climate-rl.

How to cite: Nath, P., Moss, H., Shuckburgh, E., and Webb, M.: RAIN: Reinforcement Algorithms for Improving Numerical Weather and Climate Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5159, https://doi.org/10.5194/egusphere-egu25-5159, 2025.

Machine Learning is gaining increasing attention from the scientific community in hydrological and hydraulic research. However, this field faces a consistent challenge in applying data-driven approaches due to the evident poor generalization capabilities, which are partly a result of inherent data scarcity.

We propose incorporating expert knowledge into data-driven models for river hydraulics and flood mapping by integrating physically-based information without relying on the underlying mathematical formulation (e.g., the calculation of the residuals of differential equations). This approach appears to be particularly valuable for flood simulations, where hydraulically relevant distributed parameters such as roughness, lithology, topography etc. pose significant uncertainties. The method is versatile and applicable to physical systems and scenarios in which the underlying mathematical formulation is not fully known, but expert knowledge enables the introduction of meaningful, physically-inspired constraints.

Specifically, the physical information is integrated into the model by including an additional term, weighted by the hyperparameter in the guise of a regularization term in the loss function :

Here, represents the data-driven error metric, while the physical loss term is an error metric that depends not only on the true and predicted outputs ( ), but also potentially on the inputs . Indeed, this term employs physical principles, laws, and quantities, which are not explicitly formulated in the original dataset. In this sense, we can note its similarity to data augmentation, a widely used technique in machine learning that extracts additional insights by offering alternative interpretations of the same dataset.

We clarify that this approach does not aim to replace numerical solvers or serve as an alternative numerical model, as Physics-Informed Neural Networks do: indeed, their similarity is limited to the formulation of the modified loss function.

We assessed the methodology and empirically quantified the effectiveness of the method in a simplified, well-controlled problem, evaluating the gain in generalisation capability of Neural Networks (NNs) in the reconstruction of the steady state, one-dimensional, water surface profile in a rectangular channel. We found improved predictive capabilities, even when extrapolating beyond the boundaries of the training dataset and in data-scarce scenarios. This kind of assessment is of great relevance to the application of NNs to flood mapping, where cases featuring values of the observed quantities falling out of the range of the recorded series need to be predicted.

New experiments have been also conducted on two-dimensional domains. The data-driven model was trained on a single catchment and tested on its ability to determine flooded areas in unseen catchments. Preliminary results show that an encoder-decoder model with convolutional layers exhibits improved generalization when a physical training strategy is employed. Future applications could include flood mapping for ungauged basins, leveraging similarities with other basins.

How to cite: Guglielmo, G. and Prestininzi, P.: Physically-Enhanced Training of Neural Networks for Hydraulic Modelling of Rivers and Flood Events, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5991, https://doi.org/10.5194/egusphere-egu25-5991, 2025.

Background: Agriculture is increasingly leveraging machine learning (ML) to enhance yield predictions and optimize agronomic practices. Maize, a staple crop in Ghana, offers a valuable case study for evaluating the effectiveness of diverse ML models in yield prediction and resource management.

Objective: This study aims to evaluate the predictive performance of four ML models namely Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Extreme Gradient Boosting (XGBoost) for maize yield and agronomic efficiency prediction. It also compares variable importance across these models to determine key explanatory variables.

Methods: The study utilized 4,496 georeferenced maize trial datasets from various agroecological zones in Ghana. Thirty-five explanatory variables included soil properties, climate, topography, crop management practices, and fertilizer application datasets. Model performance was evaluated using leave-one-out, leave-site-out, and leave-agroecological-zone-out cross-validation techniques. Metrics including Mean Error (ME), Root Mean Squared Error (RMSE), and Model Efficiency Coefficient (MEC) were used to compare model accuracy, while a permutation-based approach was employed to assess variable importance.

Results: XGBoost emerged as the most accurate model, achieving the lowest RMSE for yield (639.5 kg ha⁻¹) and agronomic efficiency (11.6 kg kg⁻¹), particularly for nitrogen (AE-N). RF demonstrated competitive performance, while KNN and SVM yielded inconsistent results under rigorous cross-validation conditions. Key explanatory variables identified across models included nitrogen fertilizer, rainfall, and crop genotype, underscoring their critical role in yield and agronomic efficiency outcomes.

Conclusion: XGBoost was the most robust and accurate model for maize yield and agronomic efficiency predictions, offering a reliable tool for data-driven agricultural planning in diverse agroecological settings. The findings underscore the transformative role of advanced ML techniques in modern agriculture, particularly in optimizing staple crop production in sub-Saharan Africa.

How to cite: Asamoah, E., Heuvelink, G., Chairi, I., Bindraban, P., and Logah, V.: Modelling Maize Yield and Agronomic Efficiency Using Machine Learning Models: A Comparative Analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9987, https://doi.org/10.5194/egusphere-egu25-9987, 2025.

Expert interpretation of borehole data is a critical component of geological modelling, offering essential insights into the spatial distribution of geological units within the subsurface. For large-scale regional mapping efforts, expert interpretation of all available data is impractical due to the sheer volume of boreholes. Therefore, many 3D geological subsurface models rely only on a small portion of all available data. Machine learning (ML) models can be used to automate borehole data interpretation, increasing data density. However, these automated interpretations must adhere to strict spatial and stratigraphical relationships to be consistent with the established geological knowledge of the area. Using a dataset of 1,400 boreholes with expert interpretations from the Roer Valley Graben (Southeast Netherlands), we explore how ML models can be integrated into geological modelling workflows, highlighting the challenge of ensuring compatibility with geological principles and known spatial relationships. We evaluate the model performance using traditional metrics such as accuracy, Cohen's kappa and F1 Score and newly proposed geology-inspired metrics to quantify the ability of Random Forest and Neural Network models to interpret borehole data into lithostratigraphic units while preserving key geological relationships. Our results demonstrate that while many models achieve accuracy values of 75% to 80%, Neural Networks perform significantly better in capturing the expected sequential relationships between geological units, achieving up to 96% of geological transitions between geological units that are plausible, compared to 65% for the best-performing Random Forest model selected based on traditional metrics. This study underscores the need for domain-specific metrics in evaluating model performance and the potential for ML to increase the volume of data incorporated in subsurface models.

How to cite: Garzón, S., Dabekaussen, W., De Boever, E., Busschers, F., Mehrkanoon, S., and Karssenberg, D.: Assessing the Geological Plausibility of Machine Learning Borehole Interpretations: A Case Study in the Roer Valley Graben, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10915, https://doi.org/10.5194/egusphere-egu25-10915, 2025.

Abstract:
Alluvial terraces in watersheds are geomorphic features formed by river incision and sediment deposit, representing former floodplain levels. They serve as valuable records of fluvial dynamics, climatic changes, and tectonic activity. Mapping methods for terraces that rely on field-acquired data, often involving physical or chemical analyses, are not feasible for large-scale applications. When aiming to map at the national scale, the development of a methodology that eliminates the need for such detailed information enhances scalability and broadens applicability.

We proposed a semi-automatic predictive mapping method for watershed terraces using 25m Digital Earth Model (DEM) provided by the IGN (French geographical service) and derived variables such as curvature, slope, and the difference from a base level (Raingeard et al., 2019). This method demonstrated meaningful results in the Pyrenean Piedmont for the Baïse and Ousse rivers, with the predicted map showing strong alignment with the geological reality.

In this study, we propose an automated approach for identifying alluvial terraces using relative height. Relative height is defined as the difference between the elevation derived from a DEM and the base level. Our methodology is based on the hypothesis that terraces are represented as flat areas in the relative height, where pixels exhibit similar statistical distributions. To capture these patterns, we employ a Gaussian Mixture Model, a probabilistic framework that approximates data as a combination of multiple Gaussian distributions. In this context, each Gaussian distribution corresponds to a specific alluvial terrace.

We conducted experiments on the study areas used by Raingeard et al. (2019), and the results are consistent with both the semi-automatic method and the geological reality. These outcomes provide promising prospects for the predictive mapping of superficial deposits

Reference:

Raingeard A., Tourlière B., Lacquement. F, Baptiste. J, Tissoux. H. Semi-automatic quaternary alluvial deposits mapping - Methodology for the predictive mapping of flat terrains within a watershed, by semi-automatic analysis of the Digital Elevation Model. INQUA 2019, Jul 2019, Dublin, Ireland. 2019.

How to cite: Rakotonirina, H., Lohier, T., Raingeard, A., Lacquement, F., Baptiste, J., and Tissoux, H.: Mapping Alluvial Terraces in Watersheds Using Gaussian Mixture Model on Relative Height., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11031, https://doi.org/10.5194/egusphere-egu25-11031, 2025.

Stochastic parameterisations have seen widespread use in atmospheric models. These schemes represent uncertainty by adding terms that include a random noise component directly to the equations that describe the time evolution of the model. Stochastic parameterisation development thus involves the following questions: what are the sources of uncertainty, how do we represent them, and how precisely should we formulate stochastic terms to quantify them? Common methods to quantify the uncertainty inherent in parameterisation include applying multiplicative perturbations to physics tendencies (such that the larger the tendency due to subgrid processes, the more uncertainty we should have in the tendency) or applying perturbations to physical parameters (our physics schemes often rely on parameters whose values we do not know precisely, and have complicated nonlinear responses to perturbing this set of parameters, so perturbing each one within its own specified range during the model run allows this uncertainty to feed back into the model state).

In this work we present a different approach to stochastic parameterisation. We perturb the thermodynamic profiles that constitute the inputs to parameterisation schemes. The perturbations are scaled by the degree of subgrid inhomogeneity. A representation of the subgrid inhomogeneity is estimated by a machine learning model which has been trained on coarse-grained high resolution (dx=~1.5km) model output from the Met Office Unified Model. The scheme is implemented in LFRic, the UK Met Office’s next generation modelling system, and we present results of experiments ran in single column model mode.

How to cite: Reid, H. and Morcrette, C.: A new stochastic physics scheme incorporating machine-learnt subgrid variability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11747, https://doi.org/10.5194/egusphere-egu25-11747, 2025.

Accurately forecasting water table rise (WTR) is essential for effective water resource management, infrastructure development, flood risk mitigation, and environmental conservation. This research employed multiple machine learning (ML) models, namely Ridge Linear Regression (RLR), Radial Basis Function Support Vector Machine (RBF-SVM), Linear SVM (LSVM), Random Forest (RF), and a hybrid deep learning Transformer (TR) with Bi-Long Short-Term Memory (BiLSTM), to forecast WTR one and two weeks ahead in the Muscat Governorate, Oman. A total of 19,465 high-resolution datasets, measured at half-hour intervals between December 2017 and January 2019, were utilized. The data were divided into training and testing sets, with 90% (17,976 datasets) used for training and the remaining 10% (1,489 datasets) reserved for testing. A two-way time series analysis was employed to analyze dynamic interactions between two time-dependent behaviors over time. Additionally, the rolling forecasting method was used alongside the models to capture patterns and provide updated predictions based on the most recent data trends. The results demonstrated that RLR outperformed both the individual ML models and the hybrid deep learning TR-BiLSTM models, as indicated by the NSE and RSR statistical metrics applied to the testing data. Furthermore, the one-week step-ahead forecasting achieved greater accuracy in predicting WTR compared to the two-week step-ahead forecast. However, the average computational time of the hybrid deep learning TR-BiLSTM models was notably higher compared to the standalone models. Linear models such as RLR and LSVM demonstrated accurate forecasting results due to their ability to prevent overfitting in correlated features and effectively capture the simplicity of the relationship between the data. The approach presented in this research can be effectively useful to various arid regions worldwide that are influenced by WTR.

How to cite: Elzain, H. E., Abdalla, O., Al-Maktoumi, A., Kacimov, A., and Chen, M.: Water table rise forecasting using machine and deep learning models in arid regions, Oman, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12464, https://doi.org/10.5194/egusphere-egu25-12464, 2025.

Differentiable programming enables automatic differentiation (AD) tools to compute gradients through code without manually defining derivatives. AD tools can differentiate through entire software stacks, composing many functions and algorithms via the chain rule. With models that incorporate differentiable programming long-standing challenges like systematic calibration, comprehensive sensitivity analyses, and uncertainty quantification can be tackled, and machine learning (ML) methods can be integrated directly into the process-based core of earth-system models (ESMs) to incorporate additional information from observations. Through the advent of ML, several AD tools are gaining traction. A new generation of powerful tools like JAX, Zygote and Enzyme enable differentiable programming for models of varying complexity including highly complex coupled ESMs. Here we present an overview about the perspectives of differentiable programming for ESMs, using experience from two of our applications in atmospheric modelling. First off, we set up PseudoSpectralNet, a differentiable quasi-geostrophic model in Julia with Zygote. This is a hybrid model combining neural networks with a dynamical core, showcasing how the stability and accuracy of ML models is improved by integrating a process-based dynamical core into our model. Additionally, ongoing work uses Enzyme to achieve a differentiable version of the significantly more complex SpeedyWeather.jl atmospheric model. We will discuss advantages of both approaches and give an outlook into future possibilities with differentiable models. 

