Data are a fundamental building block of science. Ever-increasing volumes and diversity of data are allowing us to solve ever more complicated research questions; yet they are also creating new challenges around efficient data management and storage. This talk focuses on geochemical data, that are relatively low in volume compared to other Earth System Science disciplines, but are highly diverse due to the large range of materials analysed and analytical techniques employed. Modern geochemical research increasingly draws on large compilations of data previously collected by multiple authors using multiple analytical methods, over years and decades. Harmonising data from such diverse sources, and ensuring consistency and comparable data quality, is a non-trivial task that requires significant investment of time and resources. As a consequence, data compilations are increasingly published in high-ranking journals. Yet often they are singular, one-time efforts for specific projects by individual authors that quickly become outdated and lose relevance. In contrast, curated synthesis databases, such as the GEOROC database for igneous geochemical rock and mineral compositions, are continuously being updated and can offer long-term consistent curation over decades. By providing free access to, and customisable search of, their comprehensive data and metadata collections, they enable the compilation of a diverse range of smaller, targeted datasets that can form the basis of many different research projects across multiple (sub)disciplines. Long-term synthesis databases are an invaluable resource for the geochemical and broader scientific community. However, despite their broad relevance and usage, many such community databases struggle to secure the required resources for database maintenance and continuous technical developments to cater to changing scientific demands. This burden can be partly alleviated through integration of databases with curated, domain data repositories. Data harmonisation is greatly aided by adherence to best practices and standards during data publication. Repositories that publish curated, discipline-specific datasets, therefore, play an important role in ensuring new analyses are sufficiently well documented to allow quality assessment and reuse by third parties. They also support data rescue and the alignment of legacy data with modern data requirements. These standards and best practices should in turn be developed based on community expertise and consensus, which requires international collaboration. In geochemistry, data providers and services from three different continents formed the OneGeochemistry initiative. OneGeochemistry promotes exchange and agreement on minimum common variables between researchers from all geochemical sub-disciplines and the more than 15 international societies, associations and science unions that govern different types of geochemical data. As a participant in the WorldFAIR project, OneGeochemistry aims to reconcile cross-domain solutions for data interoperability with domain-specific geochemical requirements. The implementation of geochemical data standards in repositories, and their broad adoption by the geochemical community, will enhance the value of data and services provided by synthesis databases, which will lead to better access to comprehensive data compilations and, ultimately, better science.

How to cite: Klöcking, M.: The importance of curated domain repositories and synthesis databases as evolving community resources for modern Earth System Science research, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9936, https://doi.org/10.5194/egusphere-egu25-9936, 2025.

Earth System Science datasets have been acquired for centuries across five broad spheres: geosphere, cryosphere, hydrosphere, biosphere and atmosphere. They vary from human observations to sensor-derived measurements ranging from nanoscale laboratory data to large-volume petascale datasets collected remotely by satellites, drones, etc. Across all spheres most datasets have their roots in three core disciplines: Geology, Geophysics and Geochemistry. Today we are generating unprecedented volumes of data and when combined with computer capacity, now at exascale, our capability to integrate and analyse data should be unparalleled.

Digital data repositories emerged around 1980 and the internet soon after. Initially data was shared by shipping on hard media. The internet soon enabled globally data sharing data, including by web services (e.g., OneGeology In 2008). Multiple global data sharing networks were envisioned, but few moved beyond those that proposed them. Machine-to-machine data sharing is still a challenge. Many spheres cannot utilise the existing capacity of computers, including the full potential of AI applications, because these cannot read the volumes of available data. 

History has repeatedly shown that revolutionary infrastructures can take decades to realise their full potential and change from being a new way of doing things to multiple ways of doing new things. 

The FAIR principles were specifically designed to increase machine-to-machine interoperability of data: they are the blueprint of WHAT needs to be done but the HOW will involve rethinking 3 key steps. 

Firstly, shift the onus on aggregating data from the consumer to repositories capable of implementing discipline-centric FAIR (meta)data standards. 

Secondly, as recommended by the WorldFAIR Second Policy Brief to the European Open Science Cloud (EOSC), change from a bibliographic approach to data stewardship to one of data engineering, where richer and more comprehensive standardised (meta)data at the datum level enables machine-to-machine access of specific variables of interest across multiple disciplinary datasets. Take a more holistic approach to standards development (e.g., Observation, Measurement and Samples Standard (ISO 19156:2023)) and identify common universals across disciplines (e.g., time, place, units of measure). Initiatives like OneGeology and Geochemistry and hopefully soon OneGeophysics can support higher--level discipline centric (meta)data standards. Standards coordination groups (e.g., CODATA, Research Data Alliance) are critical. PIDS at the object level will be essential.

Thirdly, prioritise which datasets are made fully FAIR compliant and fund their curation in repositories that offer discipline based curation. The 2019 Beijing Declaration on Research Data notes that ‘publicly funded research data should be interoperable, and preferably without further manipulation or conversion, to facilitate their broad reuse in scientific research’. The myriad of data products generated from these primary data sources can go to generalist and institutional repositories.

Revolutionary infrastructures do take time to realise their full potential. It is nearly 25 years since the early experiments using the internet to globally network data repositories. The WorldFAIR Second EOSC Policy Brief emphasises that the change to machine-actionable FAIR data ‘is one of a magnitude which will necessitate considerable resourcing, investment, and upskilling; but it will also achieve significant benefits, including creating a digitally integrated Earth to support sustainable development of our planet.

How to cite: Wyborn, L.: Rethinking HOW We Create Global Networks of Earth and Environmental Datasets to Maximise Their Potential to Underpin Integrated Research for a Sustainable Planet., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16237, https://doi.org/10.5194/egusphere-egu25-16237, 2025.

Nuclear Magnetic Resonance (NMR) is a crucial logging technique for the unconventional and complex reservoir evaluation. However, the echo spacing is always an issue of borehole NMR measurement, which limits the performance of NMR tools to acquire the short relaxation components.

In this abstract, we proposed a novel Q-Switch technique aiming at breaking through the limitation of dead-time of borehole NMR logging tool, and to achieve much shorter echo spacing. Instead of using resistors of larger resistance in parallel with the radio-frequency (RF) coil to reduce the active dead-time, an inductive coupling circuit was introduced to decrease the ringing-down time significantly after transmitting the RF pulses with high voltage. The Q-Switch circuit consists of inductive coupling coil, capacitors, resistors and active high-voltage MOSFETs. The ringing-down time of RF system was decreased by at least 10 times compared to the system without using proposed Q-switch scheme, leading to echo spacing lower to 0.3 ms under the condition with resonant frequency lower to 500 kHz.

Both simulations and experiments were in great agreements, validating the feasibility and efficiency of proposed Q-switch scheme, and proved to be promising in the borehole NMR applications.

How to cite: Luo, S., Li, X., Yu, H., Wang, Z., Xing, T., Long, Z., Che, C., Liao, G., and Xiao, L.: A Novel Q-Switch Technique for  Borehole NMR Measurement, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-121, https://doi.org/10.5194/egusphere-egu25-121, 2025.

Soil and groundwater contamination at some sites impacts downstream populations when contaminants migrate from groundwater to rivers. Predictive modeling is challenging since it is required to include detailed subsurface structure and groundwater flow models within the site, as well as watershed-scale models for large-scale transport. Now that climate change impacts are major concerns at many sites, it is important to have the capability to represent the water balance change and its impact on contaminant transport both at the site and watershed scale in a consistent manner. This study introduces a new simulation framework to couple a detailed 2D site/hillslope-scale groundwater model to the 3D watershed-scale model to describe contaminant transport from groundwater to river water within the catchment. Within the site, we estimate the contaminant discharges to the river from contaminant sources based on the Richards equation and advection-dispersion equation. The discharges are then applied as the boundary conditions to the watershed-scale model considering the width of the 2D site/hillslope-scale groundwater model and recharge rates for both models.

We demonstrate and validate our framework based on the tritium concentration datasets in surface water and groundwater collected at the Savannah River Site F-Area. Results show that the method can successfully reproduce the contaminant concentration time series in river water.

How to cite: Sakuma, K., Wainwright, H., Xu, Z., Lawrence, A., and Hazenberg, P.: Multiscale model coupling for watershed-scale contaminant transport modeling from point sources in Savannah River Site, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4077, https://doi.org/10.5194/egusphere-egu25-4077, 2025.

As one of crucial factor in atmospheric particulate matter, elemental components exhibit distinct distribution features within different particle size ranges. Crustal elements (such as Al, Si, Fe, Ca, Mg) are primarily concentrated in coarse particulate matter, whereas elements originating from anthropogenic pollution sources (such as heavy metal elements including Pb, Zn, Cd, As, Cr) are more frequently distributed in fine particulate matter. Furthermore, some specific elements may also exhibit peak concentrations in particular particle size, which is closely related to their sources and formation processes. In recent years, there are still some challenges and deficiencies. Further research is needed on the particle size distribution characteristics of complex pollution sources (such as industrial emissions and traffic emissions). Additionally, there is a need to enhance the understanding of the transformation mechanisms and health effects of elemental components within particulate matter. This study selected a typical industrial and mining city to investigate particle size distribution characteristics of elemental components in atmospheric particulate matter. Anderson Eight-Stage Particulate Impactor Sampler was used to collect atmospheric particulate matter in the urban area of Huangshi during winter and summer. Nine particle size range samples were obtained spanning from 0 to 0.4 µm, 0.4 to 0.7 µm, 0.7 to 1.1 µm, 1.1 to 2.1 µm, 2.1 to 3.3 µm, 3.3 to 4.7 µm, 4.7 to 5.8 µm, 5.8 to 9.0 µm, and 9.0 to 10 µm. Energy Dispersive X-Ray Fluorescence Spectrometry (ED-XRF) was employed to determine the concentrations of 17 elemental components, including S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Rb, Sb, Ba, and Pb. Elements Ca, S, Fe, K, Zn, Ba, and Pb were identified as the primary pollutants during the sampling period. All the elemental concentrations exhibited distinct seasonal variations, demonstrating higher levels in winter compared to summer. Each element demonstrated distinct particle size distribution characteristics with peak concentrations for most elements occurring in the 5.8 to 9.0 µm range and peaks for the remaining elements in the 0.4 to 1.1 µm range. The highest elemental concentrations in both summer and winter were mainly distributed in the 5.8 to 9.0 µm and 0.7 to 1.1 µm size ranges. In summer, most elemental concentrations were negatively correlated with relative humidity. However, in winter, there was no significant correlation with relative humidity. Rainfall had a certain scavenging effect on elements but was also influenced by other meteorological factors. Element S had the highest enrichment factor values in both summer and winter. Element Cl was highly enriched in finer particle size fractions in both seasons. Most elements were slightly enriched across all particle size fractions. Principal component analysis further identified the main sources as soil dust and wind-blown sand, coal combustion, vehicle exhaust emissions, biomass burning, mining and construction activities, and other pollution sources. These findings contribute to the formulation of effective pollution control measures and the protection of public health.

How to cite: Liu, H., Zhang, J., Zhan, C., Liu, S., Liu, T., and Xiao, W.: Size distribution of elemental components in atmospheric particulates from a typical industrial and mining city of Central China, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4641, https://doi.org/10.5194/egusphere-egu25-4641, 2025.

How to cite: Ball, C., Chang, M., Cruise, J., de Valk, C., and Kannan, V.: Industrial High Performance Computing Scalable and FAIR Workflow Opportunities for EO Operations Processing, Operations, and Archiving, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7004, https://doi.org/10.5194/egusphere-egu25-7004, 2025.

Severe convection, including thunderstorms and related phenomena like flash flooding, hail, and strong winds, can have significant socioeconomic impacts. Nowcasting, which provides real-time, short-term predictions, is vital for issuing timely warnings to mitigate these impacts. Satellite imagery is essential for monitoring convection and offering accurate predictions of storm evolution, thereby enhancing early warning systems. Ensemble forecasting, which generates multiple potential scenarios, helps better quantify uncertainties in nowcasting. However, most ensemble forecasting methods are computationally intensive and typically do not incorporate satellite images directly. The Analog Ensemble (AnEn) method, a lower-cost ensemble approach, identifies similar past weather events based on forecast data. For a given time and location, the AnEn method identifies analogs from past model predictions that are similar to current forecast conditions. Then their associated observations are used as ensemble members. Despite its advantages, AnEn struggles with locality and is sensitive to the choice of similarity metrics. This study presents an improved AnEn system that replaces forecast archives with satellite images to identify analogs of convective conditions. The system utilizes pretrained deep learning algorithms (VGG16, Xception, and Inception-ResNet) to assess image similarity. The training dataset consists of daily convection satellite images from EUMETSAT for the period 2020-2023, and the domain covers 40°N to 20°S and -20°W to 4°E. The year 2024 is used for testing, with ERA5 reanalysis of total precipitation as the verification ground-state. For a present convection satellite image this image is encoded and compared to all past encoded images of the training period using different metrics. The most similar images to the current one are then selected and their associated ERA5 total precipitation reanalysis are considered the members or our ensemble. Preliminary results indicate an average maximum precipitation anomaly of 15 mm between the analog ensemble mean and the current reanalysis, showing that the proposed system offers promising improvements in short-term forecasting.

Key words: Convection; Ensemble Forecasting; Deep Learning; VGG; Xception; ResNet; Analog Ensemble; Morocco; Nowcasting; EUMETSAT; ERA5; Morocco; Satellite Images; Remote Sensing;

How to cite: Alaoui, B., Bounoun, C., and Bari, D.: Leveraging Pretrained Deep Learning Models to Extract Similarities for the Analog Ensemble Method Applied to Convection Satellite Imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7258, https://doi.org/10.5194/egusphere-egu25-7258, 2025.

Amid global environmental degradation, understanding the spatiotemporal dynamics and trade-offs of ecosystem services (ESs) under varying land-use scenarios is critical for advancing the sustainable development of social–ecological systems. This study analyzed the Chaohu Lake Basin (CLB), focusing on four scenarios: natural development (ND), economic priority (ED), ecological protection (EP), and sustainable development (SD). Using the PLUS model and multi-objective genetic algorithm (MOGA), land-use changes for 2030 were simulated, and their effects on ESs were assessed quantitatively and qualitatively. The ND scenario led to significant declines in cropland (3.73%) and forest areas (0.18%), primarily due to construction land expansion. The EP scenario curbed construction land growth, promoted ecosystem recovery, and slightly increased cropland by 0.05%. The SD scenario achieved a balance between ecological and economic goals, maintaining relative stability in ES provision. Between 2010 and 2020, construction land expansion, mainly concentrated in central Hefei City, led to a marked decline in habitat quality (HQ) and landscape aesthetics (LA), whereas water yield (WY) and soil retention (SR) improved. K-means clustering analysis identified seven ecosystem service bundles (ESBs), revealing significant spatial heterogeneity. Bundles 4 through 7, concentrated in mountainous and water regions, offered high biodiversity maintenance and ecological regulation. In contrast, critical ES areas in the ND and ED scenarios faced significant encroachment, resulting in diminished ecological functions. The SD scenario effectively mitigated these impacts, maintaining stable ES provision and ESB distribution. This study highlights the profound effects of different land-use scenarios on ESs, offering insights into sustainable planning and ecological restoration strategies in the CLB and comparable regions.

How to cite: Jin, A.: Ecosystem Services Trade-Offs in the Chaohu Lake Basin Based on Land-Use Scenario Simulations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7785, https://doi.org/10.5194/egusphere-egu25-7785, 2025.

Seasonal ecosystems play a crucial role in environmental regulation and biodiversity by hosting complex ecological dynamics that vary with climatic conditions. The Chaschoc-Sejá wetlands are a key example of such systems in southeastern Mexico. The interaction between the lagoon system and the Usumacinta River is highly dynamic; during the rainy season, the lagoons increase in volume, reaching depths of 8 to 10 meters. However, the lagoons completely dry up during the dry season, leaving vegetation at the surface level. 

This project aims to analyze the dynamics of vegetation cover in this environment by comparing high-resolution satellite images (Planet, 3 m) and ortho-mosaics generated with a DJI Mavic 3 Multispectral drone (10 cm). By combining these datasets, we aim to improve our previous vegetation maps and obtain a more accurate and detailed assessment of the Chaschoc-Sejá Lagoon system. Understanding vegetation patterns at a larger scale during specific periods and the variations in plant life within the lagoon and along its shores is a key focus.

Data processing involved classifying vegetation cover and identifying seasonal changes using indices such as NDVI and NDWI. We also generated 3D models to estimate vegetation height. Results show that integrating both techniques significantly improves spatial resolution and temporal accuracy in monitoring these ecosystems. This study provides essential tools for managing seasonal systems and their conservation in the face of climatic and anthropogenic factors. This monitoring will aid in understanding vegetation status, identifying plant species, and contributing to managing and preserving the lagoon system.

How to cite: Nieto, J., Ramírez Serrato, N. L., Romero Herrera, A., Peralta Carreta, C., Herrera Zamarrón, G., Hernández Hernández, M. A., Hernández García, G. D. J., Olea Olea, S., Morales Casique, E., and Cortez Silva, A.: Integrating high-resolution satellite and multispectral drone Imagery for monitoring vegetation in the Chaschoc-Sejá lagoon system, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7921, https://doi.org/10.5194/egusphere-egu25-7921, 2025.

Benevento Province, located in the Campania region of Italy, may experience environmental quality impacts from neighboring developed areas such as Naples and Caserta. Previous studies have suggested that some agricultural chemicals from Naples, such as hexachlorobenzene, may be transported through the air to rural areas of Benevento. Additionally, high concentrations of polycyclic aromatic hydrocarbons (PAHs) have been detected in Naples and Caserta, making Benevento Province a potential PAH "sink." This study systematically investigated the occurrence of PAHs in soil from Benevento Province, southern Italy, and their correlations with environmental factors, soil-air exchange processes, and health risks. Over 95% of sampling sites exhibited ∑₁₆PAHs concentrations at non-polluted levels (9.50-1188 ng/g, mean = 55.0 ± 152 ng/g), and four-ring PAHs were the dominant pollutants contributing to 28.3% of ∑₁₆PAHs. The spatial distribution of PAHs presented significant heterogeneity, with hotspots concentrated near landfills. The results of Positive Matrix Factorization (PMF) model showed that the main sources of PAHs were vehicle emissions, coal/biomass combustion, and petroleum products volatilization/leakage, contributing 42.2%, 40.2%, and 17.6%, respectively. Most of PAHs correlated significantly with total organic carbon in the soil and population density, while only Benzo(b)fluoranthene (BbF) showed a significantly negative correlation with pH. The mass inventory of ∑₁₆PAHs ranged from 0.94 to 29.4 tons, averaging 2.45 tons. The synergistic effects of pollution hotspots and the persistent accumulation of PAHs in the soil suggested that the soil might act as a secondary source of PAHs. Toxicity equivalent and probabilistic risk assessments indicated that health risks from PAHs remained within acceptable limits.