How to cite: Gelbrecht, M., Klöwer, M., and Boers, N.: Differentiable Programming for Atmospheric Models: Experiences and Perspectives , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12601, https://doi.org/10.5194/egusphere-egu25-12601, 2025.

Accurate representation of turbulent processes remains a critical challenge in atmospheric modelling. Large Eddy Simulations (LES) serve as valuable tools for understanding atmospheric turbulence by explicitly resolving energy-containing eddies while parameterizing smaller-scale motions through subgrid-scale (SGS) models. In their most complex forms, these SGS parameterizations can significantly influence LES performance and computational efficiency, making their improvement useful for advancing atmospheric modelling capabilities. Neural Network based emulation of such parametrizations have proven effective in reducing the computational cost, while maintaining accuracy and stability.

Building upon recent advances in physics-informed neural networks (NN) for atmospheric modelling and emulation of physics-based processes, we present a physics-guided NN architecture for emulation of the SGS turbulence parameterizations that introduces several key innovations. Our approach uniquely combines scale-specific normalization with multi-scale feature extraction through parallel convolutional paths, distinguishing it from existing physics-guided machine learning frameworks. The deep learning-based (DL) model also incorporates physically-motivated constraints across different spatial scales while simultaneously ensuring conservation of momentum and energy.

Unlike earlier studies that focus on single aspects of physical conservation, our architecture implements a comprehensive physics-informed framework that combines Richardson number gradient handling for stability constraints, with explicit treatment of diffusion and viscosity coefficients, and scale-specific normalization for different atmospheric variables. The model was trained on limited high-resolution Radiative-Convective Equilibrium (RCE) simulations from the Met Office-Natural Environment Research Council (NERC) Cloud model (MONC), employing physics-based loss functions that enforce both conservation laws and stability constraints.

The training dataset consisted of 3-D diagnostics data from the RCE simulations, with a 64x64 km² domain and 1 km grid spacing in the horizontal. While the original simulations had 99 vertical levels with varying vertical resolution, the DL model was trained on random slices (vertical levels) chosen from the original data volume. The inputs consisted of the resolved state variables like velocity components (u, v, w) from the previous time step, the perturbations to potential temperature, mixing ratios and Richardson number, whereas the targets for the DL model were the SGS tendencies of the model prognostic fields resulting from the Smagorisnky parameterization and the coefficients of viscosity and diffusion.

The DL model's cross-regime applicability was evaluated through multiple independent test cases: (200-second sampling frequency) and different atmospheric conditions from the Atmospheric Radiation Measurement (ARM) program. The simulations from ARM atmospheric settings were mainly targeted at simulating shallow convection, with different grid/domain configurations. Results from the off-line tests demonstrate promising performance in predicting SGS and transport coefficients (viscosity and diffusion) across these varied conditions, particularly in maintaining physical consistency during regime transitions.

Our preliminary findings indicate that this enhanced multi-scale, physics-informed architecture can effectively learn SGS parameterizations from limited training data while maintaining physical fidelity across different atmospheric conditions and spatio-temporal resolutions. This approach demonstrates the potential for the development of high-fidelity, generalizable parameterizations for weather and climate models, suggesting a route forward for reducing the greater computational costs associated with more complex SGS parameterization schemes.

How to cite: Panda, S. K., Jones, T., Shahzad, M., Ellis, A.-L., and Lawrence, B.: Physics-Guided Deep Learning-Based Emulation of Subgrid-Scale Turbulence Parameterization for Atmospheric Large Eddy Simulations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13920, https://doi.org/10.5194/egusphere-egu25-13920, 2025.

Given our current computational resources, a state-of-the-art ocean model can achieve a grid resolution in the order of 10 km for realistic global simulations, meaning that small-scale convection and wind-driven mixing near the surface of the ocean with length scales of roughly 1 m cannot be explicitly resolved. However, these microturbulent processes play a fundamental role in setting the structure of the ocean stratification, govern air-sea fluxes exchange as well as tracer transport with the interior of the ocean. Therefore, we use parameterizations, models which approximate small-scale processes using large-scale variables, to represent their effects in climate simulations.

In this work, we propose NORi: a novel, physically-principled, and data-driven approach to parameterizing ocean vertical mixing using neural ordinary differential equations (NODEs). NORi uses neural ODEs (NO) to augment a simple eddy-diffusivity closure based on the local gradient Richardson number (Ri). The Ri-based diffusivity closure captures local convective- and shear-driven mixing, while the neural ODEs augment the base model with nonlocal entrainment fluxes due to convection using neural networks. NORi is designed for realistic seawater's nonlinear equation of state using TEOS-10 and explicitly represents temperature and salinity fluxes. When compared against high fidelity large-eddy simulations (LES) of different convective strengths, background stratifications, and shear conditions, NORi demonstrates excellent prediction and generalization capabilities. By design, NORi automatically satisfies tracer invariance and conservation. It also exhibits high numerical stability and accuracy owing to its online training paradigm, where neural networks are calibrated against time-integrated field variables of interest rather than on instantaneous, time-independent turbulent fluxes. When compared against other parameterizations, NORi produces deeper mixed layers which are in better agreement with the LES solution. NORi is implemented straightforwardly into Oceananigans.jl, the fastest ocean model to date without intermediate wrappers. This can be achieved owing to the cutting-edge paradigm of the Julia programming language as well as the simple, modern and flexible interface of Oceananigans.jl. Using large-scale simulations, we demonstrate that NORi is numerically stable for at least 100 years despite being trained with only a 2-day integration, is computationally efficient, and produces realistic fields which are comparable to existing parameterization.

How to cite: Lee, X. K., Ramadhan, A., Souza, A., Wagner, G., Silvestri, S., Marshall, J., and Ferrari, R.: NORi: A Novel, Physically-Principled Approach to Parameterization of Upper Ocean Turbulence using Neural Ordinary Differential Equations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14742, https://doi.org/10.5194/egusphere-egu25-14742, 2025.

The added value of machine learning for weather and climate applications is measurable through performance metrics, but explaining it remains challenging, particularly for large deep learning models. Inspired by climate model hierarchies, we propose that a full hierarchy of Pareto-optimal models, defined within an appropriately determined error-complexity plane, can guide model development and help understand the models' added value. We demonstrate the use of Pareto fronts in atmospheric physics through three sample applications, with hierarchies ranging from semi-empirical models with minimal parameters (simplest) to deep learning algorithms (most complex). First, in cloud cover parameterization, we find that neural networks identify nonlinear relationships between cloud cover and its thermodynamic environment, and assimilate previously neglected features such as vertical gradients in relative humidity that improve the representation of low cloud cover. This added value is condensed into a ten-parameter equation that rivals deep learning models. Second, we establish a machine learning model hierarchy for emulating shortwave radiative transfer, distilling the importance of bidirectional vertical connectivity for accurately representing absorption and scattering, especially for multiple cloud layers. Third, we emphasize the importance of convective organization information when modeling the relationship between tropical precipitation and its surrounding environment. We discuss the added value of temporal memory when high-resolution spatial information is unavailable, with implications for precipitation parameterization. Therefore, by comparing data-driven models directly with existing schemes using Pareto optimality, we promote process understanding by hierarchically unveiling system complexity, with the hope of improving the trustworthiness of machine learning models in atmospheric applications.

Preprint: https://arxiv.org/abs/2408.02161

How to cite: Beucler, T., Grundner, A., Shamekh, S., Ukkonen, P., Chantry, M., and Lagerquist, R.: Distilling Machine Learning's Added Value: Pareto Fronts in Atmospheric Applications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15200, https://doi.org/10.5194/egusphere-egu25-15200, 2025.

The training of machine learning models for weather and climate on multiple datasets, including local high-resolution reanalyses and level-1 observations, is one of the frontiers of the field. It promises to allow for models that are no longer constrained by the capabilities of equation-based models, as is currently still largely the case when training on global reanalyses. For example, level-1 observations contain the feedback from arbitrary scale processes and hence do not suffer from the closure problem of equation-based models. Training on observations might hence lead to machine learning-based Earth system models with reduced systematic biases, in particular for long-term climate projections. Local reanalyses are only available for a small set of regions, mainly over Europe and North America. Appropriate training might allow one to generalize the detailed process information in these to other regions or even globally. 

In this talk, we present results on the effective training with a combination of global and local reanalysis as well as level-1 observations. We consider different pre-training protocols to learn the correlations between datasets, which is critical to obtain a benefit through their combination. We use a forecasting task as baseline and study the effectiveness of different variants of masked-token modeling and more sophisticated approaches that exploit the latent space of the machine learning models. We also study different fine-tuning strategies to extract a best state estimate from multiple datasets and to generalize regional datasets globally. For this, we build on the extensive results on fine-tuning of large language models that have been developed in the last years. Our results aim to determine general principles which combination of datasets is beneficial. We also perform a detailed analysis of the physical consistency and physical process representation in the model output. Through this, we believe our work provides an important stepping stone for the next generation of machine learning-based models for weather and climate.

How to cite: Lessig, C.: Towards next generation machine learning-based Earth system models that exploit a wide range of datasets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16817, https://doi.org/10.5194/egusphere-egu25-16817, 2025.

Mesoscale atmospheric processes that are neither resolved nor parameterized in global climate models, such as slantwise convection, can have a significant impact on climate variability and change. An example of such mesoscale influence on the climate is shown by Wills et al. 2024,  who demonstrate that circulation responses to surface anomalies are increased through heat and momentum fluxes by mesoscale processes. To enable longer simulations in comprehensive global climate models that include information about subgrid mesoscale processes, a machine learning (ML) parameterization could be applied at relatively low computational cost. So far, such ML parameterizations have been primarily applied to idealized geographies (e.g., aquaplanets), and they have not targeted midlatitude mesoscale processes in particular.

In this work, we focus on midlatitude mesoscale processes over the Gulf Stream region, as simulated by variable resolution CESM2 simulations, which have 14-km resolution over the North Atlantic. Learning subgrid fluxes from this model allows a targeted parametrization of mesoscale processes leading to vertical fluxes, namely slantwise convection and frontogenesis. We use an artificial neural network to predict vertical profiles of subgrid fluxes of momentum, heat and moisture. The features (inputs) for the ML models in this work include coarse-grained atmospheric state variables at each grid point, such as the vertical profiles of horizontal winds, temperature and their horizontal shear as well as surface pressure. The vertical profile of the specific humidity and the value of convective available potential energy are included to assess the importance of moist dynamics in the determination of subgrid convectional fluxes. Our results show that moisture variables have a rather small impact, suggesting that the subgrid fluxes can be explained by dry dynamics. A greater importance is found in the horizontal differences of neighbouring momentum and temperature columns. This suggests that neighbouring column information may be essential in the prediction of subgrid-scale fluxes, e.g., through the action of shear instabilities or conditional symmetric instability. Combined with information about the vertical localization relationship of the inputs and outputs, the goal is to feed this information into an equation discovery approach, which could lead to deeper physical understanding of mesoscale momentum and energy fluxes in midlatitudes.

How to cite: Ismaili, E., Beucler, T., and Jnglin Wills, R.: Prediction and Understanding of Subgrid-Scale Vertical Fluxes by Missing Midlatitude Mesoscale Processes Using a Machine Learning Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17078, https://doi.org/10.5194/egusphere-egu25-17078, 2025.

Accurate oil spill predictions are crucial for mitigating environmental and socioeconomic impacts. Numerical models, like MEDSLIK-II (De Dominicis et al., 2013), simulate oil advection, dispersion, and transformation, but their performance depends heavily on the configuration of physical parameters, often requiring labor-intensive manual tuning based on expert judgment.

To address this limitation, we integrate MEDSLIK-II with a Bayesian Optimization (BO) framework to systematically identify the optimal parameter configuration, ensuring simulations closely match observed spatiotemporal oil spill distributions. Our optimization focuses on horizontal diffusivity and drift factor parameters, using the Fraction Skill Score as the objective metric to maximize, thus reducing the overlap between simulations and observations.

The approach is validated on the 2021 Baniyas (Syria) oil spill, demonstrating improved accuracy, reduced biases and lower computational costs compared to the standalone numerical model.

By integrating BO with the MEDSLIK-II numerical model, our method enhances oil spill prediction capabilities and provides a transferable, physically consistent optimization framework applicable to a wide range of geophysical challenges.

This work is conducted within the framework of the iMagine European project, which leverages Artificial Intelligence, including AI-assisted image generation, to advance a series of use cases in marine and oceanographic science.

References

De Dominicis, M., Pinardi, N., Zodiatis, G., & Lardner, R. (2013). MEDSLIK-II, a Lagrangian marine surface oil spill model for short-term forecasting – Part 1: Theory. Geoscientific Model Development, 6, 1851–1869. https://doi.org/10.5194/gmd-6-1851-2013

How to cite: De Carlo, M. M., Accarino, G., Atake, I., Elia, D., Epicoco, I., and Coppini, G.: Improving Oil Spill Numerical Simulations through Bayesian Optimization, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17333, https://doi.org/10.5194/egusphere-egu25-17333, 2025.

Vertical energy transport from the land surface into the atmosphere in the form of sensible and latent heat flux must be well represented in numerical weather prediction models to allow accurate estimates of near-surface atmospheric variables. Traditionally, these heat fluxes are parameterized relying on Monin-Obukhov Similarity Theory (MOST), which is based on differences in wind speed, air temperature, and humidity between adjacent measurement or model levels. Recently, Wulfmeyer et al. (2024) estimated heat flux with machine learning at much higher accuracy compared to MOST. Their ML model proposed the incorporation of additional predictor variables when estimating latent heat flux (such as solar radiation), which stands in contrast to the classical MOST approach. However, the analysis in Wulfmeyer et al. (2024) is based on a rather short data period in August 2017 at three nearby locations in Oklahoma, USA, which limits the generalizability of the results. Here, we replicate and expand the findings from Wulfmeyer et al. (2024) on a dataset from the boundary layer field site (GM) Falkenberg of the German Meteorological Service over a period of twelve years, covering various seasons and synoptic weather situations. Our findings support the role of incoming shortwave radiation not only for latent but also for sensible heat flux estimates, particularly for other parts of the year. The results thus underline the potential to develop more advanced flux parameterizations beyond MOST. In future research, we intend to investigate the role of other predictor variables, such as vapor pressure deficit or soil moisture, to assess the generalizability of the relations, to judge their performance under extreme conditions, and to derive simple but universally applicable parameterizations.

How to cite: Butz, M. V., Karlbauer, M., Beyrich, F., and Wulfmeyer, V.: Replicating Sensible and Latent Heat Flux Diagnosis with Multilayer Perceptrons on Multi-Year Falkenberg Tower Data , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17542, https://doi.org/10.5194/egusphere-egu25-17542, 2025.

Vegetation plays a crucial role in the Earth's climate system via a number of key processes, affecting the exchange of carbon, water and energy between surface and atmosphere. The complex relationship between climate change and vegetation highlights the importance of accurate and reliable vegetation models that fully capture these interactions. 

Traditional vegetation models are primarily designed to operate on CPU architectures, which restricts their ability to exploit advancements in modern parallel computing architectures such as GPUs. Furthermore, limited process knowledge and the absence of direct observations and/or quantitative theories for certain processes hinder accurate representation of these processes, which introduces uncertainties in the model results, leading to discrepancies when compared to observations. The rigid structure of these traditional models also makes integration of new processes challenging and hinders the application of advanced optimization techniques for automatic parameter tuning and objective calibration using abundant observational data due to their non-differentiable nature.

This work follows a new paradigm in vegetation modeling that integrates the robustness of traditional models with the adaptive power of machine learning techniques. The goal is to combine reliable physical components with machine learning components. As opposed to classical vegetation models, the resulting hybrid model is differentiable and the parameters of both the physical and the neural network components can be optimized jointly and efficiently using observational data.

In the proposed hybrid vegetation model, machine learning can be used to improve the computational efficiency of the model by emulating computationally expensive routines. We have implemented the key processes related to photosynthesis in LPJ in Julia. This minimal model setup is used to explore the potential of machine learning to replace the computationally expensive root-finding algorithm used in computing the optimal ratio of intercellular to ambient CO2 concentration, and hence stomatal conductance.

Machine learning can also be used for better process representation. The recently developed neural or universal differential equations offer a particularly promising methodological framework for learning the dynamics of carbon allocation to different vegetation pools using observations. The dynamics of carbon allocation to different plant components can be effectively modeled using a neural ODE approach, which utilizes observations of observable variables (e.g., Above Ground Biomass (AGB), Leaf Area Index (LAI)) to learn the dynamics of unobservable variables such as vegetation carbon pools.

How to cite: Badri, M., Hess, P., Lin, Y., Bathiany, S., Gelbrecht, M., and Boers, N.: Towards a Hybrid Vegetation Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17958, https://doi.org/10.5194/egusphere-egu25-17958, 2025.

Modelling microplastic transport through porous media, such as soils and aquifers, is an emerging research topic, where existing hydrogeological models for (reactive) solute and colloid transport have shown limited effectiveness thus far. This perspective article draws upon recent literature to provide a brief overview of key microplastic transport processes, with emphases on less well-understood processes, to propose potential research directions for efficiently modeling microplastic transport through the porous environment. Microplastics are particulate matter with distinct physicochemical properties. Biogeochemical processes and physical interactions with the surrounding environment cause microplastic properties such as material density, geometry, chemical composition, and DLVO interaction parameters to change dynamically, through complex webs of interactions and feedbacks that dynamically affect transport behavior. Furthermore, microplastic material densities, which cluster around that of water, distinguish microplastics from other colloids, with impactful consequences that are often underappreciated. For example, (near-)neutral material densities cause microplastic transport behavior to be highly sensitive to spatio-temporally varying environmental conditions. The dynamic nature of microplastic properties implies that at environmentally relevant large spatio-temporal scales, the complex transport behavior may be effectively intractable to direct physical modeling. Therefore, efficient modeling may require integrating reduced-complexity physics-constrained models, with stochastic or statistical analyses, supported by extensive environmental data. This is a sub-project (focusing on microplastics in the environment) of the Digital Waters Flagship funded by the Research Council of Finland, where we aim to create a digital ecosystem for machine learning aided hydrological modelling of various hydrosphere processes across all environmental compartments, focusing particularly on the critical zone.

How to cite: Tang, D. and Yang, X.: Modeling microplastic transport through porous media: challenges arising from dynamic transport behavior, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18139, https://doi.org/10.5194/egusphere-egu25-18139, 2025.

Permeability should be dispersed conveniently to control the aquifer's type and quality. Permeability in a variety of porous media can be determined using different methods depending on the environment and the scope of the porosity media. These days, permeability of core samples and well logging data with greater aquifer heterogeneity, artificial intelligence algorithms are well-known for estimating permeability. Machine learning and artificial intelligence have gained popularity and credibility across all scientific fields. To address the dearth of resources in geosciences generally and hydrology specifically.

As soft computing techniques, Artificial Neural Networks (ANNs) have demonstrated the capacity to estimate acceptable outputs with tolerable outcomes. The ANN model uses basic processing units, which are networks of interconnected neurons. The simplest approach is the Feed-Forward Artificial Neural Network (FF-ANN). The Middle Jurassic Hugin Formation may have been deposited as a mouth bar setting during the period of general transgression, as evidenced by fluctuating permeability values brought on by changes in the sediment supply, which varying porosity values brought on by variations in the amount of clay and size of grains.

Keywords: Artificial Neural Network, Feed-Forward Artificial Neural Network, Volve oilfield, Hugin Formation, Permeability estimation.

How to cite: Ghattas, K. and Buday, T.: Artificial Neural Network Approaches for Permeability Estimation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18810, https://doi.org/10.5194/egusphere-egu25-18810, 2025.

Machine learning (ML) has the potential to reduce systematic uncertainties in Earth System Models by replacing or complementing existing physics-based parameterizations of sub-grid processes. However, after decades of research, ensuring generalization and stability of ML-based parameterizations remains a major challenge.  We aim to minimize both epistemic and aleatoric sources of uncertainty via physically inspired, vertically recurrent neural networks (RNN) which offer key benefits such as parametric sparsity and efficient modeling of non-locality in a column. To address aleatoric uncertainty, we furthermore incorporate stochasticity and convective memory into the ML architecture. We present preliminary results using the ClimSim framework, where the physically inspired ML framework replaces a superparameterization in a low-resolution climate model.

How to cite: Ukkonen, P., Mansfield, L., and Christensen, H.: Emulation of sub-grid physics using stochastic, vertically recurrent neural networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19319, https://doi.org/10.5194/egusphere-egu25-19319, 2025.

Reliable climate projections are crucial for informed decision-making in water resource planning and management. However, selecting suitable Global Climate Models (GCMs) remains challenging due to inherent uncertainties and computational constraints. This study introduces a novel hybrid approach for GCM selection, focusing on models that exhibit consistency in projecting future climate changes and skill in representing current climate conditions, including average climate, seasonal patterns, and climatic variations. GCM performance in simulating these critical properties was evaluated for rainfall, maximum temperature, and minimum temperature using the Kling-Gupta Efficiency (KGE) metric, resulting in a structured 3×3 performance matrix for each GCM. The matrix distances, quantifying the disparities between each GCM's performance matrix and the ideal reference matrix, were used to represent overall model performance. GCMs were then ranked based on these differences using the Jenks natural breaks classification method to identify the top-performing models for ensemble construction. The proposed method was tested by selecting GCMs for Nigeria from 19 CMIP6 GCMs. Results indicate that 15 GCMs consistently projected future climate within a 95% confidence interval. Further evaluation reveals that ACCESS.ESM1.5, BCC.CSM2.MR, CMCC.ESM2, and MRI.ESM2.0 are the most suitable for simulating Nigeria's climate. The multi-model ensemble means of the selected GCMs projected a notable increase in rainfall by 10 to 40% over most of the country and maximum and minimum temperatures by 1.0 to 3.5°C and 0.5 to 4.0°C, respectively. The proposed approach offers an effective tool for GCM selection to enhance climate projection reliability.

How to cite: Kumar, S., Choudhary, M. K., and Thomas, T.: Comparative Analysis of GCM Selection Approaches for Climate Change Impact Assessment in India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1118, https://doi.org/10.5194/egusphere-egu25-1118, 2025.

Stratospheric aerosols, most of which originate from explosive volcanic sulfur emissions into the stratosphere, are a key natural driver of climate variability. They are thus one of the forcings provided by the Coupled Model Intercomparison Project (CMIP) Climate Forcings Task Team for the CMIP7 Fast Track, a set of climate model experiments designed to deliver the Intergovernmental Panel on Climate Change (IPCC) 7^th assessment cycle. In this work, we document the final version of the stratospheric aerosol forcing datasets delivered to modelling groups for CMIP7 Fast Track. Our datasets cover the 1750-2023 period to meet to the need of modelling groups who might run extended historical simulations starting in 1750 instead of 1850. We produced one volcanic stratospheric sulfur emission dataset catering for the needs of models which have a prognostic interactive stratospheric aerosol scheme, as well as a stratospheric sulfate aerosol optical property dataset required by models that cannot interactively simulate stratospheric sufate aerosols. For the satellite era (from 1979 onwards), sulfur emissions and sufate aerosol optical properties are based on the MSVOLSO2L4 and GloSSAC datasets, respectively. For the pre-satellite era (1750-1978), the emission dataset is based on ice-core datasets complemented by the geological record for small-moderate magnitude eruptions not captured in ice-core records. Although inferring emissions of these eruptions from the geological record is highly uncertain, our approach minimizes an important bias in the pre-satellite era forcing, both in terms of mean and variability. The pre-satellite aerosol optical property dataset is directly derived from emissions using an updated version of EVA_H, a reduced-complexity volcanic aerosol model. This ensures methodological consistency between our emission and optical property datasets, and maximizes consistency with methodologies used in the paleoclimate (PMIP) and volcanic forcing (VolMIP) model intercomparison projects in CMIP6. We will present extensive comparison between our CMIP7 Fast Track dataset and the CMIP6 dataset. Last, we will discuss the main challenges to improve stratospheric aerosol datasets in the future and to move to high frequency (yearly or less) extension and update instead of an ad-hoc production for each CMIP phase.

How to cite: Aubry, T., Toohey, M., Schmidt, A., Kovilakam, M., Sigl, M., Khanal, S., Chim, M. M., Johnson, B., Carn, S., Verkerk, M., Nicholls, Z., and Smith, I.: Historical stratospheric aerosol optical properties and volcanic sulfur emissions for CMIP7 Fast Track, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3831, https://doi.org/10.5194/egusphere-egu25-3831, 2025.

3D model is a 3D digital representation of objective things, which has been widely applied in fields like urban construction, disaster prevention and mitigation, medical research, biological science, industrial manufacturing, agricultural production, etc. As a special 3D model, 3D geological model possesses the characteristics of 3D model and plays a fundamental role in geological survey, mineral exploitation, underground engineering and smart city construction.With the development of intelligent sensing technology and 3D geological modeling technology, the scale of 3D geological model data increases exponentially. Meanwhile, with the pace of large-scale underground engineering and smart city continuing to increase, 3D geological model with fine large scenes is being eagerly required. The rapid growth of data and the refinement of large application scenes bring new challenges to the real-time dynamic visualization of 3D geological models. These challenges are mainly reflected in the new technical problems related to 3D geological model rendering.This study focuses on 3D geological model rendering and puts forward the corresponding solutions. The validity of the technology has been proved by the simulation test of cluster cloud environment consisting of 5 computers. The technique has been applied in the construction of 3D geological information and visualization system in transparent Xiong’an.Firstly, the data organization mode of two common structures of 3D geological model (3D geological structure model and 3D geological high-precision grid model) is analyzed, and a distributed storage strategy of 3D geological model based on MongoDB is proposed. Aiming at the characteristics of multi-layer data in z-direction of 3D geological structure model, an octree index mechanism is proposed to improve the efficiency of data scheduling according to the z-direction spatial information and layer information. The rendering optimization of a single node 3D geological model is studied. The rendering in the cloud environment still needs the cooperation of each sub-node. Therefore, the overall rendering efficiency in the cloud environment can be improved by adopting efficient rendering optimization strategies for the 3D geological model of each node and selecting an effective node scheduling strategies. Single-node 3D geological model rendering is mainly performed by transferring data from memory to GPU. The communication between memory and GPU is a bottleneck, which will affect the overall rendering efficiency. Through the strategies of visibility elimination, LOD establishment, data merging and instance rendering optimization, this thesis effectively reduces the number of drawing calls and communication times. How to optimize and improve the overall performance of 3D geological model rendering in cloud environment from a global perspective is studied, and a multi-level distributed SCMP framework is proposed, which integrates the advantages of cluster, GPU, distributed storage, etc., to maximize the distributed computing ability of existing machines and improve the rendering efficiency in cloud environment. From the experimental data, the node invocation optimization strategy with “GPU+CPU” can ensure that the frame rate of the four rendering nodes and the end-user scene in the cloud environment is stable at about 35 frames per second, and can achieve satisfactory cluster load balancing effect.