How to cite: Qu, C. and Pu, C.: Investigation of polycyclic aromatic hydrocarbons in the soils of Benevento Province, Italy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10723, https://doi.org/10.5194/egusphere-egu25-10723, 2025.

The adoption of the FAIR data principles has revolutionised data management in Earth System Sciences (ESS), yet challenges persist in achieving true machine-actionability and comprehensive implementation. 

DeSci Labs introduces two innovative tools—Decentralised Persistent Identifiers (DPIDs) and the CODEX protocol—to address the barriers to FAIR data practice implementation in general; whilst fostering widespread uptake of FAIR principles in the ESS community in particular through involvement in the FAIR2Adapt consortium.

DPIDs are globally unique persistent identifiers based on, and linked directly to, the content of the files it refers to. Each version of every file, regardless of type, is automatically assigned a cryptographic fingerprint, ensuring deterministic resolution and transparent versioning. The DPID has been specifically designed to support FAIR digital research objects. The flexibility to alias DPIDs with existing systems (e.g., DOIs) and programmatic publishing capabilities via NodesLib enhances interoperability while preserving data sovereignty.

The CODEX protocol, an open scholarly infrastructure, further complements this by enabling the storage and retrieval of FAIR digital research objects via a decentralised peer-to-peer network (IPFS). This architecture allows multiple copies of the same content to be stored by different network participants using the same PID. By empowering researchers to collaborate within an open-state repository, the protocol minimises reliance on centralised actors, ensuring long-term accessibility, data integrity, and transparency. Its modular design facilitates diverse gateway applications, maximising participation and reducing barriers to entry.

These tools address core challenges in FAIR adoption by providing robust, scalable, and interoperable solutions tailored to the ESS community. By integrating DPIDs and CODEX into data workflows, researchers can enhance data reusability, improve the provenance of research outputs, and safeguard the collective scientific record. This presentation explores how these technologies can catalyse the next decade of FAIR data practices in ESS, fostering trust, reproducibility, and innovation.

How to cite: Raijmakers, L., Hübinette, E., DeSota, E., Iman, S., and Koellinger, P.: Advancing and Supporting FAIR Principle Adoption through Innovative Social Infrastructure Tools, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13330, https://doi.org/10.5194/egusphere-egu25-13330, 2025.

In mineral exploration and geometallurgical studies, accurately segmenting mineral grains from core scanning datasets may be used to predict metal recovery. This study introduces the application of the Segment Anything Model (SAM), a cutting-edge deep learning tool, to automate the segmentation and extraction of mineral grains from Laser-Induced Breakdown Spectroscopy (LIBS) and hyperspectral core scanning datasets. SAM demonstrates high efficiency and precision in identifying mineral grains, forming the foundation for downstream analyses, including the evaluation of mineral associations, grain size distribution, and other key geometallurgical metrics. Through case studies on pegmatite deposits, this research showcases the potential of SAM to address challenges posed by mineralogically complex ore. By enabling detailed mineralogical characterisation and advancing geometallurgical methods, SAM-based grain extraction presents a transformative tool for supporting sustainable and efficient mining practices.

How to cite: Cai, Y., Manton, R., and Williams, M.: Automated Mineral Grain Extraction for Geometallurgical Studies Using Segment Anything Model (SAM) and Core Scanning Techniques, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13917, https://doi.org/10.5194/egusphere-egu25-13917, 2025.

Synthetic data has become an indispensable tool in climate science, offering extensive spatio-temporal
coverage to address data limitations in both current and future scenarios. Such synthetic data, derived
from climate simulation models, must exhibit statistical consistency with observational datasets to ensure
their utility. Among global climate simulation initiatives, the Coupled Model Intercomparison Project
Phase 6 (CMIP6) represents the latest and most comprehensive suite of General Circulation Models
(GCMs). However, the substantial High Performance Computing (HPC) resources required for these
physics-based models limit their accessibility to a broader research community. In response, genera-
tive machine learning models have emerged as a promising alternative for simulating climate data with
reduced computational demands.
This study introduces an ensemble model based on the Pix2Pix conditional Generative Adversarial
Network (cGAN) to generate high-resolution spatio-temporal maps of monthly global Sea Surface Tem-
perature (SST) with significantly lower computational cost and time. The proposed model comprises two
components: the GAN, which produces simulated SST climatology data , and the Predictor, which is
trained with the variability of the data that forecasts SST anomaly for the subsequent month using the
output data from the previous month. Both components contain the same architecture, but the training
processes are different. The predictor model can be fine-tuned with observed data for some epochs to
adopt its domain.
The ensemble model was calibrated with monthly SST observations from the COBE dataset as in-
put and output. The Empirical Orthogonal Functions (EOF) shows the model’s ability to simulate the
variabilty of the observed data. The model’s performance was evaluated using the temporal Pearson cor-
relation coefficient and mean squared error (MSE). Results demonstrate that the ensemble cGAN model
generates maps with statistical characteristics closely matching those of CMIP6 simulations and obser-
vations, achieving a mean temporal correlation coefficient around 0.5 and an MSE around 1.13 for both
cases.

How to cite: Chakraborty, D. and Mitra, A.: Simulation of Monthly Global Sea Surface Temperature Data using Ensemble GAN Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14821, https://doi.org/10.5194/egusphere-egu25-14821, 2025.

Deep neural networks have revolutionized various fields due to their remarkable adaptability, enabling them to address related tasks through retraining and transfer learning. These capabilities make them invaluable tools for diverse applications, including climate and hydrological modeling. In an earlier work (Mishra Sharma et al., 2024), we introduced a novel neural network architecture, the Max-Average U-Net (MAUNet), which leverages Max-Average Pooling to downscale gridded precipitation data to higher spatial resolutions. The model demonstrated significant improvements in resolving finer-scale precipitation features, making it well-suited for climate data applications.

In this study, we utilized the MAUNet architecture to tackle the critical task of bias correction in satellite-based precipitation estimates. Bias correction is essential for improving the reliability of precipitation data derived from satellite missions, which often exhibit systematic discrepancies compared to ground-based measurements. Specifically, we focused on correcting biases in precipitation estimates from the Tropical Rainfall Measuring Mission (TRMM) by calibrating them against high-resolution, ground-based gridded datasets from the India Meteorological Department (IMD).

Our experimental results reveal that MAUNet effectively reduces biases in TRMM precipitation estimates, achieving significantly improved agreement with ground truth data. This success is attributed to the model’s robust feature extraction and reconstruction capabilities, which enable it to learn and correct systematic errors in satellite data. The findings also highlight the potential of advanced neural network architectures in addressing bias correction challenges.

This work underscores the utility of deep learning architectures in precipitation modeling, contributing to broader goals of improving the spatial distribution of precipitation estimates. By bridging the gap between satellite observations and ground truth, the MAUNet model offers a comprehensive solution for enhancing precipitation datasets, with significant implications for climate research, hydrological studies, and policy planning.

How to cite: Mishra Sharma, S. C. and Mitra, A.: Leveraging MAUNet for Bias Correction of TRMM Precipitation Estimates, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14839, https://doi.org/10.5194/egusphere-egu25-14839, 2025.

The rising frequency and severity of landslides, exacerbated by the effects of climate change and human development in unstable areas, call for effective risk management strategies. In this context, a systematic collection of all the available data regarding geotechnical aspects, in particular geomaterial parameters, results plays a crucial role, providing a decisive contribution to define strategies for sustainable landslide risk management.

In this work, we present the translation of a geotechnical database to the aims of the project Tech4You Innovation Ecosystem – Goal 1 - Pilot Project 1, useful to identify the typical landslide scenarios, to identify sufficient knowledge for the definition of the geotechnical model and geomaterials typing in similar geo-environmental contexts. The database contains the results of laboratory tests carried out in the past by researchers at CNR IRPI in Rende, relating to 11 sites in Calabria, of which 10 in the Province of Catanzaro and 1 in the Province of Vibo Valentia.  For each site, geotechnical characterisation data of the geomaterials, which represent a key cognitive element, were grouped by type of laboratory test (grain size, indices, Atterberg limits, oedometric, direct shear and triaxial tests). We uploaded these data to validate a tool, named GeoDataTech vers. 2.0, which is an update of a previous version. In particular, we have tested the correct functioning (display, query, extraction data) with a significant sample of data. GeoDataTech vers. 2.0 can manage 2399 laboratory tests to date: 61 oedometric tests, 636 grain size, 537 indices, 78 Atterberg limits, 454 specific gravity, 512 direct shear tests and 121 triaxial tests.

This tool will be available to a wide range of stakeholders (researchers, professionals, territorial administrations, public bodies and citizens) allowing us to acquire, interrogate, export data and to upload their own files to integrate them into the database of the tool, performing advanced analyses with reference to the typification of geomaterials. By enabling the sharing of such data between researchers, practitioners and public institutions, the geotechnical tool will contribute significantly to improving disaster prevention strategies, in particular with regard to the reduction of landslide risks, thereby responding to the growing demand for accessible and interoperable data networks that increase synergic interdisciplinary research on topics such as landslide hazard.

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. This work reflects only the authors' views and opinions, neither the Ministry for University and Research nor the European Commission can be considered responsible for them.

How to cite: Scarcella, G. E., Aceto, L., and Gullà, G.: A relevant accessible and interoperable geotechnical data tool to support the landslide risk management, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15216, https://doi.org/10.5194/egusphere-egu25-15216, 2025.

Changes in land use and cover (LULC) serve as critical indicators of socioeconomic and environmental shifts induced by both natural and man-made factors. This assessment was carried out in the Imphal valley region to forecast changes in land use and land cover. In order to examine the spatiotemporal distributions of LULC, the LULC Classification was analysed using Landsat images from 2007, 2014, and 2017. The CA-Markov Chain model was used to simulate the future LULC for the year 2030 of Imphal valley region based on these the past LULCs. The model result showed that wetland herbaceous will decline by 3.3% and settlement area will expand by 28.71%. The Imphal city area is where the majority of the expanding settlement area is located. As a vital resource for future planning initiatives, this study suggests planners, environmentalists, and decision-makers to prioritise sustainable practices and make appropriate decisions for the sustainability of the region.

How to cite: Nongthouba, M., Oinam, B., and Sachidananda, K.: Prediction of landuse landcover using CA-Markov model for the valley regions of Manipur, India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15381, https://doi.org/10.5194/egusphere-egu25-15381, 2025.

GNSS (Global Navigation Satellite System) data play a crucial role in both scientific research and practical applications. GNSS datasets are used to monitor atmospheric conditions, tectonic plate movements, and Earth deformation, providing valuable insights for geodetic and geophysical studies. Although widely accessible, GNSS data often lacks the necessary structure and metadata for effective reuse, particularly for data-driven research based on machine learning. To address these challenges, we applied the FAIR (Findable, Accessible, Interoperable, Reusable) data principles to GNSS RINEX observation files hosted by the EUREF Historical Data Centre (EUREF-HDC).

The EU action plan “Turning FAIR into Reality” introduced the concept of FAIR Digital Objects (FDOs), emphasizing the need for Persistent Identifiers (PIDs) and rich, standardized metadata to ensure data can be reliably found, accessed, utilized, and cited. Building on this foundation, we developed a multi-layered FDO structure centered on GNSS RINEX data. Given the established nature of the EUREF-HDC repository, we adapted the FDO concept by prioritizing structured metadata, followed by persistent identifiers and robust (meta)data access procedures.

To implement this approach, we designed the GNSS-DCAT-AP metadata schema, assigned PIDs to both data and metadata, and developed web services enabling humans and machines alike to seamless search, retrieve, and download (meta)data. The effectiveness of our solution was evaluated using the FAIRsFAIR Data Object Assessment Metrics, demonstrating a significant improvement in FAIR compliance.

This work showcases the feasibility of transforming GNSS RINEX data into FAIR Digital Objects and could provide a practical roadmap for other geospatial data repositories seeking alignment with FAIR principles.

How to cite: Bruyninx, C., Miglio, A., Fabian, A., Legrand, J., Pottiaux, E., and Bamahry, F.: Transforming GNSS Data into FAIR Digital Objects, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16092, https://doi.org/10.5194/egusphere-egu25-16092, 2025.

Tropical Cyclones (TCs) are among the most impactful weather phenomena, with climate change intensifying their duration and strength, posing significant risks to ecosystems and human life. Accurate TC detection, encompassing localization and tracking of TC centers, has become a critical focus for the climate science community. 

Traditional methods often rely on subjective threshold tuning and might require several input variables, thus making the tracking computationally expensive. We propose a cost-effective hybrid Machine Learning (ML) approach consisting in splitting the TC detection into two separate sub-tasks: localization and tracking. The TC task localization is fully data-driven: multiple Deep Neural Networks (DNNs) architectures have been explored to localize TC centers using a different set of input fields related to the cyclo-genesis, aiming also at reducing the number of input drivers required for detection. A neighborhood matching algorithm is then applied to join previously localized TC center estimates into potential trajectories over time. 

We train the DNNs on 40 years of ERA5 reanalysis data and International Best Track Archive for Climate Stewardship (IBTrACS) records across the East and West North Pacific basins. The hybrid approach is then compared with four state-of-the-art deterministic trackers (namely OWZ, TRACK, CNRM and UZ), reporting comparable or even better results in terms of Probability of Detection and False Alarm Rate, additionally capturing the interannual variability and spatial distribution of TCs in the target domain. 

The resulting hybrid ML model represents the core component of a Digital Twin (DT) application implemented in the context of the EU-funded interTwin project.

How to cite: Donno, D., Accarino, G., Elia, D., Scoccimarro, E., and Gualdi, S.: Hybrid Machine Learning approach for Tropical Cyclones Detection, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16363, https://doi.org/10.5194/egusphere-egu25-16363, 2025.

The STARS4Water project addresses the critical need to understand the impacts of climate change and anthropogenic activities on freshwater availability and ecosystem resilience at the river basin scale. By developing innovative data services and models tailored to stakeholder needs, the project will improve decision-making processes for sustainable water resource management. A distinctive feature of STARS4Water is its focus on co-creating solutions with local stakeholders using a living lab approach, ensuring that newly developed tools remain relevant and usable beyond the life of the project.

This extension of the original project—funded with a special grant from Unitatea Executivă pentru Finanțarea Învățământului Superior, a Cercetării, Dezvoltării și Inovării (UEFISCDI) from Romania—focuses on a detailed change detection analysis to monitor and quantify land cover transformations in the emblematic Danube Delta region. The objective is to assess how environmental and anthropogenic changes have influenced this ecologically significant wetland over several decades. To achieve this, a comprehensive database of multispectral satellite images from the Landsat archive, spanning from 1985 to 2023, will be constructed. The long-term dataset enables a detailed temporal analysis, important for detecting land cover dynamics over time.

The methodology involves several key phases: (1) data collection and preprocessing of Landsat satellite images to correct errors and align imagery for consistent comparative analysis; (2) sampling and training a deep learning model using convolutional neural network (CNN) architectures, to classify various land cover types; (3) performing land cover classification on the processed images using the trained model, followed by accuracy assessment; and (4) conducting a comprehensive change detection analysis to quantify and interpret the observed transformations in land use and land cover.

The results of this analysis will deliver important knowledge on the long-term dynamics of the Danube Delta landscape, highlighting critical changes with implications for biodiversity, water management and ecosystem services. This approach will support adaptive ecosystem management and contribute to the scientific understanding of climate-related and anthropogenic changes in fragile wetland ecosystems.

Acknowlegments

This work was supported by a grant of the Ministry of Research, Innovation and Digitization, CNCS/CCCDI - UEFISCDI, project number PN-IV-P8-8.1-PRE-HE-ORG-2023-0094, within PNCDI IV.

How to cite: Scrieciu, A. and Toma, A.: Monitoring Long-Term Land Cover Transformations in the Danube Delta using Landsat Satellite Imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16549, https://doi.org/10.5194/egusphere-egu25-16549, 2025.

Abstract: Cloud cover poses a significant obstacle in harnessing multi-spectral satellite imagery for various earth observation applications including disaster response, land use and land cover mapping. To address this issue, this study investigates the potential of Prithvi WxC foundation model (Johannes Schmude et al., 2024), a deep learning architecture designed for weather and climate applications, to perform cloud gap imputation. By leveraging its ability to capture atmospheric dynamics and predict missing data, Prithvi WxC offers a promising solution.

The primary objective is to assess the accuracy and efficiency of Prithvi WxC in reconstructing cloudy pixels in Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance; MOD09 (Eric Vermote, 2015). MOD09 data provides valuable information about earth surface, cloud cover and atmospheric conditions, which are instrumental in informing the Prithvi WxC model during the finetuning and imputation process.

This research evaluates the Prithvi WxC foundation model for cloud gap imputation applications and benchmarks its performance against other foundation models, such as Prithvi EO (Jakubik et al., 2023, 2024). The process begins with preprocessing the MOD09 dataset, filtering out missing and cloudy pixels to create clean visible patches, while real-world cloudy patches are used as masks. The preprocessed data is then resampled to align with the temporal and spatial resolution requirements of both the Prithvi WxC and Prithvi EO foundation models. Through rigorous fine-tuning strategies, these models learn to reconstruct the masked regions, effectively filling the gaps caused by cloud cover. Finally, the fine-tuned foundation models are benchmarked using quantitative metrics, such as the Structural Similarity Index Measure (SSIM) and Mean Absolute Error (MAE), complemented by qualitative visual analysis.

This research explores the potential of Prithvi WxC foundation model, pre-trained on extensive weather and climate data, to improve cloud gap imputation in satellite imagery, and subsequently benchmarks it against earth observation foundation models, such as Prithvi EO. Through this evaluation, we aim to enhance scientific understanding via multi-modality and sensor-independent approaches.

 References

Johannes Schmude, Sujit Roy, Will Trojak, Johannes Jakubik, Daniel Salles Civitarese, Shraddha Singh, Julian Kuehnert, Kumar Ankur, Aman Gupta, Christopher E Phillips, Romeo Kienzler, Daniela Szwarcman, Vishal Gaur, Rajat Shinde, Rohit Lal, Arlindo Da Sil: Prithvi WxC: Foundation Model for Weather and Climate." arXiv preprint arXiv:2409.13598, 2024.

C. Roger, E. F. Vermote, J. P. Ray: https://modis-land.gsfc.nasa.gov/pdf/MOD09_UserGuide_v1.4.pdf. NASA, MODIS Surface Reflectance User’s Guide, Collection 6, 2015.

Daniela Szwarcman, Sujit Roy, Paolo Fraccaro, Þorsteinn Elí Gíslason, Benedikt Blumenstiel, Rinki Ghosal, Pedro Henrique de Oliveira, Joao Lucas de Sousa Almeida, Rocco Sedona, Yanghui Kang, Srija Chakraborty, Sizhe Wang, Ankur Kumar, Myscon Truong, Denys: Prithvi-EO-2.0: A Versatile Multi-Temporal Foundation Model for Earth Observation Applications. https://arxiv.org/abs/2412.02732, 2024.

Johannes Jakubik, Sujit Roy, C. E. Phillips, Paolo Fraccaro, Denys Godwin, Bianca Zadrozny, Daniela Szwarcman, Carlos Gomes, Gabby Nyirjesy, Blair Edwards, Daiki Kimura, Naomi Simumba, Linsong Chu, S. Karthik Mukkavilli, Devyani Lambhate, Kamal Das, Ranji: Foundation Models for Generalist Geospatial Artificial Intelligence, 2023.

Eric Vermote: MOD09 MODIS/Terra L2 Surface Reflectance, 5-Min Swath 250m, 500m, and 1km. NASA LP DAAC., NASA GSFC and MODAPS SIPS, NASA, http://doi.org/10.5067/MODIS/MOD09.061, 2015.

How to cite: Medimem, T. B., Padovani, G., Kurihana, T., Kumar, A., Melgani, F., Anantharaj, V. G., and Fiore, S. L.: Finetuning and Benchmarking an AI Foundation Model for Cloud Gap Imputation , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16634, https://doi.org/10.5194/egusphere-egu25-16634, 2025.

The urgency of the climate crisis and the need to accelerate learning and climate action requires that we build on previous knowledge, rather than replicating it. There is an abundance of knowledge on climate change adaptation and mitigation dispersed across websites, projects, platforms, and documents. There is either too much information that is not easily discoverable (sitting in silos) or it is too technical or complex, and not ‘usable’ or fit for purpose in terms (e.g. language or format). Both issues cause redundancy and sometimes replication of work, wasting resources. In the worst case, they can also cause unintended consequences such as maladaptation, or increased vulnerability. However, the issue is not a lack of information, but rather how to organise and connect such knowledge to allow people to discover what already exists and put it to effective use. As such, our goal is to make climate action knowledge findable, accessible, interoperable and reusable (FAIR) and reduce climate change knowledge silos. The recently awarded FAIR2Adapt Project aims to establish a comprehensive FAIR and open data framework for CCA and to demonstrate the impact of FAIR data on CCA strategies. By making CCA data FAIR, FAIR2Adapt will accelerate adaptation actions so that they are visible, understandable, and actionable for various purposes and different types of stakeholders. FAIR taxonomies are one approach to help tackle this issue by making climate change knowledge FAIR and by ensuring, that going forward, platforms have a way to make their knowledge FAIR and thus more reusable by the climate change community. 

The Climate Connectivity Hub and Taxonomy seek to visualize and connect online platform data (e.g. Cordis, Climate-ADAPT, weADAPT, PreventionWeb) to increase discoverability, interoperability and a shared understanding of the research results and their potential application in future policy, research and practice. It builds on past knowledge to scale up and accelerate climate action whilst also identifying key knowledge gaps. This presentation will show that: 1) taxonomies are useful supporting the interoperability of online climate knowledge and can usefully emerge from combined expert and machine learning of project results (e.g. Cordis); 2) shared vocabularies and different interpretations of language and terminology add value to project planning and implementation ; and, 3) the visualization of these elements for decision-makers, planners, researchers, policy makers, etc. can help to enable and scale accelerated climate action. 

How to cite: Bharwani, S., Witton, R., Williamson, K., and Butterfield, R.: Visualizing a climate and disaster resilience taxonomy from research evidence: scaling and accelerating knowledge interoperability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18822, https://doi.org/10.5194/egusphere-egu25-18822, 2025.

The study explores using advanced deep learning (DL) techniques for spatial-spectral analysis to detect and map peatland degradation at a granular level. Peatlands, vital carbon sinks in global ecosystems, face degradation threats that demand precise and scalable monitoring solutions. Our method combines convolutional neural networks (CNNs), fully convolutional networks (FCNs), and 3D CNNs to examine complex spatial-spectral patterns in SAR, multispectral, and hyperspectral sensor data (e.g., Sentinel-1, Sentinel-2, PRISMA) over the temperate peatland study area of the Monte Bondone region (Latitude: 46°00’48.6” N, Longitude: 11°03'14.6” E), covering an area of 40 hectares as shown in the figures.

CNNs capture spatial relationships between precipitation, temperature, vegetation, soil, and moisture, offering a detailed view of peatland composition. Using multi-dimensional, gridded data from meteorological stations and remote sensing images, CNNs identify patterns affecting peatland health. Fully Convolutional Networks (FCNs) help with spectral unmixing, isolating land cover components at the pixel level, which aids in detecting vegetation degradation and understanding ecosystem changes.

3D CNNs incorporate temporal data to classify Peatland landscapes into different degradation states. The model identifies changes over time, distinguishing between healthy, partially degraded, and fully degraded regions. Deep clustering models also classify peatland areas into degradation states, revealing trends without labeled data.

This deep learning framework supports accurate degradation mapping through spatial-spectral feature extraction, providing precise, pixel-level information to aid ecosystem management and conservation. It helps monitor peatland health and assess environmental changes across diverse landscapes.

How to cite: Kaparthi, H. V. and Vitti, A.: Deep Learning-based Spatial-Spectral Analysis for Peatland Degradation characterization, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18933, https://doi.org/10.5194/egusphere-egu25-18933, 2025.

Integrating heterogeneous hydrological datasets remains a significant challenge in environmental modelling due to variations in feature spaces, data distributions, and temporal and spatial scales across sources. This study introduces a Model-Agnostic Meta-Learning (MAML) approach to address the challenge of integrating heterogeneous hydrological datasets, leveraging a collection of datasets compiled from diverse sources. These datasets, characterized by varying features, distributions, and temporal and spatial scales, provide an ideal basis for evaluating MAML's ability to handle real-world data heterogeneity.

MAML’s unique capability to learn shared representations across datasets with minimal feature overlap and significant variability allows it to effectively transfer knowledge between subsets, offering a flexible and scalable solution for integrating hydrological data with diverse characteristics.

The proposed approach trains a base model on one subset of the data while utilizing MAML's meta-learning capabilities to adapt and transfer knowledge to other subsets with differing feature distributions. To test the model's adaptability, we simulate scenarios with varying degrees of feature overlap. Model performance is assessed using metrics such as mean squared error (MSE), both before and after fine-tuning on unseen data subsets.

Preliminary results demonstrate that MAML effectively learns shared representations across datasets, achieving significant improvements in prediction accuracy. Fine-tuning further enhances the model's adaptability, particularly for datasets with minimal feature overlap. These findings highlight MAML's potential as a powerful and flexible tool for integrating and predicting across heterogeneous hydrological datasets.

This study bridges the gap between advanced meta-learning techniques and hydrological applications, providing new insights into scalable and adaptable data integration methods for environmental sciences.

Keywords: Model-Agnostic Meta-Learning, hydrological datasets, data integration, heterogeneous data, meta-learning, environmental modelling, machine learning. 

How to cite: Slaimi, A. and Scriney, M.: Model-Agnostic Meta-Learning for Data Integration Across Heterogeneous Hydrological Datasets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19489, https://doi.org/10.5194/egusphere-egu25-19489, 2025.

The semi-arid region of Deraa Oued Noun in Morocco faces significant challenges related to water scarcity, which greatly affects the availability of groundwater resources. With recurring droughts and periods of water shortage, it is imperative to address these challenges and implement effective measures for sustainable groundwater resource management. Artificial groundwater recharge has proven to be a viable solution for alleviating water scarcity issues. By capturing and storing excess water during periods of heavy precipitation or surface water availability, artificial recharge can replenish depleted aquifers and provide a reliable water source during drought periods. However, the success of recharge projects depends on identifying suitable sites that meet specific criteria and maximize the efficiency of the recharge process.

The identification of suitable sites for artificial groundwater recharge in Daraa Oued Noun, through the integration of remote sensing, GIS (Geographic Information System), and MCDM (Multi-Criteria Decision Making) techniques, offers a promising solution to address water scarcity challenges in the context of climate change. The proposed research project aims to provide valuable and spatially explicit information for strategic groundwater resource management.

 This study was conducted in the Deraa Oued Noun district, where water shortages have been observed over the years. The research utilized geology, soil, land use, stream data, and Sentinel-2 and DEM images to develop thematic layers, including lithogeology, soil, slope, lineament density, land use, stream density, and water surface. Additionally, data on the vadose zone thickness were incorporated to enhance the analysis.

By integrating GIS and image processing techniques, these thematic layers were utilized to prepare groundwater recharge maps of the area through a weighted overlay method on a GIS platform. The results revealed that artificial recharge potential was high in the northern and western parts of the study area.

By following a systematic and rigorous methodology, including data collection, remote sensing analysis, MCDM evaluation, and site validation, this project aims to contribute to the successful implementation of artificial recharge projects in the region. By maximizing the efficiency of the recharge method, these projects will help ensure sustainable water supply, mitigate the impacts of drought, and promote long-term water security in Derâa Oued Noun and similar semi-arid regions.

How to cite: Fri, R., Scozzari, A., Haida, S., Kili, M., Erraoui, L., Chaou, J., Mridekh, A., Goumghar, L., and El Mansouri, B.: Using Remote sensing and geographic information system for delineating suitable sites for artificial groundwater recharge: A multi-criteria decision-making approach., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20324, https://doi.org/10.5194/egusphere-egu25-20324, 2025.

We present LUCIE, a data-driven atmospheric emulator that remains stable during autoregressive inference for a thousand of years with minimal drifting climatology. LUCIE was trained using 9.5 years of coarse-resolution ERA5 data, incorporating 5 prognostic variables, 2 forcing variables, and one diagnostic variable (6-hourly total precipitation), all on a single A100 GPU over a two-hour period. LUCIE autoregressively predicts the prognostic variables and outputs the diagnostic variables similar to AllenAI’s ACE climate emulator. Unlike all the other state-of-the-art AI weather models, LUCIE is neither unstable nor does it produce hallucinations that result in unphysical drift of the emulated climate. The low computational requirements of LUCIE allow for rapid experimentation including the development of novel loss functions to reduce spectral bias and improve tails of the distributions. Furthermore, LUCIE does not impose true sea-surface temperature (SST) from a coupled numerical model to enforce the annual cycle in temperature. We demonstrate the long-term climatology obtained from LUCIE as well as subseasonal-to-seasonal scale prediction skills on the prognostic variables. LUCIE is capable of 6000 years of simulation per day on a single GPU, allowing for O(100)-ensemble members for quantifying model uncertainty for climate and ensemble weather prediction.

How to cite: Guan, H., Arcomano, T., Chattopadhyay, A., and Maulik, R.: LUCIE: A Lightweight Uncoupled ClImate Emulator with long-term stability and physical consistency for O(1000)-member ensembles, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20616, https://doi.org/10.5194/egusphere-egu25-20616, 2025.

Analysis of pollen events was conducted using Hirst-type volumetric samplers in Athens and Thessaloniki in combination with CALIPSO vertical aerosol profiles. While Hirst-type ‎[1] volumetric samplers are used to confirm and characterize pollen at ground level, the understanding of pollen vertical distribution and transport is still limited. The utilization of Light Detection and Ranging (LIDAR) for identifying different pollen types is increasingly prevalent, as the depolarization ratio is related to the shape of the pollen particles while other non-spherical particle types are absent ‎[2].
Samplers are situated on the buildings’ rooftops of the Physics and Biology Departments, in Athens and Thessaloniki, respectively. Following ‎[2], intense pollen events are considered when the pollen concentration exceeds 400 grains m-3 for a minimum of two hours each day.
CALIPSO provides unique vertical profile measurements of the Earth’s atmosphere on a global scale ‎[3], with the ability to distinguish between feature types (i.e., clouds vs. aerosol) and subtypes (i.e., marine, dust, clean continental). Only case studies where CALIPSO aerosol layers were classified as marine, dusty marine, dust, or polluted dust were analyzed.
Model simulations were used to exclude the presence of other depolarizing aerosol types. HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) was used to trace the origin of the air masses. The atmospheric model RAMS/ICLAMS (Regional Atmospheric Modeling System/Integrated Community Limited Area Modeling System) was selected to describe dust and sea-salt emissions and transport.
Mean values of lidar-derived optical properties inside the detected pollen layers are provided from CALIPSO data analysis. Specifically, there are three observed aerosol layers, one over Athens (12-3-2021) and two over Thessaloniki (2-3-2020, 10-4-2020). Particulate color ratios of 0.652 ± 0.194, 0.638 ± 0.362, and 0.456 ± 0.284, and depolarization ratios of 8.70 ± 6.26%, 28.30 ± 14.16%, and 8.96±6.87 % for 12-3-2021, 2-3-2020 and 10-4-2020, respectively, were misclassified by CALIPSO as marine-dusty marine, dust and polluted dust. The pollen analysis conducted on the 12th of March 2021 indicated that the dominant pollen types were 69% Pinaceae and 24% Cupressaceae. On the 2nd of March 2020, Cupressaceae accounted for 97% of the total pollen, while on the 10th of April 2020, Carpinus represented 76% and Platanus 15%. Consequently, during periods of intense pollen presence, CALIPSO vertical profiles and aerobiological monitoring techniques may be used synergistically to better characterize the atmospheric pollen layers.

Acknowledgements
The research work was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) under the “Basic Research Financing (Horizontal support for all Sciences), National Recovery and Resilience Plan (Greece 2.0)” (Project Number: 015144).

[1] J. M. Hirst, Annals of Applied Biology 39, 157-293 (1952).
[2] X. Shang et al., Atmos. Chem. Phys. 20, 15323–15339 (2020).
[3] D. M. Winker et al, BAMS 91, 1211–1229 (2010).

How to cite: Karageorgopoulou, A., Christos, S., Thanasis, G., Χiaoxia, S., Ioanna, P., Vassilis, A., and Elina, G.: Assessment of Atmospheric Pollen Presence in Urban Areas of Greece During CALIPSO Overpasses, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21759, https://doi.org/10.5194/egusphere-egu25-21759, 2025.

The Geostationary Ocean Color Imager-II (GOCI-II), which was launched on February 19, 2020, offers an increased observation times within a day and finer spatial resolution than those of its predecessor, the Geostationary Ocean Color Imager (GOCI), which was launched in 2010. To ensure the reliability of GOCI-II data for practical applications, the accuracy of remote sensing products must be validated. In this study, we employed in situ data from Lake Taihu for validation. We assessed the accuracy of GOCI-II products, including the remote sensing reflectance inverted via two atmospheric correction algorithms (ultraviolet (UV) and near-infrared (NIR) atmospheric correction algorithms), as well as the chlorophyll a (Chl-a) concentration, total suspended matter (TSM) concentration, and phytoplankton absorption coefficient (a_ph). Our results revealed that the UV atmospheric correction algorithm provided a relatively higher accuracy in Lake Taihu, with average absolute percentage deviations (APDs) of the remote sensing reflectance across different bands of 25.17% (412 nm), 29.69% (443 nm), 22.27% (490 nm), 19.38% (555 nm), 36.83% (660 nm), and 33.0% (680 nm). Compared to the products generated using the NIR atmospheric correction algorithm, the derived Chl-a concentration, TSM concentration, and a_ph products from the UV algorithm showed improved accuracy, with APD values reduced by 16.92%, 3.32%, and 10.91%, respectively. When using UV correction, the 412 nm band performed better than the 380 nm band, likely due to the lower signal-to-noise ratio of the 380 nm band and smaller extrapolation errors when assuming a zero signal for the 412 nm band. Considering that the NIR algorithm is suitable for open ocean waters while the UV algorithm demonstrates higher accuracy in highly turbid environments, a combined UV-NIR atmospheric correction algorithm may be more suitable for addressing different types of water environments. Additionally, more suitable retrieval algorithms are needed to improve the accuracy of Chl-a concentration and a_ph in eutrophic waters.

How to cite: Zhao, M., Li, H., Li, H., Zhang, X., Ding, X., and Gong, F.: Assessment of GOCI-II satellite remote sensing products in Lake Taihu, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2237, https://doi.org/10.5194/egusphere-egu25-2237, 2025.

This study explores studies on lakes and paleolakes originating from natural effects. The main objective is to perform a bibliometric analysis of research on naturally occurring lake environments worldwide, covering the period from 2014 to 2024. Data extracted from 1687 documents in the Scopus database were analyzed using VOSviewer software. The results reveal a strict trend towards a focus on geosciences and the environment, underlined by research. This study particularly highlights the relationships between authors, co-authors, keywords, and publishers of specialized journals in this research field, thus providing essential information to guide future research and to value the role of these geological environments, which are rare in the world, based on essentially multidisciplinary geoscience approaches.

How to cite: Abbach, J., El Moussaoui, S., El Talibi, H., and Bouiss, C. E.: bibliometric analysis of natural lakes and paleolakes origin of natural events, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4994, https://doi.org/10.5194/egusphere-egu25-4994, 2025.