How to cite: Song, Y., Gao, Z., Song, G., and Li, J.: Research on Key Technologies of 3D Geological Model Rendering in Cloud Environment, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4053, https://doi.org/10.5194/egusphere-egu25-4053, 2025.

As a data-intensive science, earth system science has been focusing on the research of living environment and constituent element’s characteristics, including its forming time, location and evolution. With big data and AI boosting, there are more opportunities and challenges for geological research transformation. And it is more likely to improve geological survey and geological research by means of information method such as AI algorithm, methods, tools, software, and etc.. As for the storage, distribution and application of different format and discipline data, the key is to set up a series of rules and tools to realize the data services’ flexible using in security. It adopts hybrid data management framework to build up an unified index to support the spatial geological data finding. The matching platform is also developed to realize the geological achievements distribution. Moreover, it can significantly benefit faster and more efficient research. In all, the technique has been applied successful on Chinese geological survey information platform with great reuse adaptability on other platforms.

How to cite: cui, N. and gao, Z.: Research and application of key technologies of geological data platform, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5383, https://doi.org/10.5194/egusphere-egu25-5383, 2025.

Climate forcings are the input drivers to coupled climate models (AGOCMs) and earth system models (ESMs). They are routinely used as part of the coupled model intercomparison project (CMIP). Here we present the historical greenhouse gas forcing used in the seventh phase of CMIP (CMIP7) and compare it to its predecessors from CMIP6 and CMIP5. We show that revised methods and input data have had little effect on historical estimates of greenhouse gas forcing, but that greenhouse gas forcing has continued to increase since 2015 (the end of the CMIP6 historical experiment), even if forcing from some specific gases has decreased. Beyond the greenhouse gas forcings, there are a number of other forcings involved in CMIP. Following on from our involvement in the CMIP Forcings Task Team, we present an outline for moving towards sustained, roughly annual, releases of these forcings and discuss the challenges for realising this possibility.

How to cite: Nicholls, Z., Pflüger, M., and Meinshausen, M.: CMIP7 historical greenhouse gas forcing and steps towards sustained releases, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5765, https://doi.org/10.5194/egusphere-egu25-5765, 2025.

With the major challenges posed by climate change and significant shifts in Earth systems, the need for high-precision and diverse climate predictions has grown. These predictions aim to explore a variety of scenarios, such as the Shared Socioeconomic Pathways (SSPs). Advances in computational power have enabled the development of sophisticated coupled physical-biogeochemical-ecological models of marine systems. However, these models remain computationally intensive and energy-demanding, raising questions about the appropriate level of complexity relative to the availability of independent data for accurate calibration, and calls for simplification to reduce execution time. Here, we aim to simplify the Eco3M-MED model, which is a complex biogeochemical model representing the low trophic levels (up to mesozooplankton) in the ocean through 37 state variables, and which is intended to be run at the scale of the Mediterranean basin.

Common simplification methods include conservation analysis, lumping, time exploration, and sensitivity analysis. Since most of these simplification methods reduce or even penalize the ability to interpret model results, or require complex implementation, we have chosen a simple, classic method, based on the local sensitivity analysis (One-Factor-At-A-Time, OFAT) method that does not impair this ability. This work's originality lies in the approach adopted to obtain different declinations of the reduced model. This approach indeed benefits from an original strategy for parametrizing the Eco3M-MED model, initiated several years ago and recently implemented in practice. This strategy consists of the construction of a set of consistent parameters, resulting in the establishment of relations between the so-called core parameters and dependent parameters. Core parameters are perturbed based on the level of knowledge of each parameter. The main objective of this study is to apply this novel approach to identify the biogeochemical processes that can be removed with minimal impact on model performance, thereby enabling model simplification and reducing computational costs. We also apply the principle that a single simplified model is not necessarily the best solution, and aim instead to derive a family of simplified models associated with different usage objectives, ensuring that the simplified model reproduces certain quantities well in particular.  The criteria used to derive a simplified model from the sensitivity analysis are also subject to analysis to identify their influence on the degree of simplification. Finally, the computational efficiency and accuracy of simplified models were compared with the full model to determine optimal simplification for specific applications. Future research will focus on performing global sensitivity analysis on high-impact core parameters to assess uncertainties in both the full and simplified models.

How to cite: Zhang, Y., Baklouti, M., and Brasseur, P.: Sensitivity-driven simplification of complex ecosystem models: Integrating mechanistic insights for cost reduction and predictive accuracy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6300, https://doi.org/10.5194/egusphere-egu25-6300, 2025.

Earth system reconstructions are key to understanding past climate dynamics, environmental variability, and biogeochemical cycles, revealing how Earth responds to natural and human influences. These reconstructions require integrating geological, geophysical, and geochemical data with advanced computational models.

Recent advancements in Large Language Models (LLMs) enhance the processing of complex datasets, improving the accuracy and predictive power of Earth system reconstructions. To leverage this, we develope GeoGPT, a domain-specific LLM for geosciences, trained on open-source data. This open-source, non-profit project encourages broad collaboration among experts in broad branches of the geoscience and AI. GeoGPT helps advance Earth system reconstructions, offering deeper insights into Earth's past and future, and guiding responses to environmental challenges.

By harnessing the combined strengths of Geoscientists, AI experts, and the broader research community, GeoGPT aspires to unlock new avenues of exploration, accelerate breakthrough discoveries.

How to cite: Chen, H.: GeoGPT and Its Potential Applications for Earth system Reconstructions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7628, https://doi.org/10.5194/egusphere-egu25-7628, 2025.

Active margin topography has profoundly shaped Earth’s climate, biodiversity, and natural resource distribution over geological time. However, reconstructing paleotopography in these regions remains challenging due to the sparse and uneven distribution of proxies like stable isotope paleoaltimetry, palynology, paleobotany, and thermochronology. These traditional methods often leave large spatial and temporal gaps, with uncertainties in paleoelevation estimates reaching up to 2,000–3,000 m. To address these challenges, we introduce an innovative workflow utilizing artificial intelligence to reconstruct paleotopography at active margins since the Devonian. Using Explainable Boosting Machines (EBMs), we identify key factors such as plate kinematics, mantle dynamics, and climate that govern active margin topography. Insights from the EBM analysis guided the development of a Random Forest (RF)-based regressor which was then used to predict paleotopography through time. Our RF model achieved a mean error of 554 m when validated against present-day ETOPO elevation data. Our model highlights time-evolving subduction flux, trench migration rates, and upper mantle temperature as the primary controls on active margin topography. To validate our approach, we compare our reconstructions with existing paleotopographic models and geological proxies in two regions: the Cenozoic Andes and Mesozoic-Cenozoic Eastern China. For the Andes, our model closely matches the existing reconstructions, highlighting a ~4,000 km rapid rise of the Altiplano since the late Oligocene, driven by an increase in subduction flux (from 0.03 km³/yr to 0.10 km³/yr) and a transition in trench migration from retreating (2 cm/yr) to stationary, likely due to slab anchoring. In Eastern China, our model predicts sustained high topography (>2,500 m) during much of the Cretaceous, attributed to high subduction flux (>0.12 km³/yr) from the Pacific Plate and an advancing trench. A subsequent shift to trench retreat (-2 cm/yr) in the Late Cretaceous–Early Cenozoic led to back-arc extension and a decline in elevation to ~1000 m. Our study offers a transformative approach to bridging gaps in paleotopographic constraints, improving our understanding of the interplay between surface and interior processes. By providing a robust framework for reconstructing past landscapes, our model has significant implications for studying ecosystems, biodiversity evolution, and the metallogenesis of convergent margins.

How to cite: Singh, S. P., Seton, M., Zahirovic, S., and Wright, N. M.: Reconstructing the Topographic Evolution of Active Margins Since the Devonian Using Artificial Intelligence, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9233, https://doi.org/10.5194/egusphere-egu25-9233, 2025.

Earth system modelling is an important instrument to investigate climate change in an integrated way, taking into account the interactions between the different compartments of the Earth system. It is also an important tool to perform climate projections for different climate scenarios in order to take appropriate mitigation and adaptation measures. Such climate simulations are coordinated internationally as part of the World Climate Research Programme’s (WCRP) Coupled Model Intercomparison Project Phase 7 (CMIP7).

Climate forcings are key for defining the main drivers of climate change in climate simulations. A very important aspect of the CMIP7 intercomparison is that all participating models were run under similar experimental conditions. In particular, in using the same climate forcings in the different models.

The Alfred Wegener Institute (AWI) will participate in the CMIP7 project with two state-of-the-art Earth system models AWI-CM3 and AWI-ESM3. This is being done as part of the German contribution to the Coupled Model Intercomparison Project (CAP7). The AWI contribution to CAP7 includes the adaptation of the AWI-CM3 model to be able to use different forcing data (such as greenhouse gases, solar forcing, O₃, and aerosol forcing) to fulfil the requirements of CMIP7. One task is therefore the implementation of the climate forcing dataset for anthropogenic aerosols MACv2-SP (currently available for CMIP6plus [Fiedler and Sudarchikova, 2024]) provided for CMIP7. For this purpose, an aerosol interface will be implemented in the AWI-CM3 climate model to read and process the aerosol forcing data provided by the MACv2-SP dataset.

In this presentation we will discuss the implementation of the MACv2-SP data into the AWI-CM3 climate model and present first results of the responses to these forcings.

How to cite: Wieters, N., Streffing, J., Goessling, H., and Jung, T.: Preparing AWI-CM3 for CMIP7: Implementing anthropogenic aerosol forcing (MACv2-SP), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10752, https://doi.org/10.5194/egusphere-egu25-10752, 2025.

General circulation models (GCM) can reasonably reproduce the global mean temperature trend during the historical period. In this study, we examine the performance of CMIP6 models in reproducing the changes in land-ocean temperature contrast between 1979-2014 during the comprehensive satellite observation era. The observed land-ocean warming contrast, defined as the land warming trend divided by the ocean warming trend, is completely outside the model spread of the historical scenario, indicating that the models severely underestimate land temperature increase relative to global mean temperature warming. Even when sea surface temperatures are prescribed to observations in AMIP experiments, the land warming trend remains outside the model spread, particularly between 15S and 15N. This was shown to be because GCM overestimates the increase in specific humidity on tropical land and underestimates the drying trend on tropical land. Since future projections over land have a significant impact on human activity, improving the representation of tropical land surface processes in GCMs is essential.

How to cite: Toda, M., Kang, S., and Shaw, T.: Climate Models Underestimate Satellite Era Land-ocean Warming Contrast in the Tropics, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11291, https://doi.org/10.5194/egusphere-egu25-11291, 2025.

The CMIP ensembles represent a partial exploration of our uncertainty in the future physical climate under various scenarios for future greenhouse gas emissions. They are thus valuable tools for exploring the potential consequences of climate change for society. One aspect of this is the impact on global and national economies. In the economics literature this is often addressed through a “damage function” which relates economic damages to national, regional or global changes in temperature.

Here we will present an assessment of the economic damages implied by the CMIP6 ensemble for various nations/regions, different Shared Socio-Economic pathways, and, crucially, a variety of different damage functions. A number of important factors will be highlighted including:

• The uncertainty in economic damages which arises from the chaotic nature of the climate system, characterised by those CMIP6 models with relatively large initial condition ensembles.
• The relative consequences for economic assessments of uncertainty in the damage function, the choice of CMIP6 model (model uncertainty), chaotic uncertainty (initial condition uncertainty), and scenario uncertainty.
• How these factors vary by country and region.

CMIP6 only represents a limited exploration of uncertainty in the physical climate response and there is also considerable uncertainty in the damage functions beyond that currently explored in the literature. These represent deep uncertainty. We will present plans for future work to embed more thorough explorations of epistemic uncertainty into future analyses, including the consequences of crossing tipping points. These considerations are valuable when considering the design and implementation of the CMIP7 project. What would be the most useful design characteristics if the target were economic assessments? We will address this question in terms of both the size of initial condition sub-ensembles, the diversity of models included, and the value of a mixture of higher and lower resolution model implementations.

How to cite: Rosser, J. and Stainforth, D.:  Economic consequences of CMIP diversity and targets for CMIP7, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12178, https://doi.org/10.5194/egusphere-egu25-12178, 2025.