Rock masses are characterized by the complex hierarchical structures involving various scale levels. The deformation of rock masses is primarily controlled in weak structural layers between rocks, whereas the rock block can be regarded as a non-deformable block and can move as a whole. In consequence, a new dynamic phenomenon, namely the pendulum-type wave, has emerged, which is a kind of nonlinear displacement wave caused by the overall movement of relatively intact large-scale rock blocks. Aiming at the complex hierarchical structures of rock masses and low-frequency characteristics of pendulum-type waves, the blocky rock masses composed of granite blocks and rubber interlayers are simplified into the block-spring model and wave motion model. Based on Bloch theorem and d’Alembert’s principle, the dispersion relation and equations of motion of 1D blocky rock masses are determined. Research shows that with the increase of the rock size and geomechanical invariant, the initial frequency of the first attenuation zone gradually decreases, and only the low-frequency waves lower than that frequency can propagate in blocky rock masses, which reveals the mechanism of low-frequency characteristics of pendulum-type waves theoretically. The equivalent substitution for the two models and their errors are given, and the results show that the equivalent substitution of the two models is not universal and unconditional. Finally, the influence of hierarchical structures on the dispersion relation and dynamic response is further studied. The larger the stiffness ratio, or the higher the order of hierarchical structures, the smaller is the effect of ignoring the high-order hierarchical structures.

How to cite: Jiang, K. and Qi, C.: Research on dispersion relation and dynamic properties of pendulum‑type waves in 1D blocky rock masses with complex hierarchical structures, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5033, https://doi.org/10.5194/egusphere-egu25-5033, 2025.

Coral reefs in the Mariana Islands serve important roles for the islands’ ecology and economy, contributing to the region’s fisheries, tourism, coastal protection, education, and cultural histories. Despite their immense value, the resilience of these marine ecosystems is threatened by an array of climate-change induced stressors, including ocean acidification and coral bleaching. In response, a team from the University of Guam (UOG) launched a large-scale coral reef mapping campaign to monitor priority reef sites throughout the Mariana Islands using drone technology. The UOG team, consisting of researchers and remote pilots funded by the USGS, Pacific Islands Climate Adaptation Science Center, NASA Guam EPSCoR, and NASA Guam Space Grant, has been conducting drone-based missions to capture high-resolution imagery of priority coral reef sites across Guam and Saipan. Their efforts aim to gather aerial data of coral reefs in Micronesia, providing resource managers with essential information regarding response and recovery. Initially, the campaign used of NASA’s fluid lensing technology developed by Chirayath (2019) for coral reef mapping. This technology combines unmanned aerial systems (UAS), off-the-shelf technology, and machine learning algorithms to create detailed coral reef maps by filtering out distortions caused by light and ocean waves, resulting in clear, high-resolution imagery. In 2024, this process was augmented to employ a new methodology that strategically uses RGB sensors and low tides. This system allows the remote pilots to capture the areas and produce orthomosaic maps at much more efficient rates while maintaining high-resolution quality. By providing these datasets within a shorter turn-around time, local natural resource managers are able to get a timely snapshot of the coral reef sites – providing crucial data of the ecosystem’s health that can help inform conservation decisions. This presentation will outline the collaborative efforts between UOG and regional partners, demonstrating how drones and fluid lensing technology are innovating coral reef monitoring efforts. It will explore how the collected data can help local resource managers make informed decisions regarding coral reef management, showcase coral reefs to the general public, ultimately transforming how local communities can contribute to coral reef resilience.

How to cite: Sayama, J. and Fausto, K.: Innovating Coral Reef Mapping with Drones & NASA Fluid Lensing Technology in the Mariana Islands, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5359, https://doi.org/10.5194/egusphere-egu25-5359, 2025.

In today's changing climate, there is an urgent need to understand the adverse impacts of climate change on natural environments, infrastructures, and industries.Particularly permafrost regions in the Arctic are highly vulnerable to global warming, impacting both the environment and socioeconomic aspects. Thus, systematic monitoring of such environments, is of paramount significance. Advances in Geoinformation technologies, and in particular in Earth Observation (EO), cloud computing, GIS, web cartography create new opportunities and challenges for Arctic research examining the socioeconomic impact of climate change.The rapid advancements in EOin particular have led to an exponential increase in the volume of geospatial data that come from spaceborne EO sensors. This surge, combined with the fast developments in GIS and web cartography present significant challenges for effective management, access, and utilization by researchers, policymakers, and the public. Consequently, there is a growing need for advanced methodologies to organize, process, and deliver geospatial information that comes from EO satellites in an accessible and user-friendly manner.

Recognizing thepromising potential of geoinformation technologies, the European Union (EU) has funded several research projects that leverage advanced technologies such as geospatial databases and WebGIS platforms to streamline EO data handling and dissemination. One such project is EO-PERSIST (http://www.eo-persist.eu), which aims to create a collaborative research and innovation environment focusing on leveraging existing services, datasets, and emerging technologies to achieve a consistently updated ecosystem of EO-based datasets for permafrost applications. To formulate the socioeconomic indicators, the project exploits state of the art cloud processing resources, innovative Remote Sensing (RS) algorithms, Geographic Information Systems (GIS)-based models formulating, exchanging also multidisciplinary knowledge.EO-PERSIST innovative approach is anticipated to contribute to more informed decision-making and broader data accessibility for researchers, policymakers, and other stakeholders.

The present contribution aim is two-fold: at first, to provide an overview of EO-PERSIST Marie Curie Staff Exchanges EU-funded research project; second, to present some of the key project outputs delivered so far relevant to the selected Use Cases of the project and the geospatial database developed for assessing the socioeconomic impacts of climate change in the permafrost Arctic regions.

This study is supported by EO-PERSIST project which has received funding from the European Union's Horizon 2020 research and innovation program under grant agreement No. 101086386.

KEYWORDS:earth observation, cloud platform, Arctic, socioeconomic impact

How to cite: Tselos, G.-N., Detsikas, S. E., Kroszka, B., Grzybowski, P., and Petropoulos, G. P.: Enhancing the use of Geoinformation technologies to assess the socioeconomic impacts of climate change in the Arctic: Insights from the EO-PERSIST Project, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6240, https://doi.org/10.5194/egusphere-egu25-6240, 2025.

The U.S. territory of Guam is threatened annually by high-intensity storms and typhoons due to its location in the western Pacific Ocean. The island’s infrastructure – buildings, roads, and utilities – bear the brunt of typhoon damage, which in turn affects public health, the economy, and natural resources. Traditionally, these impacts have been observed via satellite, radar, and official weather stations.  Damages are assessed in the aftermath of the typhoon with a manual, on-the-ground approach led by the National Weather Service (NWS). This is often exhaustive and time-consuming for the assessment team. Observations from the ground can inadvertently create data gaps on damage assessments due to inaccessible areas caused by vegetative and construction debris, and flooded roads and pathways. This may not capture many impacts eligible for local or federal assistance. To address these data gaps and augment damage assessments, the University of Guam (UOG) Drone Corps program aims to assist local and federal government agencies (e.g., utility companies, public health, emergency services, and natural resource management) by collecting high-resolution aerial imagery to help prioritize and allocate limited resources. This presentation highlights the results of this novel collaboration of UOG, NWS, Guam Homeland Security (GHS), and the Office of the Governor of Guam in the creation of the damage assessment of Typhoon Mawar, which ravaged Guam on 24-25 May 2023. Following the typhoon, UOG worked with NWS to identify and capture imagery of vulnerable sites that were heavily impacted. This presentation will also share how UOG Drone Corps’ data was disseminated among other agencies as supplemental data for natural disaster recovery efforts. The presentation will conclude with a summary of the UOG Drone Corps program model as a resource for developing resiliency strategies for vulnerable island communities using advanced and emerging technologies. 

How to cite: Fausto, K. and Sayama, J.: Deploying UAV technology to assess typhoon impacts in vulnerable communities in Guam , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8063, https://doi.org/10.5194/egusphere-egu25-8063, 2025.

INTRODUCTION. InSAR or Interferometric Synthetic Aperture Radar is a technique for mapping ground deformation using radar images of the Earth's surface collected from orbiting satellites. DInSAR or Differential SAR Interferometry is an active remote sensing technique based on the principle that, due to the very high stability of the satellite orbits, it is possible to exploit the informative contribution carried by the phase difference between two SAR images looking at the same scene from comparable geometries.

AIM. In this setting, the main objective of this study is to evaluate the region near the closed rock salt mine in the south of Albania. Our input for this exercise will be two images of the land near the former rock salt mine in Dhrovjan near the Blue Eye (Saranda, Albania).

RESULTS. By combining the phases of 2 images we produce an interferogram where the phase is correlated to the terrain topography and deformation so if the phase shifts related to the topography are removed from the interferogram, the difference between the resulting products will show surface deformation patterns or cure between the two acquisition dates and this methodology is called differential interferometry Processing, Phase Unwrapping, and at the end creating the displacement map. We use in our study the difference in time with the algorithm which consists of working step by step with these operators: Read the two split products, Applying Orbit files, Back-Geocoding, Enhanced Spectral Diversity, Interferogram, TOPSAR Deburst, and Write. The resulting difference of phases is called an interferogram containing all the information on relative geometry. Removing the topographic and orbital contributions may reveal ground movements along the line of sight between the radar and the target.

The next algorithm we worked with these operators: Read the debursted interferogram, TopoPhaseRemoval, Multilook, Goldstain Filtering, and Write. At the same time from Goldstain Filtering, we add the Snaphu Export operator.

Correct phase unwrapping procedures must be performed to retrieve the absolute phase value by adding multiples of 2π phase values to each pixel to extract accurate information from the signal. In this study, we will use SNAPHU, which is a two-dimensional phase unwrapping algorithm consists of working step by step with these operators: read (the wrapped image) and read (2) the unwrapped image, Snaphu Import, PhaseToDisplacement, and Write. We can display it in Google Earth after saving it as .kmz and also make a profile of the displacements.

DISCUSSION AND CONCLUSIONS

One of the SAR Interferometry applications is deformation mapping and change detection. This work demonstrates the capability of interferometric processing for the observation and analysis of instant relative surface deformations in the radar LOS direction. When two observations are made from the same location in space but at different times, the interferometric phase is proportional to any change in the range of a surface feature directly. All three stages of the work are important and require accurate interpretation knowledge, especially when working with the Snaphu program.

KEY-WORDS

InSAR, DInSAR, Interferogram

How to cite: Belba, P.: The use of InSAR and DInSAR for detecting land subsidence in Albania, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11808, https://doi.org/10.5194/egusphere-egu25-11808, 2025.

Poincaré established a framework for understanding the dependence of a dynamical system's properties on its topology. Topological properties offer detailed insights into the fundamental mechanisms — stretching, squeezing, tearing, folding, and twisting — that govern the shaping of a dynamical system's flow in state space. These mechanisms serve as a conduit between the system's dynamics and its topology [Ghil & Sciamarella, NPG, 2023]. A topological analysis based on the templex approach [Charó et al., Chaos, 2022] involves finding a topological representation of the underlying structure of the flow by the construction of a cell complex that approximates its branched manifold and a directed graph on this complex. A pivotal feature of the cell complex that facilitates the characterization of the flow dynamics is the joining locus, upon which all the fundamental mechanisms that sculpt the flow leave a pronounced signature.

The local dimension d(x) and the inverse persistence θ(x) of the state x of a dynamical system [Lucarini et al., 2016; Faranda et al., Sci. Rep., 2017] provide information on the rarity and predictability of specific states, respectively. We demonstrate herein that these two measures, d and θ  also provide information about the localization of the joining locus.

The present work proposes a new topological method for fingerprinting a system’s nonlinear behavior using the concept of persistent generatexes. This novel approach integrates the strengths of two topological data analysis methods: the templex and persistent homologies. Rather than employing a single cell complex and a digraph to characterize the flow of the system, our approach emphasizes the localization of the joining locus through the calculation of local dimension and the inverse persistence, leading to the construction of a family of nested digraphs. The dynamical paths, namely the nonequivalent ways of travelling through the flow, are found to be the most persistent cycles; here the concept of persistence is used in the sense of the persistent homology approach [Edelsbrunner & Harer, Contemporary mathematics, 2008]. The dynamical paths give us the ‘topological fingerprinting’ of a system’s dynamics.

How to cite: Charó, G. D., Faranda, D., Ghil, M., and Sciamarella, D.: Topological fingerprinting of dynamical systems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12947, https://doi.org/10.5194/egusphere-egu25-12947, 2025.

The Er-Rich region is a focal area for understanding the geological evolution of the central-eastern High Atlas, which it covers almost entirely along a north-south transverse line. It is a hinge region between the two major tectonic structures of the High Atlas (the North and South Atlas faults), which reveal a framework of Meso-Cenozoic carbonate, detrital and magmatic rocks.

Previous studies have highlighted the complexity of mapping in this area. To date, no detailed geological map has been produced for this study area, with the exception of the old provisional 1:200,000 map of the Midelt-Rich High Atlas. Remotely-sensed mapping initiatives have also been carried out in the region, except that they do not provide a final interpretation as a geological map, supported by geological maps covering neighboring regions. A detailed geological map of the Er-Rich region, based on the results of remote sensing and field data, is therefore needed in the area. For this purpose, remote sensing geological mapping techniques have been applied to two types of satellite data: 1) Landsat 8 OLI (Optical Land Imager) multispectral optical data, and the Spot 5 panchromatic band acquired by the HRG-2 (High Geometric Resolution) instrument; 2) Sentinel-1 SAR data with dual polarisation (HV-HH).

All the data underwent several pre-processing or correction stages using appropriate software, in particular radiometric and atmospheric correction for Landsat 8 OLI (Optical Land Imager) images using ENVI software. The corrected product of the three Landsat 8 OLI scenes covering the region were then spatially enhanced using the Spot 5 panchromatic band to produce a multispectral image with a high spatial resolution of 5 m using ENVI software. The Sentinel-1 radar data were pre-processed using SNAP toolbox software by applying a series of corrections.

The results obtained by applying the Optimum Index Factor (OIF) method and Principal Component Analysis (PCA), allowing us to select the most significant colored compositions. Moreover, this combination enabled us to delineate with great precision the large outcrops of carbonate rocks (limestones, marl), siliciclastic rocks (conglomerates, sandstones and silts) and magmatic rocks (igneous intrusions).

The lineaments were extracted manually by visual interpretation of Sentinel-1 radar images, after applying directional filtering folowing four general orientations (N0, N45, N90, N135), enabling us to generate a synthetic structural map of the region.

The results obtained were compared with data from geological maps of adjacent areas and approved by field observations, leading to the production of a high-precision geological map, compiled with pre-existing geological literature.

How to cite: Hdoufane, M., Zafaty, O., and Ettaki, M.: Integrated remote sensing data and field investigations for geological mapping and structural analysis in the Er-Rich area (High Atlas, Morocco), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13782, https://doi.org/10.5194/egusphere-egu25-13782, 2025.

How to cite: Mendes, J. and Portier, M.: Uniform Data Access Layer: Advancing Data FAIRness in FAIR-EASE, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15569, https://doi.org/10.5194/egusphere-egu25-15569, 2025.

The impacts of global change, such as extreme heat and water scarcity, are increasingly threatening urban populations. Evapotranspiration (ET) plays a vital role in mitigating urban heat islands and reducing the effects of heat waves. It also serves as a proxy for vegetation water use, making it a critical tool for designing resilient green cities. Despite its importance, high-resolution mapping of urban ET that captures both spatial and temporal dynamics remains limited. This study focuses on the Paris metropolitan area, analyzing ET variability across multiple spatial scales (from 10 m to 10 km) using Sentinel-2 data from the Copernicus system. The Normalized Difference Vegetation Index (NDVI) is calculated with observation scale of 10 m, and then used as a proxy for ET. Universal Multifractal analysis, which have been widely used to characterize and model geophysical fields extremely variable across wide range of space-time scales, are implemented on this new data set. This framework is parsimonious since it basically relies on three parameters only: the mean intermittency codimension C₁, the multifractality index a and the non-conservation parameter H.  Specifically, the multifractality index α (1.3–1.5) and the mean intermittency codimension C₁ (~0.02) were derived to quantify the spatial and temporal heterogeneity of ET. The analysis, spanning 2019–2023, revealed noticeable temporal and spatial variability in ET. The study focuses on a square region of approximately 60 km × 60 km within the area around Paris. This region was further divided into multiple portions of size ranging from 2 to 10 km to assess potential variability over the studied areas. By incorporating both yearly and monthly data, the analysis captured seasonal trends as well as interannual variability, with higher variability observed during the summer months, driven by increased vegetation activity and water demand. Spatially, yearly data was analyzed and ET variability was most pronounced in densely populated areas, such as central Paris, where anthropogenic influences dominate. In contrast, forested areas and urban parks demonstrated significantly more stable ET patterns, underscoring the moderating effect of vegetation cover. These findings highlight the critical role of urban greening in mitigating extreme variability and stress on urban ecosystems.

How to cite: Zhu, S., Gires, A., Schertzer, D., Tchiguirinskaia, I., and Maksimovic, C.: Temporal and Spatial Dynamics of Urban Evapotranspiration in Paris: A Multiscale Perspective, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16203, https://doi.org/10.5194/egusphere-egu25-16203, 2025.

The ESA Directorate of Earth Observation Programmes has been actively leveraging satellite-based environmental information to address fragility contexts, focusing on areas such as environmental crimes, crimes against humanity, cross-border crimes, and onset of crises. Over the past decade, ESA has explored digital intelligence crime analysis by employing advanced data mining and machine learning tools to uncover hidden patterns and relationships in historical crime datasets, enabling better detection, prediction, and prevention of criminal activities.

Despite these advancements, the integration of Earth Observation (EO) capabilities into investigative practices remains limited. This is due to several challenges, including low awareness of EO's potential, a lack of illustrative use cases showcasing its benefits, inconsistencies in satellite data collection compared to investigative needs, high costs of very high-resolution imagery, and restricted access to national intelligence sources. To overcome these barriers, ESA has been investigating strategies to systematically incorporate EO-derived information into investigative frameworks also as legal evidence, aiming to enhance situational awareness and support stakeholders in developing procedures to exploit EO and OSINT for addressing international crimes and assessing fragility contexts, in cooperation with international organizations including Interpol, UNODC and ICC.