A common set of simulation is important for intercomparison between different models. The ‘entry-card’ to participate in CMIP is to perform the baseline DECK simulations (1850-control, 1pctCO2, abrupt-4xCO2, historical-amip). However, performing the control and historical simulations from an 1850 baseline is prohibitively expensive for high-resolution, fully-coupled, general-circulation-models (GCMs). Therefore, HighResMIP chose to use a shorter experimental protocol based on 1950 conditions alongside a shorter spin-up length and simplified aerosols. Because of this difference in protocol, it is not clear exactly how the HighResMIP simulations relate to the other CMIP simulations. In this study we analyse the control and historical simulations of the HadGEM3-GC3.1 model, which performed control and historical simulations based on both 1950 and 1850 baselines. Our results show that the absolute temperature is sensitive to the different experimental protocol, but the anomalies are much more comparable. This opens an interesting discussion on whether climate change should be discussed in terms of absolute values or anomalies. The difference in the absolute value (and mean state) is largely due to the different aerosol scheme used in CMIP and HighResMIP for this particular model. The second phase of HighResMIP no longer require models to use Easy Aerosol, so modelling centres should use the same aerosol scheme if they would like their HighResMIP simulations to be comparable to CMIP simulations.

How to cite: Lai, M. and Roberts, M.: 1950-control vs 1850-control: How do HighResMIP simulations relate to CMIP simulations?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12926, https://doi.org/10.5194/egusphere-egu25-12926, 2025.

The projected changes in the hydrological cycle under global warming remain highly uncertain across current climate models. Here, we demonstrate that the observational past warming trend can be utilized to effectively constrain future projections in mean and extreme precipitation on both global and regional scales. The physical basis for such constraints relies on the relatively constant climate sensitivity in individual models and the reasonable consistency of regional hydrological sensitivity among the models, which is dominated and regulated by the increases in atmospheric moisture. For the high-emission scenario, on the global average, the projected changes in mean precipitation are lowered from 6.9% to 5.2% and those in extreme precipitation from 24.5% to 18.1%, with the inter-model variances reduced by 31.0% and 22.7%, respectively. Moreover, the constraint can be applied to regions in middle-to-high latitudes, particularly over land. These constraints result in spatially resolved corrections that deviate substantially and inhomogeneously from the global mean corrections. This study provides regionally constrained hydrological responses over the globe, with direct implications for climate adaptation in specific areas.

How to cite: Dai, P., Nie, J., Yu, Y., and Wu, R.: Constraints on regional projections of mean and extreme precipitation under warming, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14119, https://doi.org/10.5194/egusphere-egu25-14119, 2025.

The global sea surface warming pattern emerged since the early 1980s is characterized by a boomerang-shape warming in the western Pacific, the basin-wide warming in the Indian Ocean north of 30°S, and a triple-stripe warming in the North Atlantic. This pattern can be obtained with or without El Niño/La Niña signals, indicating the independence of El Niño/La Niña, and is the leading EOF with the El Niño/La Niña signals removed. A negative phase of this pattern started emerging in the early 1980s, switched to positive phase in the 1990s, and has been becoming more prominent for the past few years.

CMIP models have been found to have difficulty simulating observed global sea surface temperature (SST) trend, especially the cooling trend in the tropical eastern Pacific. However, the cooling trend in the tropical eastern Pacific in the past four decades is statistically insignificant in our trend analysis adopting a more stringent signal detection method (namely, the False Discovery Rate, FDR). By applying the same trend detection and EOF approach to the simulated SST in the historical simulations of forty CMIP6 models by removing El Niño/La Niña signals, we detected in the ensemble mean SST a trend pattern closely resembling the observed, which also changes from negative to positive phases in the late 1990s and continues becoming more positive into 2014. Whereas each model has slightly different performance in simulating this trend pattern, the ensemble time series of corresponding trend pattern in each model correctly reflects the emergence and enhancement of the warming pattern in the past four decades. However, this model ability seems to be masked by the large fluctuations of El Niño/La Niña, an intrinsic climate mode contributing large internal variability to the global domain, and its temporal fluctuations cannot be synchronized in the coupled models in the historical experiments, which are strongly driven by continuously increasing radiative effect of greenhouse gases concentration. On the other hand, The models seem to be capable of simulating the emergence of the global ocean warming pattern in response to the prescribed increasing greenhouse gas concentration.

How to cite: Hsu, H.-H. and Chen, Y.-L.: CMIP6 Models Properly Simulate the Emergence of Global Ocean Warming Pattern, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14140, https://doi.org/10.5194/egusphere-egu25-14140, 2025.

GeoGPT is a non-profit domain-specific Large Language Model for geosciences, trained based on open-source data. It provides an effective solution to the challenges of managing large data volumes, complex formats, and low efficiency in the utilization of books and papers in the field of paleontology. Its powerful data extraction capabilities will significantly enhance the efficiency of extracting, analyzing, and building databases for data of various formats, sizes, and origins. This enables scientists to construct online fossil datasets and empowers paleontologists to develop innovative tools such as paleontological classification assistants. Not only does this accelerate scientific research progress, but it also makes the acquisition and application of paleontological data, such as invertebrate fossils, more convenient, ultimately driving comprehensive progress in the field of paleontology.

How to cite: Xiang, Z.: GeoGPT: Transforming Paleontology with AI-Powered Data Extraction and Analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14239, https://doi.org/10.5194/egusphere-egu25-14239, 2025.

One of the major challenges faced by the geotectonic community is how to determine the paleolongitude of continents and tectonic plates as we try to reconstruct Earth’s tectonic history back in time, because classic paleomagnetic record is only sensitive to paleolatitude.  Torsvik et al. (2014) previously used mantle structure as a reference frame for palaeolongitude constraints back in Earth history, assuming that the two equatorial and antipodal large low shear velocity provinces (LLSVPs) observed in present-day Earth’s lower mantle are fixed and stable ancient structures unrelated to plate tectonic history and subduction geometry. However, such an assumption is inconsistent with true polar wander (TPW) record (Li et al., 2004, 2023), the cyclic occurrence of global mantle plume activity coupled with the supercontinent cycle (Li et al., 2008; and Zhong, 2009), and geodynamic modelling results (Zhong et al., 2007; Zhang et al., 2010; Flament et al., 2017).

In a recent paper of Li et al. (2023), we utilized palaeomagnetically interpreted TPW record, particularly inertia interchange true polar wander (IITPW) events, and global mantle plume record, to develop a dynamic global mantle reference frame that not only provides a first-order mantle dynamic evolution for the past 2 billion years, but also for the first time provides a way to trace the longitudinal change of continents and tectonic plates back in time. In particular, through the recognition of newly-defined type-1 and type-2 IITPW events coupled with plume record checking, we are now able to hypothesis that: (1) in periods with type-1 IITPW, the concerned supercontinent had developed its own degree-2 mantle structure (e.g., the antipodal LLSVPs divided by concurrent circum-supercontinent subduction girdle); (2) in periods with type-2 IITPW, a young supercontinent or multiple plates during the assembly of that supercontinent were moving over a legacy degree-2 mantle structure of the immediate ancestor supercontinent prior to the maturity of its own mantle structure. In our model, Nuna (lifespan 1600–1300 Ma) assembled at about the same longitude as the latest supercontinent Pangaea (lifespan 320–170 Ma), with an equatorial degree-2 mantle structure starting to exist as early as ca. 1700 Ma. Rodinia (lifespan 900–720 Ma) formed through introversion assembly over the legacy Nuna subduction girdle either ca. 90^◦ to the west or to the east before the subduction girdle surrounding it generated its own degree-2 mantle structure by ca. 780 Ma (but not before 800 Ma). Pangea assembled over the subduction girdle of legacy Rodinian degree-2 mantle structure, with its own degree-2 mantle structure (the one we still observe today) formed no much earlier than 270 Ma.

References

Flament, N., Williams, S., Müller, R.D. et al., 2017. Nat. Commun. 8, 14164.

Li, Z.-X., Liu, Y. and Ernst, R., 2023. Earth-Sci. Rev. 238, 104336.

Torsvik, T.H., van der Voo, R., Doubrovine, P.V. et al., 2014. Proc. Natl. Acad. Sci. 111 (24), 8735–8740.

Zhang, N., Zhong, S., Leng, W., Li, Z.-X., 2010. J. Geophys. Res. Solid Earth 115(B6), B06401.

How to cite: Li, Z.-X.: Absolute longitudinal constraints for palaeogeographic reconstruction based on a dynamic mantle reference frame, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15897, https://doi.org/10.5194/egusphere-egu25-15897, 2025.

Back in 2017, solar forcing recommendations for the 6th round of the Coupled Model Intercomparison Project (CMIP) were provided which covered, for the first time, all relevant solar irradiance and energetic particle contributions. Since then, this dataset has been extensively used in climate model experiments and has been tested in various intercomparison studies. Further, new datasets have been come available. An International Space Sciene Institute (ISSI) Working Group has been established to review these recent achievements in order to define the strategy for building a revised solar forcing dataset for the 7th round of CMIP. After receiving community feedback on this strategy, a historical solar forcing dataset for CMIP7 has been recently constructed. Major changes with respect to CMIP6 include the adoption of the new Total and Spectral Solar Irradiance Sensor (TSIS-1) solar reference spectrum for solar spectral irradiance and an improved description of top-of-the-atmosphere energetic electron fluxes, as well as their reconstruction back to 1850 by means of geomagnetic proxy data. Solar irradiance varaibility in the reference forcing dataset is based on historical reconstructions generated with the new empirical NASA NOAA LASP (NNL) Solar Spectral Irradiance Version 1 model, NNLSSI1. In adition, an alternative solar irradiance dataset, based on SATIRE, is provided for sensitivity experiments. In this talk we will discuss the applied modifications with respect to CMIP6 and their implication for climate modeling. Ongoing activities on solar forcing uncertainty quantification and the construction of future solar forcing scenarios will also be summarized.

How to cite: Funke, B., Dudok de Wit, T., Haberreiter, M., Marsh, D., Ermolli, I., Kinnison, D., Nesse, H., Seppälä, A., Sinnhuber, M., Usoskin, I., Asikainen, T., Bender, S., Chatzistergos, T., Coddington, O., Koldoboskiy, S., Lean, J., van de Kamp, M., and Verronen, P.: Solar forcing for CMIP7, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15908, https://doi.org/10.5194/egusphere-egu25-15908, 2025.

Paleomagnetism provides the main quantitative tool for reconstructing Earth’s paleogeography. Apparent polar wander paths (APWPs), derived from paleomagnetic data, trace the motion of tectonic plates relative to the Earth’s rotation axis through geological time, providing a paleogeographic framework for studying the evolution of Earth’s interior, surface, and atmosphere. Traditionally, APWPs are calculated from study-mean paleomagnetic poles that are assigned equal weight, regardless of the number of paleomagnetic sites used to compute it and the uncertainties in the position or age of the pole. Here, we introduce the next generation of APWPs that are calculated from site-level paleomagnetic data instead of from study-mean poles. This alternative approach assigns larger weight to larger data sets and allows the incorporation of spatial and temporal uncertainties. We demonstrate the advantages of this new method with recently published APWPs based on compiled (Gallo et al., 2023) and simulated site-level data (Vaes et al., 2023). We show how the latter, a global APWP for the last 320 Ma, provides more reliable estimates of the apparent polar wander rate of all major tectonic plates, and discuss its implications for the rate and magnitude of true polar wander since 320 Ma. In addition, we introduce APWP-online.org: an online, open-source environment that provides user-friendly tools to compute site-level APWPs and to use them to quantify relative paleomagnetic displacements. We showcase how these tools are currently used to compute site-level APWPs, e.g., for the North China block and Tibetan terranes. Finally, we provide future directions for the construction of APWPs and highlight opportunities for improving their quality and resolution.

How to cite: Vaes, B. and van Hinsbergen, D.: Reconstructing paleogeography using site-level apparent polar wander paths, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16299, https://doi.org/10.5194/egusphere-egu25-16299, 2025.

The provision of solar forcing datasets for CMIP7 includes a dataset with scenarios from the present to 2300. This dataset contains daily values of the same variables as in the historical solar forcing for CMIP7, namely: solar spectral irradiance, medium energy electrons, solar energetic protons and galactic cosmic rays. In contrast to CMIP6, which had only two scenarios, for CMIP7 we will provide a large ensemble of scenarios to avoid selection bias.

Let us stress that we are providing scenarios, not forecasts: the reconstructions vary randomly in time, but their statistical and spectral properties are fully consistent with historical variations, providing realistic surrogates for solar forcing.

In this presentation we explain how historical observations are used to build these surrogate reconstructions. This process involves several steps, starting with the 14C reconstructions of past solar activity. These will be described in detail, together with the first version of the dataset. 

How to cite: Dudok de Wit, T., Funke, B., Haberreiter, M., Marsh, D., Ermolli, I., Kinnison, D., Nesse, H., Seppälä, A., Sinnhuber, M., Usoskin, I., Asikainen, T., Bender, S., Chatzistergos, T., Coddington, O., Koldoboskiy, S., Lean, J., van de Kamp, M., and Verronen, P.: Solar forcing for CMIP7: making of future scenarios, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17597, https://doi.org/10.5194/egusphere-egu25-17597, 2025.