Recent developments in EO technology and methodologies have created significant opportunities for more impactful applications. ESA has focused on tailoring EO-based services and OSINT to meet the case-sensitive requirements of security and development end-users, enabling better integration of EO-derived insights into intelligence models. These efforts include developing advanced EO information products that go beyond routine offerings, testing and evaluating these products in collaboration with end-users, and demonstrating their value in operational settings.

The GDA Fragility, Conflict, and Security initiative has been a cornerstone of ESA’s work, involving partnerships with International Financial Institutions (IFIs) to co-design tools that provide precise and timely information. These tools have supported initiatives aimed at reducing inequalities, promoting economic development, and enhancing environmental safety in fragile and conflict or post conflict-affected areas. By combining geospatial data with diverse data sources, ESA has delivered customized analyses and reports to improve emerging threats analysis and decision-making processes.

Several ESA initiatives have demonstrated the benefits of EO services for assessing fragility risk exposure, characterizing dynamic needs in fragile contexts, planning post-conflict reconstruction, and managing natural resources. ESA constantly engages with stakeholders, including the OECD, security organizations, and humanitarian actors, and its community of industries and research centres to promote the adoption of EO in international development, humanitarian aid, and peacebuilding. Through these efforts, ESA continues to advance the role of EO in supporting justice, accountability, and sustainable recovery in fragile settings.

How to cite: Corvino, M. and the Michela Corvino: Leveraging EO for Security and Resilience, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17156, https://doi.org/10.5194/egusphere-egu25-17156, 2025.

The Canary Eddy Corridor is a dynamic region of mesoscale eddy activity, playing a critical role in the transport of physical properties (heat and salt) and biogeochemical properties (nutrients, larvae, plankton) in the eastern North Atlantic. This study investigates the Lagrangian evolution of the trapping capacity of mesoscale eddies according to their lifecycle phases and vertical structure (surface vs. subsurface eddies).

We combine OceanParcels (an open-source Python toolbox) and an eddy identification and tracking algorithm with the GLORYS12V1 reanalysis product and altimetry data from AVISO to simulate particle release and track trajectories within eddies. Applying the eddy tracking algorithm at surface and subsurface levels in GLORYS12V1 reveals that subsurface eddies with a surface signal exhibit subsurface rotational velocities at the eddy core that occasionally exceed those of surface eddy cores. This highlights the potential misrepresentation of eddy transport capacity when relying solely on altimetry data, without accounting for the vertical structure, which can be better resolved through a combination of model outputs and observational data, such as non-standard Argo float configurations. Furthermore, a detailed analysis of the eddy lifecycle phases shows that mature eddies exhibit substantially greater trapping depths compared to their growth and decay stages. These findings align with earlier modeling analyses of dipoles originating south of Madagascar, which also highlight enhanced trapping depths in mature eddies.

The results provide a comprehensive view of the trapping capacity of mesoscale eddies throughout their lifecycle and vertical structure, emphasizing their critical role in biophysical coupling, ecological connectivity, and the transport of biogeochemical properties, as well as microplastics and other pollutants.

Acknowledgments: The first author is grateful for the internship grants ERASMUS +, AMI-MESRI, and TIGER. 

How to cite: Vacca, D., Aguiar-González, B., and Morris, T.: Lagrangian Evolution of the Trapping Capacity of Mesoscale Eddies in the Canary Eddy Corridor: A Numerical Modeling Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17965, https://doi.org/10.5194/egusphere-egu25-17965, 2025.

The abandonment of agricultural land in India, especially paddy fields, has emerged as a significant challenge for food security and ecosystem sustainability in the country. Although rice production is vital for national food security, research on paddy land abandonment in India remains limited. Some Indian states have reported an alarming decline in paddy cultivation area over the past two decades. The study employs the Udupi district of Karnataka, India, a high-rainfall coastal region where paddy has traditionally been the dominant crop and where paddy land abandonment has been observed, as the study area. This study addresses crucial research gaps by framing these objectives for the study: (1) developing a deep learning framework that utilizes both intensity and phase information from polarimetric Synthetic Aperture Radar (SAR) data for abandoned paddy land detection, (2) leveraging recurrent neural networks (RNNs) to capture temporal patterns in abandonment, and (3) demonstrating an automated, all-weather monitoring approach that overcomes the limitations of traditional optical remote sensing in tropical regions.

Conventional monitoring approaches struggle with persistent cloud cover in tropical regions which limits effective assessment of abandonment patterns. SAR data provides unique capabilities for continuous monitoring under all weather conditions, making it particularly well-suited for tropical regions. However, previous studies have primarily underutilized SAR's potential by concentrating solely on backscattering intensity from ground range detected (GRD) products, overlooking the valuable phase information that could offer deeper insights into land use changes.  In this study, we employ Sentinel-1 Single Look Complex (SLC) data, which offers both intensity and phase information. Considering the temporal nature of paddy land abandonment, we developed a deep learning framework utilizing RNNs viz. LSTM, BiLSTM and BiGRU to effectively capture time-series patterns in the data. This framework analyzes backscattering coefficients (VV and VH polarizations) and polarimetric parameters (entropy, anisotropy and alpha angle) derived from SLC data collected during the Kharif seasons from 2017 to 2024. We carried out extensive ground truth data collection of active and abandoned paddy lands to train and validate our models. The backscattering coefficients were processed through orbit correction, radiometric calibration, TOPSAR deburst, multi-looking, speckle filtering and terrain correction. For deriving the polarimetric parameters, after basic preprocessing steps, the covariance matrix was generated followed by the polarimetric decomposition of the phase-preserved data. Results indicate that our RNN models show promising performance in detecting temporal patterns of paddy land abandonment. The method exhibits a robust ability to produce reliable abandoned land maps in regions prone to cloudy and rainy conditions. Future research should explore polarimetric features across various vegetation types in abandoned lands, expand the methodology to other agricultural systems, and examine the impact of socio-economic and topographical factors on abandonment patterns to support evidence-based land management policies.

How to cite: Kasture, S., Hegde A, A., and Umesh, P.: Deep Learning based Paddy Land Abandonment Detection Using Multitemporal Polarimetric SAR Patterns, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18606, https://doi.org/10.5194/egusphere-egu25-18606, 2025.

The presence of rain in wind farms involves several modeling challenges, as the momentum exchanges between turbulent wakes and the particle phase present subtle phenomena. For instance, rain droplets are typically large enough to exhibit inertia relative to the air carrier phase. Under these conditions, it has been found that the gravitational settling of particles in turbulent flows may be either enhanced or hindered compared to stagnant conditions. While this has significant implications for rainfall transport, ash pollutants, and pollen dispersion, very few studies have been conducted in field conditions. Moreover, the scaling laws and non-dimensional parameters governing this phenomenon have not yet been properly identified, and determining which configurations result in the enhancement or hindrance of settling velocity remains an open question.

We propose a hybrid experimental/numerical approach. Field data from a meteorological mast located at a wind farm in Pays d’Othe, 110 km South-East of Paris, France, were used to characterize the background turbulent flow through a set of sonic anemometers. Additionally, disdrometers were employed to characterize the settling velocity of raindrops, discriminating by particle size. Numerical simulations complement this data analysis. Specifically, 3D space and time vector fields that realistically reproduce the observed spatial and temporal variability of wind fields are generated using multifractal tools. Then, 3D trajectories of non-spherical particles are simulated and their settling velocity derived.

Our findings indicate that the presence of turbulence significantly hinders the settling velocity of raindrops in turbulent environments. Our study covers several distinct rainfall events, allowing us to analyze the influence of turbulent flow properties on this phenomenon.

How to cite: Obligado, M. and Gires, A.: Rainfall Dynamics in Wind Energy Scenarios, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19070, https://doi.org/10.5194/egusphere-egu25-19070, 2025.

Land degradation is a critical challenge to sustainable development, impacting ecosystems, economies, and communities globally. As part of the FAIR-EASE Earth Critical Zone (ECZ) pilot, this study develops a tailored Land Degradation Assessment tool based on the Trends.Earth approach. The tool aims to enhance data accessibility, integration, and usability across environmental domains, supporting decision-making and policy frameworks aligned with the United Nations Sustainable Development Goals (SDGs).
Building upon the robust Trends.Earth implementation, we can integrate customized workflows and datasets to reflect regional variability in degradation indicators, including vegetation productivity, soil health, and land cover changes. Our approach prioritizes FAIR (Findable, Accessible, Interoperable, and Reusable) principles to ensure broad usability and collaboration across scientific and policy communities.
Preliminary results demonstrate the tool's capacity to enhace the detail of the analysis and to identify degradation hotspots. Furthermore, the integration of open-source geospatial tools and standards supports a scalable framework applicable to diverse environmental contexts.
The tool is designed to be embedded within the LandSupport platform, a geospatial decision support system, further enhancing its accessibility and integration into decision-making processes for land management.
This work contributes to advancing interdomain digital services and illustrates the potential of FAIR principles in addressing complex environmental challenges. We invite feedback from the community to refine, expand and customise the tool's application, fostering collaboration for sustainable land management.

How to cite: Mauriello, I. E., Langella, G., Terribile, F., and Miralto, M.: Customizing Trends.Earth for land degradation assessment in the earth critical zone: a FAIR-EASE approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19225, https://doi.org/10.5194/egusphere-egu25-19225, 2025.

Over thirty years ago, Y. Kagan speculated that seismicity could fruitfully be considered as “the turbulence of solids”.  Indeed, fluid turbulence and seismicity have many common features: they are both highly nonlinear with huge numbers of degrees of freedom.  Beyond that, Kagan recognized that they are both riddled with scaling laws in space and in time as well as displaying power law extreme variability and – we could add – multifractal statistics.

Kagan was referring to seismicity as usually conceived, as a sudden rupture process  occurring over very short time periods.  We argue that even at one million year time scales, that the movement of tectonic plates is “quake-like” and is quantitatively close to seismicity, in spite of being caused by relatively smooth mantle convection. 

To demonstrate this, we develop a multifractal model grounded in convection theory and the analysis of the GPlates data-base of 1000 point trajectories over the last 200 Myrs.  We analyzed the statistics of the dynamically important vector velocity differences where Dr is the great circle distance between two points and Dt is the corresponding time lag.  The longitudinal and transverse velocity components were analysed separately.  The longitudinal scaling of the mean longitudinal difference follows the scaling law <Δv(Δr)> ≈ Δr^H with empirical H close to the mantle convection theory value  H = 1.  This high value implies that  mean fluctuations vary smoothly with distance.  Yet at the same time,  the intermittency exponent C₁ is extremely high (C₁ ≈ 0.55) implying that from time to time there are enormous “jumps” in velocity: “Plate quakes”.  For comparison, laminar (nonturbulent) flow has H = 1 but is not intermittent (C₁ = 0), whereas fully developed isotropic fluid turbulence has the (less smooth) value H = 1/3 (Kolmolgorov) but with non-negligible intermittency C₁ ≈ 0.07 and seismicity has very large C₁ ≈ 1.3.  Our study thus quantitatively shows how smooth fluid-like behaviour for the longitudinal velocity component can co-exist with highly intermittent quake-like behaviour.

Whereas the longitudinal component is well modelled by (highly intermittent) convection, the transverse velocity is well modelled by Brownian motion.  In the temporal domain both components (including their strong correlations) display such diffusion behaviour (i.e. with classical exponent H = ½), but are highly intermittent (C_1time = C_1space/2 ≈ 0.27).  Finally, the extreme velocity differences (that appear as occasional spikes in the velocities) have power law probability tails; the “Guttenberg-Richter” exponents in the seismology literature.

The advection - diffusion model is based on an underlying multifractal space-time cascade process.  Using mantle convection theory, we show how the driving multifractal flux (ψ) is related to vertical heat fluxes, expansion coefficients, densities, viscosities and specific heats. Taking typical values predict driving fluxes very close to the observed mean <ψ> ≈ 1/(400 Myrs).  Trace moment analysis shows that the outer space-time scales of the cascade process are ≈17000 km in space and ≈ 50Myrs in time.   Whereas the former corresponds to half the Earth’s circumference, the latter is the typical time required for a plate to randomly “walk” the same distance.

How to cite: Lovejoy, S., Spiridonov, A., and Balakauskas, L.: The turbulence of solids: a multifractal plate tectonic model with Guttenberg-Richter plate “quakes” , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20653, https://doi.org/10.5194/egusphere-egu25-20653, 2025.

Gold deposits in New Brunswick, part of the Canadian Appalachians, formed during various stages of the Appalachian orogeny. Significant regional-scale transcurrent faults that are locally controlling cogenetic magmatic, include the Restigouche, Rocky Brook-Millstream, McCormack-Ramsay Brook, McKenzie Gulch, and Moose Lake faults, played a crucial role in shaping the geological framework and enabling the focusing of mineralizing fluifds in northern New Brunswick. The mineral systems approach is applied here to link conceptual models of mineralization processes with available exploration data, aiming to achieve effective mineral prospectivity mapping (MPM). This method is designed to streamline exploration efforts, minimizing both time and cost, which are key priorities in the mineral exploration industry. A machine learning-based data-driven approach was utilized to evaluate 18 predictor maps with a pixel size of 200 meters. These MPM maps integrated diverse features, including geochemical indicators for Au, As, Sb, Zn, Pb, Cu, and Mo in till to define geochemical anomalies, airborne radiometric data for K, eU, and eTh, as well as aeromagnetic and LiDAR datasets, to interpret geological characteristics, structural features, faults, intrusive and extrusive units, and lithological contacts. A series of edge enhancement filters, including Reduced to Pole (RTP), first vertical derivative (FVD), tilt derivative (TDR), and analytic signal (AS), were applied to the dataset, followed by a 3D inversion. Our results show that bimodal felsic to mafic intrusive and extrusive igneous systems exhibit a strong magnetic response, a conclusion validated through correlation with drill core assay data. Moreover, this study utilized principal component analysis (PCA) of till data to determine pathfinder and indicator elements associated with gold mineralization. A MPM model was created for epithermal gold mineralization using a Support Vector Machine (SVM), incorporating the known gold occurrences and deposits of the area. The performance of the resulting MPM maps was evaluated using the area under the receiver operating characteristic curves (AUC-ROC). The study concludes that SVM is a robust tool for mineral exploration, providing a data-driven approach to identifying new mineral deposits with greater accuracy and efficiency.

How to cite: Mami khalifani, F., Lentz, D., and Walker, J.: A Machine Learning-Based Framework for Mineral Prospectivity Modeling: Predicting Epithermal Gold Mineralization in Northern New Brunswick, Canada, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1121, https://doi.org/10.5194/egusphere-egu25-1121, 2025.

Mongolian society and food production depends heavily on livestock farming, which is usually practiced with nomadic systems. Consequently, movement patterns of herders are crucial in respect of finding sufficient forage and sustainable use of pastures. In this study, a combination of InSAR, optical and weather time series data has been explored as a tool for spatio-temporal grazing monitoring. To detect movement patterns, a machine learning (ML) based method to detect breakpoints in vegetation condition has been developed and compared to the widely-used Breaks For Additive Season and Trend (BFAST) algorithm. The results have been validated using test sites spread across the entire eastern Mongolian steppe ecosystem, covering different grassland use intensities. The results indicate that (1) ML method performed superior compared to BFAST, detecting 41.5% of breakpoints. (2) Breakpoints in summer pastures mainly occurred from April to June, while on winter pastures, they emerged in October, November, and the following February and March. (3) Regarding spatial prediction, the model developed in this study predicts breakpoints in areas distinguish between summer and winter camps, However, there is insufficient data to conclusively attribute the occurrence of pasture breakpoints to herder movements.

How to cite: ji, S.: Can Vegetation Breakpoints in Eastern Mongolia grassland be detected using Sentinel-1 coherence time series data?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1321, https://doi.org/10.5194/egusphere-egu25-1321, 2025.

This study explores the application of score-based generative diffusion models for mapping sea surface Suspended Particulate Matter (SPM) of the Dutch Wadden Sea using satellite-derived images, focusing on their comparative efficacy against state of the art deterministic methods such as 4DVarNet, UNet, and DInEOF. Although deterministic deep learning approaches provide robust reconstructions, they often struggle with probabilistic uncertainty and extreme values of overly complex real-world scenarios. Our findings indicate that diffusion models, when conditioned with 4DVarNet and DInEOF, offer improved performance over DInEOF and UNet. Although slightly less accurate than 4DVarNet, this discrepancy is not a significant concern, as the primary goal extends beyond merely maintaining accurate reconstructions. Instead, our approach aims to provide a comprehensive view of the distribution through the samples. Our results show that diffusion models are able to generate the tail of the distribution, thereby capturing extreme values more effectively. And they assist in identifying areas of high uncertainty, particularly when the samples show inconsistencies. Furthermore, unlike typical 2D diffusion models, this study employs a 3D approach, incorporating 2D spatial and 1D temporal dimensions, allowing the model to capture dynamic physical changes over time and enhance the accuracy of probabilistic predictions of the image time series.

How to cite: Nguyen, T.-T.-N., Lakra, M., Jourdin, F., and Fablet, R.: Score-based Diffusion Models for the Space-Time Interpolation of Sea Surface Turbidity, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4051, https://doi.org/10.5194/egusphere-egu25-4051, 2025.

Geospatial monitoring of water quality is essential for managing and protecting groundwater resources, particularly in agricultural regions where nitrate contamination poses significant environmental and public health risks. This study presents a novel methodology for generating absence points in geospatial binary classifications applied to nitrate levels in groundwater across Odense, Denmark. We developed machine learning designed to generate absence points using multiple approaches for binary classification: random, buffer-based, similarity-based, and Maxent-based methods. The integration of maximum entropy into the absence generation workflow allowed us to identify low-susceptibility zones, improving the accuracy of binary classification. The dataset comprised geospatial nitrate concentration levels derived from environmental, hydrological, and anthropogenic variables. Spatial data included high-resolution land-use maps and hydrological parameters. Model evaluation was conducted using Random Forest, with results indicating that the Maxent-based approach consistently outperformed other methods across all metrics, including precision (0.96), AUC (0.96), and TSS (0.91). This method proved particularly effective in handling the challenges associated with presence-only data and produced the most reliable predictions for nitrate contamination in groundwater. The findings underscore the importance of leveraging advanced absence generation techniques to enhance model performance in geospatial classification modeling.

How to cite: Naghibi, S. A., Ahmadi, K., and Berndtsson, R.: Bridging Data Gaps in Water Quality Modelling: A Machine Learning Framework for Absence Point Generation in Geospatial Binary Classifications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6262, https://doi.org/10.5194/egusphere-egu25-6262, 2025.

Deep learning approaches for geological subsurface reconstruction typically require extensive training datasets, limiting their practical application in geosciences where data acquisition is costly and sparse. We present a methodology using sparse convolutional autoencoders that effectively learns from synthetically generated training data while maintaining strong generalization to real-world scenarios. Our model is trained exclusively on synthetic basin boundary configurations and corresponding forward-modeled Vertical Electrical Sounding (VES) responses, thereby eliminating reliance on extensive real-world training datasets. Through transfer learning, the model achieves high reconstruction accuracy with as few as 1000 synthetic training examples. Systematic tests reveal the model preserves strong performance beyond its training distribution, suggesting it learns robust heuristic approximations and remains effective beyond the training range of 3–50 input points.

The trained model was applied to the Huancayo tectonic basin in the Peruvian Andes. There, the 300 to 350-m deep subsurface geometry of the tectonic basin was sucessfully modeled on basis of data input from 41, newly acquired VES logs along two cross sections of 12- and 14-km long.  Surprisingly, the reconstruction also revealed previously unidentified fold and thrust systems, for which the model was not explicitely trained, while also maintaining physical consistency with field measurements.

Our results demonstrate that sparse convolutional autoencoders, when trained on synthetic datasets, can effectively bridge the gap between data-hungry deep learning methods and data-sparse geological applications.

How to cite: Uribe-Ventura, R., Barriga-Berrios, Y., Barriga-Gamarra, J., Baby, P., and Viveen, W.: A new data-efficient, deep learning-based methodology for geological subsurface reconstructions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6404, https://doi.org/10.5194/egusphere-egu25-6404, 2025.

How to cite: Kanevski, M.: Environmental Data Exploration and Modelling Using Extreme Learning Machine, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7462, https://doi.org/10.5194/egusphere-egu25-7462, 2025.

As Artificial Intelligence (AI) and Machine Learning (ML) methods evolve at an explosive pace, there is an increased need to handle geological data challenges if we wish to (continue to) ride the AI/ML wave. Most geological data is at its core geospatial data in different shapes and formats. A few examples are polygon-based geological maps, geophysical and remote sensing raster grids and a plethora of sample analyses with coordinate data. Applications of geological data are as varied as the Earth is vast. However, these differing geospatial data formats in geology significantly limit the interoperability of datasets in an analytical ML context. Multivariable analyses of geological data often involve extensive spatial interpolation and projection headaches. Consequently, we must first solve geospatial data challenges to fully tackle inter- and intradisciplinary geoscience problems with ML and AI predictions on multivariable cross-disciplinary geological data. At our company, Scandinavian Highlands, we are building a platform and database structure to break down these geospatial format barriers using a hexagonal discrete global grid system called H3. The H3 grid represents all positions on Earth’s surface by hexagon (and 12 pentagon) cells at different levels of coarseness, ie. resolutions, down to 1 m² cell area. The resolutions are bound together by a systematic parent cell to children cells hierarchy. We process different types of geospatial geological data (raster and vector) to an H3 grid representation at the appropriate resolution for the given dataset. In doing this, we create a database structure where different geological data layers can be seamlessly merged into a single feature “stack” table for AI/ML purposes at either local, regional or global scales and across individual dataset resolutions. In this presentation, we demonstrate the hexagonal multiple feature stack concept in action, from simple grouped/filtered visualisation, regression and descriptive statistics to dimension reduction techniques (e.g. PCA and t-SNE), clustering and other supervised and unsupervised methods. Furthermore, all analysis results can be assessed spatially on the map, grounding them on the Earth’s surface and in real-life decision-making use cases.

How to cite: Traun, M. K., Sandø, F., and Jensen, S. L.: Tackling Spatial Multiple Features AI/ML Problems in Geology with Hexagons, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11150, https://doi.org/10.5194/egusphere-egu25-11150, 2025.

Effective land monitoring and land use classification are critical for proper management of resources especially in heterogeneous and climate diverse areas. Consequently, this study seeks to test the hypothesis that the integration of Sentinel-1 radar and Sentinel-2 optical data enhances the degree of discrimination of crops in major farming areas of Morocco from the years 2020 to 2022. A three-dimensional coordinate system was established which included a series of processing stages that started with cloud masking, scaling of reflectance, and radar optical integration. At each year’s end, temporal averages and composites were created using selected Sentinel-2 spectral bands B2, B3, B4, B8, B11, B12 and Sentinel-1 VV & VH dual polarization channels. Ground truth samples from four major crops; Baley, Crop, D. Wheat and S. Wheat were used as the training set in a Random Forest classifier. The results for the three agricultural zones indicated high overall accuracies greater than 80% for each year, with the application of a combination of radar and optical data sets contributing greatly towards the ability to differentiate the crops located in cloud folded and spectral overlapping areas. Many classes had high consumer accuracy (≥70%) levels, yet several crops, like D. Wheat, had poor producer accuracy, possibly due to the uneven distribution of ground truth data sets. The small amount of Kappa coefficients between 0.50 and 0.60 also indicate moderate agreement similar to the validation data and thus more accurate ground truth and class targeted feature detection is needed. This study emphasizes the relevance notes of the multi-sensor data fusion technology for crop monitoring and also landcover classification which contributes to precision farming and resources management. Future work will focus on including temporal characteristics as well as state-of-the-art machine learning techniques to solve class balance issues and improve classification performance.

How to cite: choukri, M., laamrani, A., and chehbouni, A.: Optimal Use of Multi-Sensor Data for Precision Agriculture: Sentinel-1 and Sentinel-2 Fusion in Crop Classification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12795, https://doi.org/10.5194/egusphere-egu25-12795, 2025.

Spatial autocorrelation, the relationship between nearby samples of a spatial random variable, is often overlooked in machine learning models, leading to biased results. We investigated various methods to account for, address, and integrate spatial autocorrelation for modelling and prediction of soil organic carbon (SOC) using random forest models. We created and evaluated five different RF models to incorporate spatial structure through methods like buffer distances, KNN/RFSI coordinates, GWRFR, and kriging/RFRK. These were compared against a baseline models that did not have any added spatial components. Cross-validation showed slight improvements in accuracy for models considering spatial autocorrelation, while Shapley Additive Explanations confirmed the importance of spatial variables. However, no decrease in spatial autocorrelation of residuals was observed. The raster-based models exhibited enhanced prediction detail, but high-resolution validation data availability limited thorough validation. The findings emphasize the value of incorporating spatial autocorrelation for improved SOC prediction in machine learning models. We applied the models to predict SOC for the whole of Estonia in 10m raster resolution. Computational differences provided additional insights into pragmatic choices of models.

How to cite: Kmoch, A., Choi, J., Harrison, C. T., and Uuemaa, E.: Spatial autocorrelation in machine learning for modelling soil organic carbon, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16038, https://doi.org/10.5194/egusphere-egu25-16038, 2025.

JupyterLab is a web-based interactive development platform that is widely used in the Earth science community. Using Jupyter Notebooks, it is possible to perform data analysis tasks, annotate and visualize results in a way that is easy to reproduce, present and share with others. JupyterLab allows the use of “extensions”, which add functionality to the platform. One of these is Jupyter-AI [1], which allows the use of Large Language Models (LLMs), such as ChatGPT, Claude Sonnet and Ollama, within the JupyterLab environment, through  a chat interface or directly within notebooks. By integrating LLMs into JupyterLab, it is possible to leverage their code generation capabilities to assist a user to translate their analysis tasks from an idea to actual executable code in an efficient manner. One drawback of using these LLMs in tasks involving spatio–temporal data is that the models typically do not have access to the data necessary for the analysis task and will often resort to generating fictional data or using placeholders in the code that they create. This requires the user to adapt the provided code to their data, which removes some of the utility provided by the LLM.

In this context we make use of FrevaGPT, an approach for using LLMs in climate data analysis that allows for quick, complex and reproducible analyses of data sets, such as decadal climate model forecasts. Leveraging LLM’s capability to write code and using few-shot prompting (in-context learning) allows the LLM to utilize Freva [2,3] (Free Evaluation Framework), a data search and analysis platform, which provides a standardised interface to spatio-temporal datasets hosted on an HPC cluster [4].  

FrevaGPT integrates seamlessly into Jupyter-AI and, by making use of the Freva library, combines the code-generating capabilities of LLMs with contextual understanding of how to access relevant datasets on the HPC cluster. This in addition with FrevaGPT’s ability to execute generated code in an isolated environment on an HPC node, annotating and explaining any intermediate results, as well as automatically correcting errors encountered along the way, could serve as a starting ramp for researchers to efficiently produce new analysis products based on spatio-temporal climate data. 

This PICO will include examples of using FrevaGPT within JupyterLab to analyse spatio-temporal datasets from the climate of the past, as well seasonal to decadal climate predictions.

References:

[1] Jupyter-AI GitHub Repository: https://github.com/jupyterlab/jupyter-ai
[2] Kadow, Christopher, Sebastian Illing, Etor E. Lucio-Eceiza, Martin Bergemann, Mahesh Ramadoss, Philipp S. Sommer, Oliver Kunst, et al.. 2021. “Introduction to Freva – A Free Evaluation System Framework for Earth System Modeling”. Journal of Open Research Software 9 (1): 13. https://doi.org/10.5334/jors.253.
[3] Freva GitHub Repository: https://github.com/FREVA-CLINT/freva
[4] Public Freva Instance: https://www.freva.dkrz.de/

How to cite: Oertel, F., Lucio Eceiza, E., Willmann, S., Wentzel, B., Bergemann, M., and Kadow, C.: An AI-Copilot for JupyterLab for climate data analyses using FrevaGPT, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16905, https://doi.org/10.5194/egusphere-egu25-16905, 2025.

Aerial imagery has emerged as a powerful tool for environmental analysis and decision-making, enabling us to gain valuable insights. We present a comprehensive approach for performing semantic segmentation on aerial images of illegally dumped construction waste. We focus on the detection and analysis of the waste content to utilize it for circular economy. Leveraging the Segment Anything Model (SAM) developed by Meta, we produced highly accurate masks from aerial drone images. We created a dataset of over 46,000 manually labeled masks, which serve as ground truth for training and evaluation. Then we fine-tuned the ResNet-50 classification model together with the deep learning model. Our methodology combines the prediction of the classification model with these detailed masks to produce the final waste stream map. The map offers a comprehensive understanding of the open area allowing for further potential stocks analysis and economic evaluation. Overall, we achieved 86% detection accuracy on our full dataset, where for common classes the accuracy is higher. The waste identification can be used for economic and environmental decisions-making necessity of cleanup operations. The results also allow better planning of potential untapped stocks and treatment of different waste streams, aiding in local circular economy and waste management strategies. Our model development can serve the waste management and recycling sectors as well as municipal and national policy makers. 

How to cite: Mager, A., Blass, V., Gorun, A., Tsur, Y., and Shahar, M.: Leveraging AI for Material Identification in Unauthorized Dumps for Circular Economy Applications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18720, https://doi.org/10.5194/egusphere-egu25-18720, 2025.

How to cite: Meuer, J., Witte, M., Timmreck, C., and Kadow, C.: Efficient Large Ensemble Generation of Climate Model Output Using Latent Diffusion and Spatio-Temporal Transformers, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19037, https://doi.org/10.5194/egusphere-egu25-19037, 2025.

How to cite: Bueechi, E., Reuß, F., Pikl, M., Lukas, V., Trnka, M., Homolova, L., and Dorigo, W.: From a regional to field scale - transfer learning for Earth observation based crop yield forecasts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19281, https://doi.org/10.5194/egusphere-egu25-19281, 2025.

Agriculture, a cornerstone of human civilization, exerts a significant impact on natural resources to fulfill societal food demands. Climate change, exacerbated by anthropogenic activities and environmental consequences, poses a critical threat to agricultural productivity. While modern agronomic practices have enhanced yields, they have also resulted in detrimental consequences such as habitat loss, reduced biodiversity, and resource depletion.

This study investigates crop suitability in Hassan District, India, by integrating Artificial Intelligence (AI) with Geographic Information Systems (GIS). Eight key geo-climatic and pedological factors, relatively stable over time, were considered. Determining optimal land use for targeted crop cultivation is crucial in the face of climate change and global food security concerns.

Geospatial technologies and Sequential AI have demonstrated significant potential in addressing agricultural and environmental challenges through data-driven approaches. This research assesses the suitability of land for paddy, maize, and gram cultivation during the kharif season in Hassan District. A weighted metric approach was employed within a GIS environment, utilizing a Sequential Artificial Neural Network (ANN) model. Initially, an equal-weighted arithmetic mean was used to evaluate seven criteria encompassing soil, climate, and topographic factors. Likewise, criterion weights were derived from a sequential regression model, reflecting their relative importance in crop suitability prediction.Slope, soil depth, and rainfall emerged as the most influential factors, collectively accounting for 76% of the total weight. The results demonstrated an improvement in site suitability assessment compared to conventional methods, highlighting the efficacy of this integrated AI-GIS approach.

How to cite: Shivamurthy, V., Palat Ebrahim, M., and M Betegeri, V.: Geo Intelligence: A Key to Sustainable Paddy and Maize Production in Hassan District, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19971, https://doi.org/10.5194/egusphere-egu25-19971, 2025.

This study presents a comprehensive exploration of the collection and analysis of diverse geological and geophysical datasets from the eastern sector of the Russian Arctic. By leveraging advanced machine learning (ML) techniques, including convolutional neural networks, decision trees, and classical regression models, we provide insights into both data acquisition—encompassing geological, gravimetric, magnetic, and other parameters—and the subsequent analysis and interpretation of these data.

The research is structured around three primary objectives:

• Data Collection and Structuring: A systematic approach to the acquisition and organization of information on the geological and geophysical conditions in the eastern Russian Arctic.
• Application of Machine Learning Techniques: Employing cutting-edge ML methods to analyze and interpret the collected datasets.
• Findings and Practical Implications: Highlighting key results and conclusions, with an emphasis on their practical applications in Arctic geological and geophysical research.

This work aims to introduce conference participants to innovative ML methodologies in geophysical data analysis and emphasizes the significance of employing diverse approaches to enhance understanding and application. The study also underscores the broader potential of these methods for application in other regions and global-scale research.

How to cite: Lisenkov, I. and Soloviev, A.: Analysis of Geological and Geophysical Data in the Eastern Russian Arctic Using Machine Learning Techniques, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20280, https://doi.org/10.5194/egusphere-egu25-20280, 2025.

Temperature plays a critical role in climate systems and resource management. Understanding spatiotemporal evolution of the temperature is vital for effective climate adaptation and resource management. Traditional models often treat spatial and temporal aspects separately, limiting their ability to capture the full correlation between these dimensions. This study evaluates various time series and machine learning models, including Holt-Winters, SARIMA, TSLM, NNAR, and ANN, using a daily dataset from 30 meteorological stations in Apulia region (Italy) from 1982 to 2023. These models are assessed based on RMSE and MAE metrics. The best models are then integrated with spatiotemporal kriging of the residual data, with results showing that the hybrid approach outperforms traditional methods. This generated high-resolution predictive maps provide valuable insights into temperature trends, supporting better decision-making in agriculture, water management, and climate resilience.

Funding information

Financial support from ICSC–National Research Center in High Performance Computing, Big Data and Quantum Computing, funded by European Union–NextGenerationEU”
Project name: PNRR-HPC; Project code: CN00000013; CUP: C83C22000560007. 

How to cite: Iqbal, N., De Iaco, S., and Palma, M.: Integrating Machine Learning and Time Series Models for Spatiotemporal Temperature Prediction: A Case Study from Apulia, Italy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20435, https://doi.org/10.5194/egusphere-egu25-20435, 2025.

Integrating fine-scale measurements with broad-scale monitoring is a challenge for environmental monitoring, but it is a critical advancement in the face of increasing climate variability. We addressed this challenge by integrating fine-scale measures from Unoccupied Aerial Systems (UAS) to train broad-scale satellite imagery via machine learning algorithms. We applied this integration to detect how the spatial patchiness of bare ground varies over five years across a 100 km² semi-arid landscape in southern Arizona, USA. We used the Largest Patch Index (LPI) as the measure of spatial patchiness of bare ground. Our findings reveal three key advances in monitoring spatial patchiness over time and across a large landscape. First, the UAS-trained satellite estimates of LPI effectively represented the expected bare ground response to extreme climate events, where LPI increased during severe drought (-2.47 Standardized Precipitation-Evapotranspiration Index (SPEI)) and LPI decreased during exceptional wet periods (+1.95 SPEI). Second, the estimates of LPI were consistently 30-60% greater at lower and drier elevations, validating the ability to represent known ecological gradients. Third, and most notably, we confirmed that LPI is a scale-sensitive measure that differs between 3-m and 30-m grids, and that the magnitude of the differences is inversely related to the density of data in the satellite imagery. LPI was greatest using the 30-m grid Landsat 8 data with a density of 0.02 B/m² and LPI was least when using the 3-m grid PlanetScope data with a density of 0.9 B/m². But we found intermediate LPI values when resampling PlanetScope to 30-m grid while maintaining the greater data density. This previously unrecognized role of data density enriches the understanding of scale effects in landscape pattern analysis. In the end, we demonstrated a practical solution for integrating fine-scale UAS and broad-scale satellite observations via machine learning to support broad-scale environmental monitoring.

How to cite: Ponce-Campos, G. E., Heilman, P., Norton, C. L., Gao, S., Crimmins, M. A., and McClaran, M. P.: Integrating Unoccupied Aerial Systems and Satellite Data to Map the Patchiness of Bare Ground at the Landscape Scale, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-46, https://doi.org/10.5194/egusphere-egu25-46, 2025.