A Digital Twin (DT) is a data-driven model of a physical entity with two-information flows that enables the direct interaction between both. DTs of the natural environment are typically constructed by fusing multi-modal measurements of some physical phenomena using Artificial Intelligence (AI) methods. The physical entity interacts with the DT through natural changes whilst the DT interacts with the physical entity through automated changes in sensing systems and through decision-making processes. Large-scale DTs of the Earth system are currently in development through initiatives such as Destination Earth (DestinE) whilst small-scale DTs for local monitoring are in development for numerous applications such as hazard warning, agriculture and eco-hydrology. Currently these systems are being developed independently yet combining them offers opportunities for calibrating large-scale DTs and improving the resolution of large-scale DTs by replicating the dynamics of smaller systems using AI methods. In this contribution, we develop a new concept through which to link small- and large-scale DTs in order to automate an agile sensing system that can respond to natural environmental variability and directly measure changes of interest. Large-scale DTs are built primarily through Earth Observation (EO) data and describe regional to global scale changes in the Earth system whilst small-scale DTs simulate local variability using in situ sensors such as Terrestrial Laser Scanners (TLS). Linking the two means the large-scale DT can inform small-scale DTs by adapting their measurements (e.g. spatial and temporal resolution, focus area of interest, specific physical measurements) in response to regional changes in, for example, weather patterns. We focus on the following components: 1) using the small-scale DT to downscale the large-scale DT and ‘zoom’ into areas of interest; 2) using both the small- and large-scale DT to automatically detect changes in the environment and acquire new measurements without human intervention; and 3) using the small-scale DTs to calibrate large-scale DTs. With the increasing development of digital twin technology in the environmental sciences, our new concept will enable better integration of DTs and improve monitoring performance, which can improve decision-making. 

How to cite: Durrant, A., Harcourt, W. D., Höfle, B., Weiser, H., and Tabernig, R.: Linking small- and large-scale Digital Twins: A concept , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18187, https://doi.org/10.5194/egusphere-egu25-18187, 2025.

Paleoclimate information has played a key role in demonstrating how the Earth System responds to a variety of external forcings and how the earth’s climate is tightly related to atmospheric greenhouse gas concentrations. Although no strict analogue of possible future climate states exists, testing our understanding of the earth system, as embedded in earth system models, for conditions widely different from the historical period, is made possible by the existence of paleoclimate and paleoenvironmental reconstructions. Since its start in 1995, PMIP, the Paleoclimate Modelling Intercomparison Project (https://pmip.lsce.ipsl.fr/), has fostered and coordinated model-model and model-data comparisons for key periods: the mid-Holocene, ~6000 years ago, the Last Glacial Maximum (LGM), 21,000 years ago, the last two millennia, the last interglacial, the mid-Pliocene warm period (MPWP) were the key periods for PMIP4-CMIP6, with specific targets for each period. For instance, the enhanced monsoons and response of the northern high latitudes for the mid Holocene, the fate of Arctic sea ice and climate of the last interglacial, large spatial gradients and equilibrium climate sensitivity for the LGM and MPWP. In addition, each of these periods stood as reference for further PMIP experiments aimed to better understand the response of the climate system to external forcings.

For the next CMIP phase, PMIP continues to contribute studies on the responses to external forcings. This poster will present the targets for the FastTrack last interglacial experiment (abrupt-127k) as well as future opportunities related to other periods (e. g. Kageyama et al., 2024). We look forward to discuss with the CMIP and PMIP communities to plan further cross-cutting work and analyses.

Acknowledgements and cited reference.

We are acknowledging the help of the PMIP community in building PMIP over the years.

Kageyama M, et al., (2024) Lessons from paleoclimates for recent and future climate change: opportunities and insights. Front. Clim. 6:1511997. doi: 10.3389/fclim.2024.1511997

How to cite: Kageyama, M., Brierley, C., and Peterschmitt, J.-Y. and the the PMIP community: Earth system responses to external forcings : opportunities from paleoclimate studies and the Paleoclimate Modelling Intercomparison Project (PMIP) for CMIP, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18925, https://doi.org/10.5194/egusphere-egu25-18925, 2025.

We present the CMIP7 solar forcing dataset and its validation both for the effects of solar irradiance and particle forcing. In particular we present the results from first simulation runs that use the new CMIP7 as well as the previous CMIP6 dataset for the period 2002-2012, covering two solar maxima and a deep solar minimum. Specifically, we present simulation runs carried out with the chemistry-climate models WACCM, SOCOL, EMAC, ICON and KASIMA to determine the response to the solar SSI and particle forcing. The performance of the CMIP7 recommendations with respect to atmospheric radiative heating and composition will be evaluated both compared to the CMIP6 recommendations, and to satellite observations of atmospheric trace gases. The different responses and their implications will be discussed.

How to cite: Haberreiter, M. and the CMIP7 Solar Forcing Validation Team: CMIP7 solar forcing validation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20295, https://doi.org/10.5194/egusphere-egu25-20295, 2025.

The past topographic evolution of the Tibetan-Himalayan orogen holds the key to understanding interactions between Earth, Climate and Life processes since deep times. This has been hindered so far by the lack of accurate paleogeographic reconstructions of the orogen through time based on well-dated reliable proxies of past elevations. In the sedimentary archives of the basins formed in the orogen during the collision, recovered fossil content including pollen, fish and mammals yielded first order estimates on elevations based on environmental conditions of nearest living relatives while leaf physiognomies provided more direct constraints. Stable isotope composition from ancient meteoric waters preserved in pedogenic carbonates and biomarkers have been recovered and interpreted in terms of paleoelevations assuming past meteoric lapse rates. Outside of the basins in the high massifs, synkinematic hydrous silicates preserving ancient rainfalls have been used for paleoaltimetry purpose, notably in the Himalayas. Despite these significant efforts, the new paleoelevation datasets have led more to controversy than consensus. Fierce debates currently involve several international groups. Widely different topographic growth scenarios have been proposed with end-members ranging from a high Plateau prior to the onset of the India-Asia collision (“Proto-Tibetan Plateau”), to a much more recent - mostly Miocene - uplift and the preservation of broad low elevation valleys late until the Neogene.

As part of the starting TIBETOP project (funded by the french ANR) we propose here a state-of-the-art review of paleoelevation proxies across the Tibetan-Himalayan orogen, ranging from surface records in sedimentary basins to deeper crustal rocks now exhumed in the relief and mountain belts bordering these basins. We present a compilation and reappraisal of the existing regional paleoelevation data including revised provenance, stratigraphy dating, and stable isotope data in basin records as well as structural context, exhumation and fluid-rock deformation interactions at different interfaces of the continental crust. The TIBETOP project thus aims to produce a set of interactive paleogeographic reconstructions through time with associated datasets constraining the Himalayan-Tibetan orogen since the India-Asia collision. These will be improved through the project to include new data and updated paleogeographic reconstructions made available to modelers of climatic, biotic and surface processes to enable testing the above cited fundamental hypotheses on the role of mountain and plateau building on Earth System processes over geologic time.

How to cite: Dupont-Nivet, G., Fidalgo, J.-C., Moreau, K., Rivera, L., Feng, Z., Fang, X., Lavé, J., and Licht, A.: Tibetan Plateau paleogeographic reconstructions during the India-Asia collision: from paleoelevation proxies to geodynamic models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21621, https://doi.org/10.5194/egusphere-egu25-21621, 2025.

The integration of big data, cloud models, and extensive knowledge to drive new knowledge discovery through data is a new paradigm for research in the field of Earth sciences. Although the advancement of big data technologies and infrastructures has simplified data acquisition, deep-time geoscience still faces challenges such as fragmented data, difficulties in visualization, and insufficient computing power. To assist the broad community of geoscientists, we propose the "Deep Platform," a one-stop online research platform that utilizes cloud computing and advanced technologies. The platform provides open access to deep-time geoscientific data, knowledge, models, and computing power. It is designed to promote collaborative innovation and discovery among global geoscientists. The "Deep Platform" represents a significant advancement in geoscientific exploration, fostering global collaboration and advancing a data-driven research paradigm within the framework of open science.

How to cite: Hu, L.: DEEP Platform: Empowering Global Geoscientists  in Data-Driven Research Era, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21952, https://doi.org/10.5194/egusphere-egu25-21952, 2025.

Understanding microbial activity in snowpacks is essential for unveiling the dynamics of cold ecosystems, yet little is known about how this activity changes between day and night. To address this knowledge gap, we conducted a 24-hour study on a snowpack located in the French Alps, sampling snow at five-hour intervals across different layers —from the surface to the basal layer in contact with soil.

For each layer and time point, we sampled snow to assess microbial activity using omic techniques, like metagenomic and metatranscriptomics, coupled to the analysis of environmental parameters, including sunlight duration, snow pH and temperature. Our results revealed significant diurnal variations: sunlight, pH and temperature fluctuated throughout the 24-hour period, with microbial activity showing corresponding changes. For example, algae affiliated with Chlorella and Volvox, or fungi affiliated with Rhizophagus and Penicillium, showed different transcriptomic responses to diurnal changes in surface and basal samples. These findings highlight the influence of environmental factors on microbial processes in snow and provide the first insights into how microbial activity adapts to the diurnal cycle in snowpacks.

This study contributes to understanding microbial dynamics in snow-covered ecosystems, shedding light on the interplay between microorganisms and their environment over short temporal scales.

How to cite: Schivalocchi, F., Darfeuil, S., Crouzet, A., Martins, J., Tusha, D., Jaffrezo, J. L., Piot, C., and Larose, C.: Unraveling Microbial Activity in Alpine Snow using metatranscriptomics: A 24-Hour Study of Diurnal Variations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1190, https://doi.org/10.5194/egusphere-egu25-1190, 2025.

The metabolic potential and activity of deep-sea microbes have not been fully explored by metatranscriptomics using the samples obtained by different sampling methods. Here, we report active deep-sea microbes obtained by the methods of Multiple in situ Nucleic Acid Collection (MISNAC), in situ microbial filtration and fixation (ISMIFF), in situ microbial filtration without fixation (ISMIFU) and the Niskin bottle at 1,038-m depth in the South China Sea. Higher biodiversity and different dominant active microbial taxa in the metatranscriptomes were detected in the MISNAC and ISMIFF samples, compared with the other two approaches. The transcriptional profiles of 40 conserved genes were similar between the MISNAC and ISMIFF samples, while expression of a quarter of these genes was not detected in the ISMIFU sample. Genes related to the CO oxidation and nitrification processes were highly transcribed in the MISNAC and ISMIFF transcriptomes, whereas genes for chemotaxis and low-oxygen adaptation were highly transcribed in the Niskin samples. Overall, our result highlights the importance of in situ sampling and preservation for more precise quantification of the ecological function of active deep-sea microbiomes.

How to cite: Wang, Y. and He, Y.: Transcriptional difference of deep-sea microorganisms under different sampling methods, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1494, https://doi.org/10.5194/egusphere-egu25-1494, 2025.

The use of glacial rock flour as an agricultural soil amendment gained increasing interest due to its fine particle size, potential to deliver crop-essential nutrients, and capacity for Mg or Ca-rich silicates to enhance rock weathering to capture atmospheric CO₂ in carbonate form. Additionally, some studies have observed that rock flour amendments can reduce N₂O flux. Plant growth, carbon storage, and soil emissions are all influenced by microorganisms in soil, which actively participate in biotic weathering processes that release plant-essential nutrients like phosphate, potassium, and sulfur from minerals. Furthermore, microorganisms drive nitrogen and carbon cycling, which influences soil fertility and greenhouse gas emissions. Thus, the unknown impact of glacial flour application to soil microbial communities must be investigated. Our study aimed to assess a French agricultural soil microbial community responds to varying glacial rock flour application rates during a 12-week microcosm experiment. The granitic glacial flour selected for study originated from Mer de Glace (French Alps). To understand the microbial community’s response, we focused on taxonomic shifts, relative abundance of genes related to nitrogen cycling and nutrient access, and geochemical shifts between baseline and 12-week samples. We hypothesized that glacial flour amendment would select for a community that would reduce nitrogen losses through N₂O and demonstrate improved ability to access flour-bound nutrients particularly at higher application rates compared to the control soil. To test this hypothesis, we conducted a 12-week microcosm study where glacial rock flour was added to 50 g of agricultural soil at rates of 0, 0.5, 2, 5, 10, 20, 30, 50, 80, 115, and 157 t ha^-1. Each treatment had 12 replicates and one replicate per treatment was destructively sampled each week for analysis. For the higher application rates (30-157 t ha^-1), a quartz powder control was included to account for potential changes in soil structure. Replicates remaining in the experiment were watered once per week up to 80% of water holding capacity to simulate agricultural irrigation or rainfall events. DNA was extracted from all samples for downstream analyses and subsamples from baseline and endpoint were retained for geochemical analyses. Quantitative polymerase chain reaction (qPCR) was performed on the 16S and 18S genes to quantify bacterial and fungal abundance, respectively. Metabarcoding of the v3-v4 region of the 16S rRNA gene (rrs) was done to track taxonomic changes in the bacterial population over the course of the experiment. Inorganic nitrogen species were quantified in the baseline and 12-week samples. Preliminary results showed that bacterial communities exhibited differential growth in response to amendments above 5-10 t ha^-1 compared to those below, with shifts occurring at week 4 and week 10 of the experiment. Glacial flour application of 30 and 50 t ha^-1 resulted in the lowest percent loss of inorganic nitrogen from baseline to week 12 compared to other application rates. These initial findings indicate that glacial rock flour application rates may significantly influence the soil microbial community, with important implications for nitrogen cycling and nutrient accessibility.

How to cite: Sampsell, K., Wild, B., Vogel, T., and Larose, C.: Soil microbial community response to glacial rock flour amendment: insights from a microcosm experiment, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3250, https://doi.org/10.5194/egusphere-egu25-3250, 2025.

Microbiological activity on glacier and ice sheet surfaces can be a major factor responsible for their darkening. Among microbes, pigmented snow- and glacial ice-algae increase light absorption, further accelerating melting and supporting the development of pigmented algal blooms on the Greenland Ice Sheet (GrIS). The relationship between carbon-fixing algae and carbon-respiring heterotrophic microorganisms influences the amount and composition of organic matter (OM). Yet, the dynamics of the OM derived from these microbes on the GrIS remain unclear. To address this gap, we incubated algae-dominated snow and ice surface samples in situ in vented bottles under light and dark conditions. We evaluated the initial microbial community composition (via 16S and 18S rRNA gene sequencing) and characterized the changes in both dissolved and particulate OM (DOM and POM) via ultrahigh-resolution Fourier transform ion cyclotron resonance mass spectrometry. We show that glacier ice-algae habitats dominated by Ancylonema, have higher abundance of highly unsaturated and aromatic compounds resistant to bio- and photo-degradation. In contrast, snow-algae habitats dominated by Chloromonas, are enriched in bioavailable and more photosensitive unsaturated aliphatics and sulfur- and phosphorus-containing compounds. Light exposure increased water-soluble DOM compounds derived from POM, which accounted for large proportion of the initial DOM composition of both algae dominated habitats. Of these initial DOM pools, up to 50% were heterotrophically degraded in the dark, while light alone photodegraded less than 20%. The significant accumulation of light-absorbing aromatics from both POM and DOM pools at the end of the ice-algae experiments, emphasize ice-algae larger effect on altering glacier color compared to snow-algae, and thus on decreasing glacier albedo and accelerating melting.

How to cite: Rossel, P. E., Antony, R., Mourot, R., Dittmar, T., Anesio, A. M., Tranter, M., and Benning, L. G.: Organic matter variability in algal dominated habitats on the Western Greenland Ice Sheet , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4787, https://doi.org/10.5194/egusphere-egu25-4787, 2025.

How to cite: Álvarez-Gandía, Y. D., Lewis, C., Barker, B. M., Catalán-Dibene, J., Kaufeld, K. A.  ., Kollath, D., Lauer, A., Mead, H., Oltean, H., Ramsey, M., Romero-Olivares, A., Bartlow, A. W., and Gorris, M. E.: Understanding soil properties conducive for Coccidioides ssp. presence in the United States , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7658, https://doi.org/10.5194/egusphere-egu25-7658, 2025.

Groundwater health is increasingly threatened by climate change, which alters precipitation patterns, leading to groundwater recharge shifts. These shifts impact subsurface microbial communities, crucial for maintaining ecosystem functions. In this decade-long study of carbonate aquifers, we analyzed 815 bacterial 16S rRNA gene datasets, 226 dissolved organic matter (DOM) profiles, 387 metabolomic datasets, and 174 seepage microbiome sequences. Our findings reveal distinct short- and long-term temporal patterns of groundwater microbiomes driven by environmental fluctuations. Microbiomes of hydrologically connected aquifers exhibit lower temporal stability due to stochastic processes and greater susceptibility to surface disturbances, yet they demonstrate remarkable resilience. Conversely, isolated aquifer microbiomes show resistance to short-term changes, governed by deterministic processes, but exhibit reduced stability under prolonged stress. Variability in seepage-associated microorganisms, DOM, and metabolic diversity further drive microbiome dynamics. While shifts in DOM influence the potential functions of the microbiome, its overall functional potential demonstrates high temporal stability and resilience over time, largely due to functional redundancy. These findings highlight the dual vulnerability of groundwater systems to acute and chronic pressures, emphasizing the critical need for sustainable management strategies to mitigate the impacts of hydroclimatic extremes.

How to cite: Wang, H., Herrmann, M., Schroeter, S. A., Zerfaß, C., Lehmann, R., Lehmann, K., Ivanova, A., Pohnert, G., Gleixner, G., Trumbore, S. E., Totsche, K. U., and Küsel, K.: Balancing Act: Groundwater microbiomes' resilience and vulnerability to hydroclimatic extremes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8846, https://doi.org/10.5194/egusphere-egu25-8846, 2025.

The microbiologically-driven darkening of bare ice surfaces on the western Greenland Ice Sheet is significantly enhancing melting, contributing to sea level rise. Among microorganisms, purple-brown pigmented glacial ice algae (mainly members of Ancylonema alaskanum and Ancylonema nordenskiöldi) are key contributors to the ice surface darkening and the associated surface albedo reduction. It is known that the glacial ice algae actively replicate and spread across vast ice surface areas during the summer melt season. However, the metabolic pathways driving the glacial ice algal bloom development are still poorly understood. To address this knowledge gap, we used an untargeted endometabolomics approach to explore the dynamics and metabolic potential of glacier ice algal blooms and the role of the environment on their metabolic responses. We analyzed glacial ice algae-dominated surface ice samples from various locations across the western Greenland Ice Sheet using high-resolution mass spectrometry to annotate the metabolome of the algae-dominated samples. Combined with physical and chemical environmental data describing their constantly changing habitat (e.g., temperature, light response, cell numbers) we derived novel insights into the metabolic activity of the glacial ice algae and their biochemical adaptations to glacier conditions. Our data contribute to improving our understanding of the link between ice darkening and microbial activity.

How to cite: Majumder, A., Jaeger, C., Lisec, J., E Rossel, P., Tranter, M., M Anesio, A., and G Benning, L.: Metabolic profiles of glacial ice algal-dominated habitats across the western Greenland ice sheet., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12311, https://doi.org/10.5194/egusphere-egu25-12311, 2025.

Wastewater-based epidemiology (WBE) is increasingly recognized as a pivotal tool for tracking community health trends, including viral pathogens and antimicrobial resistance (AMR). This study integrates the surveillance of SARS-CoV-2 genes and AMR in wastewater samples collected from wastewater treatment plants (WWTPs) in Dehradun, India, during the COVID-19 pandemic. By combining genomic and molecular analyses, this research offers a dual perspective on two significant public health threats: the emergence of SARS-CoV-2 variants and the escalating burden of AMR. Weekly wastewater samples were collected from eight WWTPs between June 2022 and July 2023. SARS-CoV-2 RNA was quantified using real-time PCR assays targeting N, S, and ORF-1ab genes. At the same time, AMR was assessed through 16S rRNA gene sequencing and qPCR to detect resistance genes across multiple antibiotic classes, including aminoglycosides, β-lactams, macrolides, and tetracyclines. Seasonal variations, gene abundance, and correlations between SARS-CoV-2 and AMR markers were analyzed to understand the dynamics of these health risks in the urban environment. In this respect, SARS-CoV-2 analysis revealed 68 distinct lineages, dominated by Omicron recombinant variants XAP, XBB.1.16.1, and XBB.1.22 in March and April 2023, making up more than 50% of the total abundance. Such variants carried mutations that could increase transmissibility, underlying the importance of wastewater monitoring in tracking viral evolution. Meanwhile, AMR surveillance highlighted significant seasonal trends in the abundance of antibiotic-resistance genes (ARGs). Tetracycline resistance surged to 34.35% during the monsoon season at the Kargi WWTP, compared to 12.98% in winter. In contrast, macrolide resistance peaked at 35.87% in winter and decreased to 15.35% during the monsoon season. Resistant genes, such as tetX, ermF, blaOXA-50, and aadA1, were frequently detected, with aminoglycosides and tetracyclines consistently showing high resistance levels across sites and seasons. The simultaneous presence of SARS-CoV-2 RNA and ARGs in wastewater underscores the role of WWTPs as reservoirs and conduits for emerging public health threats. Climatic factors, anthropogenic activities, and proximity to healthcare facilities impact the distribution of resistant genes and viral variants.  Notably, the effective removal of SARS-CoV-2 genes in municipal WWTPs (~50% gene reduction) highlights the possibility of targeted interventions to mitigate pathogen spread. However, the continued presence of resistant genes despite treatment raises concerns about environmental and public health risks. This study illustrated the potential for integrated viral and AMR wastewater surveillance to deliver community health intelligence in real-time. Thus, by monitoring SARS-CoV-2 variants alongside AMR trends, WBE can be an early warning system for emerging health threats, informing public health policy and environmental management strategies.

Keywords: antimicrobial-resistance; Covid-19; resistant genes; wastewater-based epidemiology

How to cite: Dogra, S. and Kumar, M.: The co-occurrence of Viral Pathogens and Antimicrobial-Resistance (AMR) markers from urban wastewater treatment plants in India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13669, https://doi.org/10.5194/egusphere-egu25-13669, 2025.

Learning about the metabolic activities and adaptations of deep-sea microbes is a challenging task, because the collection and retrieval of samples from the deep ocean induce RNA degradation and alteration of microbial communities. Here, we employed a in situ DNA/RNA co-extraction device to collect 18 time-course nucleotide acid samples for winter and summer seasons in the South China Sea to generate metatranscriptomes and metagenomes with the minimal possible sampling perturbation. Between the two seasons, the most active eukaryotic microbes were Ciliophora, whereas the most abundant but inactive eukaryotic microbes were Retaria. In the winter, autotrophic microorganisms contributed to organic matter production by CO₂ fixation associated with nitrification. In the summer, the primary source of energy originated from heterotrophic microorganisms that can utilize alkanes, aromatic compounds and carbohydrates, partially relying on anaerobic respiration in the particles. This may relate with nutrient source variations as reflected by the different levels of microbial network complexity between two seasons. Altogether, we uncovered the metabolic activities and adaptations of active microbial groups in two seasons with in situ metatranscriptomes, paving the way to identification of the real microbial contributors to element cycles in the deep ocean.

How to cite: He, Y., Baltar, F., and Wang, Y.: In situ sampling uncovers seasonal variability in community structure and metabolism of active deep-sea microbes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14323, https://doi.org/10.5194/egusphere-egu25-14323, 2025.

Abstract: Soil contamination from excessive pesticide use is a global issue, threatening human health and environmental sustainability. Pesticides disrupt soil microbiomes, leading to a decline in beneficial microorganisms, impaired nutrient cycling, and long-term ecosystem disturbances. Microorganisms play a crucial role in environmental preservation by breaking down xenobiotics, including pesticides. However, the pathways of pesticide degradation by microorganisms are not well understood due to limitations in current culturing techniques. To address this knowledge gap, we utilized 16S rRNA V3-V4 metagenomic sequencing to analyze farming soils in Dehradun with a history of pesticide application. Our results revealed a relative abundance of the phyla Proteobacteria, Acidobacteria, Firmicutes, and Actinobacteria in contaminated zones. Bacillus, Solibacter, and Nitrospira were the most prevalent taxa, indicating nitrogen and carbon fixation and regulation of biogeochemical cycles in extreme environments. Predictive metagenome analysis showed that core-degrading orthologs involved in membrane transport, the TCA cycle, carbohydrate metabolism, and xenobiotic degradation (such as atrazine and chlorocyclohexane degradation) were prevalent in contaminated soils. Our findings highlight the implications of abundant microbes in contaminated soils through comprehensive metagenomic approaches, paving the way for further research on gene expression frequencies and major enzyme assays for pesticide degradation.

Keywords: Metagenomics, Microbial diversity, Pesticide degradation, Taxonomy

How to cite: Mahapatra, D. M., Sinha Roy, S., and Goyal, N.: Decoding Soil Microbiomes: Metagenomic Insights into Pesticide-Contaminated Agro-Ecosystems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14807, https://doi.org/10.5194/egusphere-egu25-14807, 2025.

Glacier ice algae of the streptophyte genus Ancylonema live on glaciers globally, including the Greenland Ice Sheet, and bloom despite low temperatures, low nutrient availability, and very high light intensities. In polar regions, the long polar night also imposes additional abiotic stressors. However, the cellular mechanisms responsible for Ancylonema’s resistance and adaptation to high light stress or to prolonged darkness during the polar winter are not known. We addressed this knowledge gap by evaluating the functional responses of a Greenland Ice Sheet Ancylonema-dominated microbiome to in-situ light conditions and continual darkness during a 12-day period using amplicon sequencing, metatranscriptomics, and metaproteomics. The microbial community did not substantially change during the 12 days of dark incubation; however, heterotrophs became more transcriptionally active in the dark. Metatranscriptomic and metaproteomic analyses showed that Ancylonema cells underwent high oxidative stress in the light. However, after 12 days in darkness, the algal cells retained functional photosynthetic machinery but downregulated their expression of early shikimate pathway enzyme transcripts. Transcriptional reprogramming linked to sugar uptake and phytohormone signalling was also identified in the dark, providing an insight into the first steps towards algal cell survival through the polar night. These results give us a novel understanding of the gene expression dynamics of glacier ice algae under changing light conditions, providing important clues regarding their adaptation to a harsh and extremely variable environment.