Traditional approaches for classifying hydrological surfaces - water, turbid water, salt pan, snow, and ice - usually rely only on one remote sensing dataset (often optical data like Sentinel-2). They face limitations under cloud cover areas and often confuse similar surface types (snow & salt pan, water & shadows). To overcome this, the study explores the use of Convolutional Neural Networks that can integrate spatial context, trained with multiple data sources like SAR (e.g., Sentinel-1), optical imagery, and exogenous inputs (weather, elevation). 

Deep Neural Networks are well-suited for texture extraction in remote sensing imagery and can efficiently handle inputs with multiple spectral bands. However, processing data from various sensor modalities introduces the challenge of aligning these inputs within a shared feature space where correlations can be effectively captured. To address this challenge, we developed a classical encoder-decoder architecture and explored the use of multiple encoders feeding into a single shared decoder. Two types of encoder families – EfficientNet and Swin Transformer – and two types of decoders – UNET and FPN – alongside various fusion methods were tried and showed similar performances.

For this study, a global multimodal database was gathered using open-source data from the Copernicus program. Initial trials with 17 labelled scenes (50GB) showed poor generalisation capabilities, leading to the extension of the dataset to 57 different scenes worldwide. Additional products were integrated, including Sentinel-1 (GRD VV+VH) data, 30m digital elevation models (ASTER GDEM), and meteorological data (from the ECMWF) to build the final 350GB database. Segmentation masks were generated semi-automatically (using a first version of our DL network) and then refined through visual inspection of Sentinel-2 images.

Results showed improved classification performance for all target classes when elevation data was included, and a dedicated dual-encoder-decoder model architecture proved particularly effective. On the other side, the integration of Sentinel-1 SAR data did not improve performance, likely due to the low temporal correlation between Sentinel-1 and Sentinel-2 acquisition (3-days average). Similarly, adding meteorological information did not enhance results, as our experiments showed that the model consistently disregarded scalar inputs regardless of integration approach.

Our model demonstrated notable robustness on the global database and was compared to existing CNES classification chains, including SurfWater (surface water detection) and Let-It-Snow (snow segmentation in mountains). Classification performance was comparable to SurfWater, though snow classification showed limitations in comparison to Let-It-Snow, particularly in the French Pyrenees.

The findings from this study underscore the potential of a multimodal approach in improving hydrological surface classification, particularly by incorporating data such as elevation. Future work could focus on increasing the volume of labelled data used to train the network to further enhance the model’s global applicability and precision across varied geographic and climatic conditions. Additionally, to fully leverage SAR imagery, reworking the database with more precise, directly annotated products would be essential. Finally, other approaches have to be tried in order to take into account meteorological data, for example using seasonality or more complex inputs. Other exogenous data could be added like terrain shadows.

How to cite: Eynard-Bontemps, G., May, S., Derksen, D., Dublé, N., Coquard, P.-J., and Audenino, P.: Hydrological surfaces classification with Deep Learning using multiple sensors and exogeneous data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1195, https://doi.org/10.5194/egusphere-egu25-1195, 2025.

The rapid expansion of Plastic-Mulched Landcover (PML)，characterized by its relatively small size and short lifespan, necessitates precisely mapping PML using High-Resolution Remote Sensing Imagery (HRRSI). However, the high costs and limited temporal resolution of acquiring HRRSI pose significant challenges for precise PML identification. Remote Sensing Image Super-Resolution (RSISR) offers a viable solution by reconstructing high-resolution images from lower-resolution inputs, enhancing PML detection capabilities. This study, based on hybrid attention transformer, develops a Multi-Scale Gated Feedforward Attention Network (MSG-FAN) for super-resolution reconstruction of Sentinel-2 data to meter-level resolution. The main contributions include: (1) Construction of a PML RS dataset comprising 5300 pairs of 10-m Sentinel-2 and corresponding 2.5-m Gaofen-2 and Planetscope images from eight globally selected plastic-mulched planting regions. (2) Development of the MSG-FAN model, which enhances multi-channel, multi-scale and global attentions by integrating Gated Multi-Scale Feedforward Layer (GMS-FL), Top-k Token Selective Attention (TTSA) module and Global Context Attention (GCA) module. (3) Demonstration that MSG-FAN outperforms nine state-of-the-art deep learning-based super-resolution networks, achieving an average PSNR 30.81 and SSIM of 0.7287. Our proposed MSG-FAN model advances RSISR techniques and addresses critical challenges in monitoring plastic-mulched planting regions.

How to cite: Lu, L. and Du, Y.: A Multi-Scale Gated Feedforward Attention Network for Super-Resolution Reconstruction of Remote Sensing Images in Plastic-Mulched Planting Regions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2475, https://doi.org/10.5194/egusphere-egu25-2475, 2025.

Learning rich and robust representations of Earth Observation (EO) data is critical for effective and accessible geoanalytics. While the ever-growing volume of EO data suggests high potential for self-supervised learning, most approaches are limited to fixed scales, resolutions, or modalities—thus failing to generalize beyond their original sensor configurations. To address these shortcomings, we introduce AnySat, a novel multimodal framework capable of self-supervised training on multiple, diverse EO datasets simultaneously.

AnySat’s design centers on two key innovations. First, we propose a Joint Embedding Predictive Architecture (JEPA) adapted for multimodal EO. Unlike pixel-level reconstruction methods, JEPA operates in latent space—making it inherently more resilient to cloud cover, time-of-day shifts, and varying acquisition angles. Second, scale-adaptive spatial encoders allow a single network to handle variable spatial and temporal resolutions. Notably, more than 75% of AnySat’s 100M parameters are shared across all supported modalities, scales, and resolutions, enabling the model to fully exploit diverse training corpora—a fundamental requirement for developing a true EO foundation model.

To train AnySat, we compile GeoPlex, a collection of five multimodal datasets (PASTIS-HD, TreeSatAI-TS, PLANTED, FLAIR, and S2NAIP), aiming for diversity: 11 distinct sensors including radar and optical modalities, 0.2–250 m resolution, single-image and time series, and 0.3-2600 ha per input sample. Thanks to its versatility, a single Anysat model can learn powerful representations by training from all five datasets simultaneously. We only use cross-modal alignment as a source of self-supervision, and do not require labels for pretraining.

We fine-tune and evaluate our model on the datasets of GeoPlex, as well as four external datasets to evaluate generalization.  We report state-of-the-art results on seven downstream tasks, including land cover mapping, crop-type classification, tree-species identification, deforestation detection, and disaster mapping. Notably, AnySat yields significant performance gains across multiple benchmarks, such as +2.8 mIoU on PASTIS-HD, +3.6 mIoU on SICKLE, +11.0 accuracy on TimeSen2Crop, and +10.2 IoU on BraDD-S1TS.

A major benefit of AnySat is its high performance when performing linear probing with fixed representations—even for semantic segmentation tasks. This combination of versatility, generalizability, and ease of use positions AnySat as a valuable tool for practitioners facing diverse sensor types, specialized data distributions, and limited annotations.

How to cite: Astruc, G., Gonthier, N., Mallet, C., and Landrieu, L.: AnySat: a Multi-Resolution/Modality/Scale Earth Observation Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4523, https://doi.org/10.5194/egusphere-egu25-4523, 2025.

How to cite: Fauzan, M. A., Virro, H., and Uuemaa, E.: Deep learning for detecting landscape features in agricultural lands, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5955, https://doi.org/10.5194/egusphere-egu25-5955, 2025.

Groundwater quality is a critical concern in agricultural regions, where nitrate contamination poses environmental and health risks, and ammonium levels play a pivotal role in nitrogen cycling processes. This study introduces a multi-task learning (MTL) framework designed to jointly predict nitrate and ammonium levels in groundwater, addressing the interdependencies between these variables. Conducted in Odense, Denmark, the study leverages spatial and temporal data, including hydrological, environmental, and anthropogenic variables, alongside land-use maps. The MTL approach outperforms traditional single-task models by capturing shared environmental and hydrological variables. By sharing information across tasks, the model identifies overlapping spatial, enabling robust predictions even in data-scarce scenarios. Additionally, the shared layers of the MTL model reduce overfitting, improving generalizability and providing deeper insights into the drivers of groundwater quality. The dataset used in this study includes geospatial nitrate and ammonium measurements, which were modeled alongside predictor variables such as land use, soil characteristics, and topographical variables. Model evaluation metrics demonstrated the superiority of the MTL approach, with increased accuracy, R², and reduced root mean squared error (RMSE) compared to separate models. The results highlight the potential of MTL to improve predictions and foster integrated groundwater management strategies. This study underscores the importance of advanced machine learning techniques in environmental modeling, showcasing a novel approach to jointly predict interrelated water quality variables.

How to cite: Ahmadi, K. and Naghibi, A.: Integrated Multi-Task Learning Framework for Groundwater Nitrate and Ammonium Prediction in Odense, Denmark, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6248, https://doi.org/10.5194/egusphere-egu25-6248, 2025.

How to cite: Li, Z. and Chen, B.:  Mapping Nationwide Essential Urban Land Use Categories by Integrating Multimodal Deep Learning and Multi-source Geospatial Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6757, https://doi.org/10.5194/egusphere-egu25-6757, 2025.

For the analysis of the evolution of landscapes, it is required to determine not only current states of the Earth’s surface, but to gain knowledge about past states, too. Sources of information for historic land cover are historic remote sensing imagery, and scanned historic topographic maps. To make the contained information explicitly available for subsequent computer-aided spatio-temporal analysis, classification techniques can be exploited. Against this background, multi-modal land cover classification from maps and aerial orthoimages is developed in the context of the Gauss Center (Gauss Centre, 2025), aiming to benefit from both the textural and geometrical details contained in aerial images, as well as the small intra-class variability in topographic maps.

The proposed deep learning-based classifier is a variant of a UPerNet (Xiao et al., 2018) with four down-sampling stages and takes aerial orthoimagery and topographic maps of the same epoch as an input. Each input modality is processed by an individual encoder, either based on convolutions, e.g. a ResNet (He et al., 2016), or on attention mechanisms, e.g. a Swin Transformer (Liu et al., 2022). This results in uni-modal map features and uni-modal aerial image features at four levels of detail. As the aerial images provide finer details about the texture and boundaries of the land cover objects, the aerial features of the first three stages are directly presented to the decoder, while the highest level aerial image features are fused with those of the topographic maps in a mid-level fusion. To focus on the most relevant features of the two modalities in spatial and feature dimension both, locally and globally, features are weighted by attention weights that are learned following the strategy in (Song et al., 2022). The lower-level aerial features and the high-level multi-modal features are presented to the decoder to predict to multi-modal land cover.

Experiments are conducted on two multi-modal datasets; one for binary building classification and one for multi-class vegetation classification. Both datasets consist of pixel-aligned aerial orthoimages, topographic maps and reference data at a ground sampling distance of 1 m. For all experiments, weights obtained in a pre-training on ImageNet (Russakovsky et al., 2015) are selected for the two encoder branches, while all remaining network weights are randomly initialized based on variance scaling (He et al., 2015). Training is proceeded utilizing the ADAM optimizer (Kingma & Ba, 2015) with standard parameters and a learning rate of 10-2 until the validation F1-score does not improve for 30 epochs. For both datasets, multi-modal predictions are compared to uni-modal predictions. Furthermore, attention-based feature extraction is compared to the one based on convolutions. The achieved mean F1-socres are the highest for the multi-modal variants of the classifier, where a higher score of 90.1% can be achieved utilizing convolutions on the building dataset (multi-modal, attention: 86.9%; aerial, convolution: 89.2%; map, convolution: 84.6%) and attentions are to be preferred for vegetation classification, resulting in a mean F1-score of 83.0% (multi-modal, convolution: 82.2%; aerial, attention: 82.1%; map, attention: 54.0%).

How to cite: Dorozynski, M., Rottensteiner, F., Dahms, T., and Hovenbitzer, M.: Leveraging multi-modal classification of historical aerial images and topographic maps to derive past land cover, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8714, https://doi.org/10.5194/egusphere-egu25-8714, 2025.

The integration of remote sensing and deep learning has revolutionized environmental monitoring, leveraging cutting-edge technologies to assist the decision-making processes in resource management and offering advanced tools for rapid disaster response. Our work employs satellite imagery to address pressing challenges in Earth observation, integrating multi-sensor, multi-resolution, and multi-temporal data for studying the aftermath of disastrous events by means of deep learning models, capable of handling such diverse data modalities.

We focused on the segmentation of wildfire-affected areas, using multispectral images from the Sentinel-2 satellites combined with the information from the Copernicus Emergency Management Service, in particular the geolocation and impact assessments, for more that 100 events occurred mostly in the European Mediterranean region. This dataset is further enriched with the observations from the Sentinel-1 and Sentinel-3 satellites, ensuring a comprehensive representation of the effects of each wildfire event by integrating measurements from multiple sensors with varying resolutions and revisit time. To streamline the workflow, a custom library based on the SentinelHub API has been developed, facilitating the download, preprocessing, and combination of data from different sources.

The study is performed on time-series of images, incorporating pre-event and post-event data, processed with a deep learning approach that combines Convolutional Long Short-Term Memory (ConvLSTM) layers in a UNet-like architecture. The results demonstrate the effectiveness of our model in accurately segmenting the affected areas, thus providing actionable insights for emergency management and recovery. Furthermore, the varied dataset, which comprises wildfire events occurring in diverse geographical conditions, enhances the robustness and generalizability of the described methodology.

This work is supported by ICSC – Centro Nazionale di Ricerca in High Performance Computing, Big Data and Quantum Computing, funded by European Union – NextGenerationEU, and it has been carried out within the Spoke 2 (“Fundamental Research and Space Economy”) as part of the activities in the Working Group 6 (“Cross-Domain Initiatives and Space Economy”) under the flagship use-case “AI algorithms for (satellite) imaging reconstruction”.

How to cite: Anastasi, G. A., Piparo, G., and Tricomi, A. R.: Advancing environmental monitoring through deep learning: wildfire segmentation using time-series of images from the Sentinel constellation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9376, https://doi.org/10.5194/egusphere-egu25-9376, 2025.

Many applications in Earth system sciences require continuous, gap-free data sets. However, remote sensing data in particular are plagued by gaps due to clouds, incomplete coverage, or low-quality flags. Gap-Filling in remote sensing data often requires model architectures that are tailored specifically to underlying dataset characteristics such as scale, resolution or range of values. This limits the transferability to other gap-filling scenarios. Training these models is further hindered by the lack of adequate training samples, as they must be gathered from gap-afflicted data themselves. In this work, we present a spatiotemporal, univariate and multiscale gap-filling method that is independent of any specific dataset. A modular implementation allows for the customization of system parameters, so that the method can be adjusted and applied to various datasets, even outside the Earth Science domain. By employing a patch-wise gap-filling approach, introducing masked loss functions, and providing effective methods for synthetic gap generation, we are able to leverage gap-afflicted datasets and gather large amounts of training samples from them. To demonstrate the flexibility of the system, we perform gap-filling on multiple climatic variables from Earth System Data Cubes (ESDC) (Mahecha et al. 2020) using a 3D CNN architecture, making this the first global-scale gap-filling solution on ESDC. By capturing both spatial and temporal relations, the model is able to generate predictions that are coherent on large scale and across patches, thus demonstrating the potential of the patch-wise gap-filling framework and the use of 3D CNN architectures for spatiotemporal gap-filling tasks.

How to cite: Zimmer, C., Neumann, A., Mahecha, M., and Umlauft, J.: PATCH-FILL: Multiscale and Univariate Gap-Filling in Remote Sensing Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10406, https://doi.org/10.5194/egusphere-egu25-10406, 2025.

Crop maps provide valuable insights for a range of applications, including water resources management, crop yield prediction, and the planning of domestic and foreign policies. Information on seasonal or yearly agricultural land cover can help governments and organizations make informed decisions to address agricultural challenges and promote environmental sustainability. However, large-scale land cover mapping remains a significant challenge due to the high computational demands of processing remote sensing data, especially when using high-resolution imagery for large-scale applications such as country-wide mapping.

This high computational requirement can be partially mitigated by performing object-based classification, where data is summarized into segments or fields. A challenge in object-level mapping is selecting an appropriate method for analyzing and interpreting the data. For instance, convolutional neural networks (CNNs), a commonly used deep learning algorithm, are not directly applicable in their basic form because they require a gridded structure. Graph Neural Networks (GNNs) present a novel approach, effectively analyzing relationships between objects represented as nodes and edges in a graph. The ability of GNNs to capture complex relationships between segments in a non-grid structure offers distinct advantages, such as handling irregular or non-Euclidean data and exploiting spatial and temporal dependencies within a region. This makes GNNs particularly well-suited for high-resolution remote sensing tasks where traditional grid-based methods may struggle with spatial context and object interactions.

This study applies GNNs to multitemporal Sentinel-1 Interferometric Wide Swath data (VV and VH polarizations), leveraging ten-day composites from May to September to capture seasonal crop growth dynamics. Training, testing, and validation datasets cover 40×40 km², 20×20 km², and 20×20 km² areas, respectively, within North Rhine-Westphalia, Germany. Sentinel-1 images are segmented using the Felzenszwalb-Huttenlocher algorithm, grouping pixels into objects. Each segment’s average backscatter values are calculated, and crop class labels are assigned using InVeKos ground truth data, which includes field boundaries and crop information. This data is transformed into a graph, where nodes represent segments, and edges define adjacency. The GraphSAGE framework is employed to train the GNN model.

Performance comparisons include segment-level and pixel-level neural networks (NNs). Preliminary results show that GNNs achieved the highest accuracy (88.01%), outperforming segment-level NN (86.02%) and pixel-level NN (78.89%). GNNs also demonstrated efficient computational performance, with shorter inference times (0.19 seconds) compared to pixel-based methods (10.7 seconds), and generated more homogeneous maps, minimizing the salt-and-pepper effect.

These results highlight the potential of GNNs for scalable, object-based mapping at high resolution. The approach will be expanded to classify cropland across Germany, generating a 10-meter spatial resolution crop map. By leveraging the temporal dynamics of Sentinel-1 data and incorporating 2022 data, this method offers an efficient and robust framework for large-scale applications in crop management, land-use monitoring, and resource planning.