How to cite: Feord, H. K., Keuschnig, C., Trivedi, C. B., Mourot, R., Zervas, A., Turpin-Jelfs, T., Tranter, M., Anesio, A. M., Adrian, L., and Benning, L. G.: Survival strategies of supraglacial algae-dominated communities in the transition from high light to continual darkness on the Greenland Ice Sheet, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16283, https://doi.org/10.5194/egusphere-egu25-16283, 2025.

Deep Underground laboratories are unique environments for exploring the impact of ultralow radioactivity on living organisms. They also provide unique features for running long term controlled low-dose experiments.

Evolution is an on-going process, and it can be studied experimentally in organisms with rapid generations. The E. coli Long-Term Evolution Experiment (LTEE) is an ongoing study in experimental evolution begun by Richard Lenski at the University of California, which has been tracking genetic changes in 12 initially identical populations of asexual Escherichia coli bacteria since 24 February 1988 on more than 60.000 generations.

A first evolution experiment conducted at Modane Underground Laboratory with the same E. Coli strain and the same growth medium used by Richard Lensky and collaborators has shown no change in the fitness trajectory over 500 generations when radiative background was reduced by a factor 6 from 150 to 26 nGy/hr. Monte-Carlo simulation of the experimental set-up showed that 40K in the E. Coli culture medium (Davis Medium) was the almost exclusive source of radioactivity to the bacterial strains, representing 99% of the dose received.

Potassium has three naturally occurring isotopes: 39K (93.258%) and 41K (6.730%) are stable, while 40K (0.012%) is radioactive, with a half-life of 1.25 billion years. As 40K in the nutritive medium was the main obstacle to the reduction of the dose received by the bacterial strains during this experiment, depleting 40K in the potassium used to feed the bacteria would reduce significantly the dose received and allow exploring further the ultralow radioactivity frontier. Reciprocally, enriching the potassium in 40K would increase the dose absorbed by the bacteria without changing any other physico-chemical parameters.

We therefore propose to compare the fitness trajectories over 1000 generations of the same E. Coli strain using Davis Medium (DM) nutritive media either enriched or depleted in 40K. Although the isotopic composition of natural Potassium is very stable, potassium enriched in 39K and depleted 10 times in 40K can be purchased from commercial vendors for less than 10 € per milligram.  To enrich natural potassium in 40K, a promising approach is through neutron irradiation.

Repeating the same experiment using DM nutritive media that differ only by the isotopic composition of the potassium allows isolating the sole impact of radiation on the evolutionary path of the bacteria. Increasing 40K isotopic fraction increases proportionally the absorbed dose and radiation induced mutations are expected to modify the strain evolutionary paths when they exceed the spontaneous mutation rate.

These experiments could be performed in several Deep Underground Laboratories to compare the observed fitness trajectories and quantify the reproducibility of the observed evolutionary paths.  

How to cite: Breton, V., Fois, G., Insa, C., Maigne, L., and Biron, D.: Deep Underground Long Term Evolution Experiments, a key to understand the impact of low doses on living organisms, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18428, https://doi.org/10.5194/egusphere-egu25-18428, 2025.

The ablation area of the Greenland Ice Sheet (GrIS) is a biome driven by microbial activity. During the summer melt season, the weathering crust of the ice becomes a wet living skin dominated by eukaryotic glacier ice algae, particularly Ancylonema spp., which accelerate ice melt through their dark pigmentation. Cryoconite holes, formed by sediment melting into the weathering crust, also dominate the landscape of the ice surface. They are primarily inhabited by cyanobacteria as the main primary producers and also host a diverse community of bacterial, fungal and other microeukaryotic heterotrophs. This study investigates the active microbial communities and functionality of the weathering crust and cryoconites using Total RNA metatranscriptomics. With this approach, we describe the full diversity of ice surface microbial communities; assembling, annotating and analyzing jointly full-length 16S rRNA and 18S rRNA genes in addition to transcriptomes. We conducted a seasonal study over a 21-day period during the ablation season, sampling ice and cryoconite habitats. Samples were collected from five cryoconite holes and five 2-meter patches of the weathering crust ca 25km inland on the GrIS, near Ilulissat. Biomass from cryoconite holes and ice surfaces was collected at solar noon on seven sampling days during the summer. The findings highlight the dynamics and spatial variability of very different microbial communities between the weathering crust and cryoconite holes. Notably, the weathering crust is dominated by eukaryotic biomass, and spatial variability is significant; cryoconites are far more diverse, dominated by prokaryotic interactions and relatively stable temporally.  A snowfall in late summer provided a window of opportunity to show that cryoconites communities are robust, while the functionality of the weathering crust, including genes associated to carbon, nitrogen and phosphorus cycling all responded to snowfall. The Total RNA approach in this study provides a powerful insight into the entire active microbial community and their functionality on glacial surfaces.

How to cite: Zervas, A., Perini, L., Feord, H., Jaarsma, A., Sipes, K., Tranter, M., Benning, L. G., and Anesio, A. M.: TotalRNA sequencing reveals active community and functional dynamics on the surface of the Greenland Ice Sheet, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18535, https://doi.org/10.5194/egusphere-egu25-18535, 2025.

The cryosphere encompasses a wide range of perennial microbial habitats including glaciers, lakes, seas, rivers, and soils. These habitats pose intrinsic seasonally-variable challenges for microbial populations, whose activity may be temporarily constrained. Microbial dynamics in these environments during the summer and (to a lesser extent) spring have been extensively studied, however the winter season remains largely unexplored. During the winter period, microbial activity may be constrained by freezing temperatures, limited availability of liquid water, and the absence of light and thus photosynthetic carbon input. Critical aspects of microbial activity, including metabolic processes, winter-specific community profiles, and their unique functional roles in ecological processes, are still poorly understood. To address this knowledge gap, we examined microbial activity during the winter months in a range of aquatic and terrestrial Arctic habitats. Using metatranscriptomic techniques, we identified active microorganisms and uncovered their core metabolic requirements for sustaining activity. We also employed BONCAT (bioorthogonal noncanonical amino acid tagging) to assess the ratio of live to dead microbes across different habitats and utilized qPCR and RT-qPCR to quantify organism abundance. Our results revealed significant differences in community composition, abundance, and activity across environments. Notably, glacial snow and lake slush snow exhibited high RNA-to-DNA ratios, with distinct differences in microbial diversity. Lake slush snow, in particular, displayed a more uneven microbial community compared to its snow counterpart. In contrast, soil showed very low activity despite a high DNA content. Among the ice cores, glacial ice exhibited both high diversity and moderate microbial activity. Overall, our findings suggest that microbial communities in winter are active, with activity levels varying across different habitats. These variations may be driven by factors such as differences in microbial seeding sources and the availability of free water. Despite limited energy reserves, we suggest that winter microbial communities contribute to the mineralization and recycling of biomass and elements, playing a crucial role in sustaining ecological processes in the Arctic cryosphere.

How to cite: Singh, H., Bradley, J. A., Vogel, T. M., and Larose, C.: Exploring microbial activity and metabolic requirement during the polar winter of the Arctic, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20154, https://doi.org/10.5194/egusphere-egu25-20154, 2025.

How to cite: Rai, K. and Singh, G.: Advanced Backscatter Modeling for Enhanced Detection of Forest Disturbances in the Indian Subcontinent, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-188, https://doi.org/10.5194/egusphere-egu25-188, 2025.

Forest structure is shaped by forest management practices, land-use changes and forest disturbances including droughts, fires, storms and insect outbreak. It plays an important role in climate by modifying the carbon-water-energy exchanges with the atmosphere, and affects the capability of forests to undergo future disturbances in a changing climate. Given the importance of forest structure for the climate, land surface models are moving towards explicit representations of forest structure and management strategies. We present a new procedure to initialize forest diameters over Europe and document its implications for simulations of future forest carbon sinks. The simulated diameters for each grid cell covered by forests are initialized toward the diameter from a forest inventory. To this end, a 300-years semi-analytical spinup was carried out to bring the soil carbon and nitrogen pools into equilibrium until the European forests were clearcut. Then, a 150-years biosphere simulation over Europe was performed to build a look-up-table of simulated diameters. For each grid point, the year associated with the simulated diameter that is the closest to the observation is selected, enabling the production of new initial state files over Europe. The new initialization procedure makes the initial state of forest more realistic and therefore is expected to have significant influence on the evolution of the forest carbon sink. In this work, we will assess the effect of the initialization procedure on the simulated land carbon sink and we will evaluate the representation of the diameters in the ORCHDEE LSM. The method could be further extended to initialize other forest state variables such as height or aboveground biomass.

How to cite: Remaud, M., Jeong, J., Marie, G., Flores, O., Naudts, K., and Luyssaert, S.: On the assessment and initialization of European forest state variables in Land Surface Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2268, https://doi.org/10.5194/egusphere-egu25-2268, 2025.

Knowledge of forest ecosystem pattern and process responses to climate change and anthropogenic pressure requires innovative tools that combine monitoring and modelling of tree growth dynamics to account for a more sustainable management of forest resources and ecosystem services. In this context, Digital Twins (DTs) emerge as powerful tool to allow a better interpretation of complex models, summarizing a large amount of data and knowledge into a comprehensive 3D visualization. A Digital Twin is an evolving and comprehensive representation of a physical object, in our case trees, which involves three key elements: a digital representation of the object, an evolving set of data and a dynamic adjustment of the object data. However, due to the structural complexity of the forest stand, and the lack of adequate growth historical data series useful to build and validate the simulations, the full potential of Digital Twin frameworks has yet to be realized in forest field. Our work aims to develop a system that simulate forest growth and spatial patterns through a process-based single tree model and represent the outputs into a 3D immersive and interactive environment, able to reproduce the stand structure of Mediterranean forests. An individual based spatially explicit model has been developed to simulate the biomass growth within a time step of one year and while an immersive 3D dynamic environment enables the user to interact with trees (e.g. tree marking, logging). Competition among trees has been modelled computing the tree influence on surroundings space using a distance-biomass dependent approach. We implemented a set of allometric equations to convert tree biomass into size attributes (e.g. stem diameter, total height) to appropriately represent the modelled forest stand in the 3D environment. The use of DTs can assist forest experts and policymakers in managing complex systems like Mediterranean forests, by simulating several management scenarios and analysing their long-term impacts on forest ecosystem dynamics. Furthermore, process-based models coupled with an immersive 3D representation could help to better understand the forest ecosystem functioning.

How to cite: Fornaro, R., Giannino, F., Heatfield, D. N., Minopoli, V., Aquino, A., Rita, A., Saracino, A., and Saulino, L.: Forest digital twin: coupling field data, mathematical modelling and 3D representation of a Mediterranean forests  , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6985, https://doi.org/10.5194/egusphere-egu25-6985, 2025.

The continued introduction of non-native insect species, coupled with the rising threat of extreme wildfire events, poses significant risks to terrestrial ecosystems and the services they offer globally. However, the impact of invasive phloem-feeding insect species on fire severity is not well understood, particularly in terms of how they influence fire behaviour and the likelihood of crown fire ignition. Two experimental designs were set up to investigate how the alien tortoise scale (Toumeyella parvicornis) outbreaks have influenced fire behaviour dynamics and canopy surface reflectance in the Mediterranean P. pinea stands severely burnt in the summer of 2017. We combined Rothermel’s model for fire surface spread and Van Wagner’s crown ignition model to simulate fire behaviour and employed data from the Landsat 8 collection to detect canopy wilt symptoms related to T. parvicornis outbreaks. Simulating fire behaviour in single-storied P. pinea stands indicated that all predicted fires were surface fires. An uncertainty analysis concerning the inputs of the canopy fuel attributes model revealed that fires in thinned stands were entirely classified as surface fires. In contrast, in unthinned stands, only 62.7% were surface fires, with 37.3% categorised as conditional fire types. Among the Landsat 8 reflectance bands, only NIR, Green, and SWIR 2 were sensitive to the abundance of T. parvicornis. Based on these sensitive bands, two-band NIR-multiplied vegetation indexes were significantly associated with the abundance of T. parvicornis from the fall generation onward, when sooty mould consistently covered canopy needles. The divergence between observed and predicted fire behaviour underscores the need to investigate the processes and variables linked to T. parvicornis feeding activity on the trees to improve fire behaviour prediction. Understanding how insect outbreaks can modify fire behaviour in Mediterranean stands is crucial for effective management at stand and landscape levels. The satellite vegetation indexes based on sensitive reflectance bands represent an essential tool for an early recognition of insect outbreak distribution on large spatial scale.

How to cite: Saulino, L., Garonna, A. P., Rego, F. C., Rita, A., Aquino, A., Liuzzi, G., Fornaro, R., Pinelli, E., Silvestro, R., Rossi, S., and Saracino, A.: Outbreaks of invasive phloem feeding Toumeyella parvicornis modified fire behaviour and canopy surface reflectance in Mediterranean Pinus pinea forests, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7375, https://doi.org/10.5194/egusphere-egu25-7375, 2025.