How to cite: Donmez Altindal, E., Leonhardt, J., Roscher, R., Heckelei, T., and Storm, H.: Graph Neural Networks for Crop Cover Mapping, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11619, https://doi.org/10.5194/egusphere-egu25-11619, 2025.

Climate change is fundamentally altering Earth's natural systems, from shifting weather patterns and sea level rise to increasingly frequent extreme events. Understanding and responding to these changes demands continuous, reliable observations of our planet. While Earth-observing satellites have collected terabytes of data in recent decades with ever-increasing temporal, spatial, and spectral resolution, synthesizing these diverse data sources into homogeneous, long-term records remains a significant challenge for climate monitoring and situational awareness. 

We address this challenge with Instrument-to-Instrument Translation (ITI), an artificial intelligence framework that learns to translate between different satellite imaging domains. Building on unpaired image-to-image translation techniques, ITI overcomes a fundamental limitation in satellite data integration - it does not require the instruments to observe the same location at the same time. This flexibility enables ITI to perform instrument intercalibration, enhance image quality, mitigate sensor degradation, and achieve super-resolution asynchronously across multiple wavelength bands to enable multi-vantage point observations

Building on ITI's proven success in harmonizing solar observations, we extend the framework to address the unique challenges of Earth observation and atmospheric monitoring. More specifically, we demonstrate ITI’s capability by harmonizing observations from two geostationary weather satellites with complementary coverage: the Meteosat Second Generation (MSG) monitoring Europe and Africa with 11 spectral bands, and the Geostationary Operational Environmental Satellite (GOES-16) observing the Americas with 16 spectral bands. For this, we developed rs_tools, a comprehensive software package that streamlines the creation of machine learning-ready datasets, and adapted the ITI pipeline to handle the specific complexities of Earth observation data, e.g. missing observations of visible bands at night. 

Our results reveal good agreement between the ITI-translated imagery and actual high-quality observations, especially for infrared spectral channels. We conduct a multi-faceted performance analysis using image quality metrics (PSNR, histogram distributions, power spectra) across varying spatial scales, spectral bands, and geographic features (land/ocean). The unique overlap in MSG and GOES-16 coverage over the Atlantic Ocean enables additional validation through paired metrics (MSE, Pearson correlation, SSIM) after projecting both observing systems into a common reference frame.

The ITI tool is available as open-source software for the research community, and can easily be adapted to novel datasets and research applications. This research outcome is supported by NASA award 22-MDRAIT22-0018 (No. 80NSSC23K1045) and managed by Trillium Technologies, Inc.

How to cite: Jungbluth, A., Freischem, L., Johnson, J. E., Jarolim, R., Schirninger, C., and Spalding, A.: Instrument-to-Instrument translation: An AI tool to intercalibrate and homogenize observations from Earth-observing satellites, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12622, https://doi.org/10.5194/egusphere-egu25-12622, 2025.

As primary producers in Earth's system, plants drive global matter and energy fluxes. Understanding the global distribution of plant functional traits and their biodiversity is, therefore, critical for understanding ecosystem behavior and Earth system dynamics in the face of climate and global change. However, we lack observations for various plant functional traits, such as plant height, leaf size, and nitrogen content, at a global scale.

These data gaps could be addressed through citizen science projects, where thousands of individuals have already recorded millions of plant photographs for species identification purposes. While these photographs do not include direct information about plant traits, trait data for thousands of plant species can be accessed from scientific databases. By linking these two data sources—crowd-sourced plant photographs and trait information from scientific databases—through plant species, we can supervise computer vision models to infer plant traits from plant images. The principle of "form follows function" suggests that a plant's appearance can provide valuable insights into its functional properties.

To assess the potential of citizen science data for plant trait estimation, we propose testing the feasibility of using weak and noisy labels for effective trait prediction. Considering that different plant traits are not independent of each other, we leverage multi-trait learning. Additionally, our approach incorporates plant images along with ancillary environmental data, such as soil conditions and Earth observation satellite data, to provide crucial context on factors like climate or land surface properties.

To fairly evaluate model performance, we curate a clean dataset spanning diverse geographic regions, as well as taxonomic and phylogenetic groups. We conduct a comprehensive study on the resilience of trained models across these distribution shifts. Furthermore, we assess which traits can be effectively learned from noisy labels and explore the extent of trait transferability under different conditions.

Our findings indicate that models trained on noisy data can, to a notable extent, predict a series of plant traits, including plant height, leaf area, and specific leaf area. This approach provides an efficient, scalable, and non-destructive method for estimating important plant functional traits. It could lay the groundwork for large-scale biodiversity monitoring and ecosystem assessment, with the potential to revolutionize how we track the functional properties of ecosystems at a global scale.

How to cite: Sharma, A., Lusk, D., Trost, J., and Kattenborn, T.: PlantTraitNet: A Multi-Modal, Multi-Task Approach to Learning Global Plant Trait Patterns Using Citizen Science Data and Noisy Labels, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12681, https://doi.org/10.5194/egusphere-egu25-12681, 2025.

Urban dwellers today frequently document their daily experiences through digital photography, sharing these moments across popular social networking platforms. While these platforms host vast collections of images, the geographical data associated with many photos is often imprecise or missing entirely. Accurately determining the geographic coordinates of user-submitted photographs adds substantial value to these visual records, offering practical applications in city development planning, architectural studies, and public safety monitoring. Despite its potential benefits, the process of precise image geo-localization remains technically complex and challenging. This study presents an innovative approach to geo-localize crowd-sourced images in urban settings, addressing the limitations of traditional methods. By combining street-view panoramas and satellite imagery through a novel contrastive learning framework, we significantly improve localization accuracy. Using Hong Kong as a case study, we demonstrate substantial improvements over existing approaches, reducing median and average errors by 77.4% and 63.6%, respectively. Surprisingly, our findings reveal that satellite imagery alone outperforms street-view data in geo-localization tasks, challenging previous assumptions. This research not only advances the field of urban image geo-localization but also provides a valuable multi-source benchmark, paving the way for future innovations in urban sensing, mapping, and analysis across various disciplines.

How to cite: Hou, Q., Hou, C., Zhang, F., and Weng, Q.: Crowd-sourced images geo-localization method based on multi-modal deep learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14877, https://doi.org/10.5194/egusphere-egu25-14877, 2025.

Geostationary satellites such as GOES, GK2A, Himawari, and MTG/MSG have been providing a wealth of observational data over the past few decades, capturing the evolution and movement of clouds and precipitation systems with unprecedented detail. This extensive record has been instrumental in advancing our understanding of atmospheric dynamics and supporting the continuous monitoring of extreme weather events and natural disasters. Despite their capability to observe wide areas and the advantage of covering the entire globe with data from just three satellites, utilizing these geostationary satellites to model the Earth system on a global scale has proven challenging due to differences in observation intervals, spectral channels, and coverage footprints. To address these challenges, we propose a zero-shot Video Frame Interpolation (VFI) framework designed to harmonize imagery from multiple geostationary satellites. This method adapts the Many-to-Many Splatting VFI model, originally developed for RGB video processing, to work with single-channel infrared satellite imagery. By generating intermediate frames at high temporal resolutions (e.g., 2–5-minute intervals), our approach enables near-synchronous global coverage with improved temporal consistency. This method offers two key benefits. First, it enhances uniform sampling across satellites, creating a cohesive global view that is particularly valuable in data-sparse regions. Second, the interpolated frames improve the ability to capture and track critical meteorological phenomena. In addition, we address practical considerations such as computational efficiency, consistency with radiative or brightness temperature fields, and the robustness of zero-shot generalization. Our findings suggest that this zero-shot VFI framework can significantly advance global nowcasting providing a pathway to more accurate and timely Earth system modeling.

How to cite: Choi, Y., Ryu, H. G., Seo, M., and Kim, D.: Harmonizing Multisatellite Geostationary Observations Using Zero-Shot Video Frame Interpolation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16027, https://doi.org/10.5194/egusphere-egu25-16027, 2025.

Clouds affect Earth’s radiation balance by reflecting incoming sunlight (cooling effect) and trapping outgoing infrared radiation (warming effect). Their vertical distribution in the atmosphere significantly influences their radiative properties and overall climate impacts. However, how clouds will respond to climate change remains unknown: cloud feedbacks are the largest source of uncertainty in climate projections. Global 3D cloud data can help reduce these uncertainties, improve climate predictions, and support better decision-making.

Clouds are observed globally from space using satellites, which provide insights into their distribution, structure, and evolution. Observations from the Cloud Profiling Radar (CPR) aboard NASA’s CloudSat mission have provided valuable information on the vertical distribution of clouds. However, its long revisit times (~ 16 days), narrow swath (1.4 km) and observations limited to the same local time each day hinder our ability to study the temporal evolution of clouds or their diurnal cycle. In contrast, imaging instruments observe larger regions with higher temporal resolution but only offer a top-down view with limited vertical information.

In this work, we apply deep learning to images observed by geostationary satellites paired with vertical cloud profiles to extrapolate the vertical profiles beyond the observed tracks. Specifically, we use 11-channel imagery from the MSG/SEVIRI instrument, colocated with CPR vertical profiles. First, we pre-train models using self-supervised learning methods, specifically (geospatially-aware) Masked Autoencoders, applied to MSG/SEVIRI data from 2010. The pre-trained models are then fine-tuned for the 3D cloud reconstruction task using paired image-profile data. As only a small fraction of images overlap with CloudSat observations, the pre-training step enables us to exploit the full information contained in the MSG/SEVIRI images. We find that pre-training consistently improves reconstruction performance, particularly in complex regions such as the inter-tropical convergence zone. Notably, geospatially-aware pre-trained models incorporating time and coordinate encodings outperform both randomly initialized networks and simpler U-Net architectures, leading to improved reconstruction results compared to previous work.

In the future, we plan to extend this method to longer time periods and apply it to ESA’s EarthCARE data, once available, to further improve 3D reconstructions and enable the development of long-term 3D cloud products.

How to cite: Girtsou, S., Freischem, L., Bintsi, K.-M., Castiglione, G., Diaz Salas-Porras, E., Eisinger, M., Johnson, E., Jones, W., Jungbluth, A., and Massant, J.: Reconstructing 3D cloud fields from multispectral satellite images using deep learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16891, https://doi.org/10.5194/egusphere-egu25-16891, 2025.

Remote sensing data from satellites offer real-world observations on large spatial scales without incorporating model biases and model simplifications such as contained in reanalysis datasets. Numerical weather prediction models benefit largely from data with high temporal and spatial resolution, as provided by Earth observation remote sensing missions. Yet, while geostationary (GEO) satellites provide data at high temporal, e.g., 15 minutes, but low spatial resolution, e.g., 5 km, low earth orbit (LEO) satellites deliver data at low temporal, e.g., 16 days, but high spatial resolution, e.g., 90 m. In this research study, we therefore train a combination of a masked autoencoder and a ResNet model to learn a mapping from GEO to LEO Land-Surface Temperature (LST) products. The model receives the coarse-resolution 5 km LST from the Copernicus Global Land Service (apart from other static inputs) to approximate the fine-grained 70 m LST product from NASA’s ECOSTRESS mission. We use the spatial domain extent over Europe defined by the Land Atmosphere Feedback Initiative (LAFI). In theory, our algorithm allows the generation of 70 m LST estimates at a temporal resolution of 15 minutes. However, missing or corrupted input patches, when covered by clouds or in the event of missing sensor coverage or outages, challenges this optimal resolution. Therefore, we aim at 70 m daily LST estimates across continental Europe. We will present examples of the super resolution results from different biome regions across Europe, highlighting the potential and limitations of our approach.

How to cite: Traub, M., Karlbauer, M., Hellwig, F. M., Jagdhuber, T., and Butz, M. V.: Land-Surface Temperature Super Resolution from Geostationary to Low Earth Orbit Satellite Products with a Masked Autoencoder, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18922, https://doi.org/10.5194/egusphere-egu25-18922, 2025.

Frequent cloud cover and terrain-induced shadows pose significant challenges for reliable forest monitoring. Traditional monitoring methods, such as ground-based observations and aerial surveys, often suffer from low temporal resolution, making it difficult to track seasonal changes or detect sudden forest anomalies, such as windthrow damage. Earth Observation (EO), particularly Sentinel-2 imagery, offers the potential for high revisit rates and global coverage, but these advantages are diminished by the persistent presence of clouds and shadows, particularly during winter months in mountainous areas. The tasks of forest anomaly detection and windthrow damage assessment particularly benefit from the increased temporal resolution provided by cloud and shadow free Sentinel-2 imagery. 

The SAFIR project aims to develop a scalable and robust framework for comprehensive forest monitoring, with a focus on resilience in complex terrains, including mountainous regions. Within the project and to fully leverage the advantages of EO, it is crucial to implement effective preprocessing techniques to address cloud and shadow disturbances. These challenges can be overcome by employing a method that predicts missing image information by reconstructing the albedo. This process involves integrating spatial, spectral, temporal, and physical priors into the image restoration, allowing for the extraction of meaningful information from partially obscured satellite measurements. 

This contribution introduces a concept for a modular deep learning framework designed to process cloudy or shadowed satellite images and predicting the corresponding albedo values. The framework consists of two core modules: a shadow remover and a cloud remover. Both modules undergo pretraining on large cloud-free satellite datasets to build robust spatiotemporal embeddings. They are subsequently fine-tuned using physics-based methods to improve accuracy in restoring obscured and clouded image areas. Unlike traditional approaches that prioritize visual clarity, this framework is optimized for machine learning. The objective is to create enhanced data products for downstream forest monitoring applications. The effectiveness of this approach is validated by comparing the results with non-enhanced Sentinel-2 data, making the downstream tasks a methodological validation step.  

Validation is also conducted using multimodal data, integrating satellite imagery with high-resolution Unmanned Aerial Vehicle (UAV) data. The planned UAV campaigns, conducted in Portugal, Germany and Austria, capture low-altitude imagery at 120 m. Hence, they provide ground-truth validation by revealing surface conditions beneath cloud cover. This validation step supports the fine-tuning of the image restoration models and ensures that restored satellite images align closely with real-world conditions.  

By leveraging heterogeneous data sources, including high-quality in situ UAV data, this contribution introduces a scalable concept for high-frequency satellite monitoring. The framework aims to go beyond experimental setups and achieve operational deployment in the GTIF initiative by ESA, making EO more efficient. 

How to cite: Beltrame, L., Salzinger, J., Lampert, J., and Fanta-Jende, P.: Towards a Scalable Deep Learning Framework for Forest Monitoring under Challenging Conditions with Multimodal Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19171, https://doi.org/10.5194/egusphere-egu25-19171, 2025.

Urban heat islands (UHI) exacerbate health and environmental challenges, disproportionately affecting vulnerable populations. This study identifies high-risk areas for UHI effects in the Americas, including their metropolitan regions, using a suitability analysis model. It highlights the interplay between urban expansion, social vulnerability, and climate stress, emphasizing the urgency of addressing these issues in rapidly urbanizing contexts.

High-resolution satellite imagery and geospatial data were used to build the model. Key criteria included population density (dasymetric layers from WorldPop), relative wealth index, land surface temperature (LST) from MODIS, land cover from MODIS, PM2.5 and NO2 concentrations (Sentinel-5P), and road network layers derived for the analysis. Each criterion was reclassified, transformed to a common scale, and weighted equally to ensure consistency and comparability.

The suitability index was generated using raster algebra (weighted sum), producing a continuous map where higher values indicate greater susceptibility to heat stress and lower socioeconomic status. The analysis revealed spatial patterns that highlight areas with high potential impacts due to UHI characteristics.

The suitability index serves as a tool for identifying priority areas for targeted interventions and climate mitigation actions. This integrative approach highlights the need for sustainable urban development policies that reduce socio-environmental disparities and promote resilience in vulnerable communities

How to cite: Rocha, T.: Identifying Urban Heat Island Risk Areas with Vulnerable Populations: A Suitability Analysis Approach to support Health Interventions in the Americas, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-742, https://doi.org/10.5194/egusphere-egu25-742, 2025.

The digital twin is a useful tool for scientists and decision makers to understand the present (what now), explore future trajectories (what next) to to investigate the future impacts of current risk mitigation actions (what if), or of a system. Working at the local scale allows detailed physics to be implemented in an approach that better captures the complexity of the study site (city, watershed, etc.) in an approach that complements the global scale. The availability of very high-precision spatial products (optical, 3D, thermal, etc.) enables this high-precision local analysis anywhere on the Earth.This growing interest is leading a number of actors to build digital twins at the local scale. However, building this type of representation requires a dedicated effort from the user, usually a scientist, which prevents him from focusing on the scientific added value he could bring with his thematic expertise.
The Digital Twin Factory (DTF, 2024-2026) project, coordinated by the French National Centre of Space Studies (CNES), aims to provide users with a framework capable of building, deploying and operating a digital twin at the scale of the site. It is designed as a Digital Twin as a Service API (PaaS) to abstract the underlying infrastructure, with possibility of accessing both the HPC resources and usual Cloud providers. The DTF also provides users with methodological building blocks to access (catalogue harvester), manipulate (ingester, data processing pipeline), visualize and analyze (plot, dashboarding) the data. In this way, the instantiators of the digital twin can focus on their thematic expertise and deploy their physical solvers with access to multi-source data.
While high performance computing resources can be made available to run these physical models, parametric studies or climate trajectories may require high cost and long simulation times. Partial or full data based surrogate model is an approach that can overcome this barrier and provide results in a reactive manner. Part of the DTF's work is therefore aimed at providing users with methodological building blocks for surrogate modelling, based on the expertise of the scientific community.
This contribution presents the multi-layered architecture of the DTF project, its different components and the services offered to users. We illustrate this work with the construction of the digital Twin of Nokoué Lake in Benin that integrates flood forecasting, pollution control, salinity management, long-term risk evolution, risk governance, and adaptation measures. Satellite data are used as input for a hydrodynamic code, on which first developments of surrogate models are presented.

How to cite: Xavier, T., Derksen, D., Martin, V., and Brunet, P.-M.: Building a framework for the design and deployment of digital twins: the Digital Twin Factory project, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1299, https://doi.org/10.5194/egusphere-egu25-1299, 2025.

An Earth System Digital Twin (ESDT) is a dynamic, interactive, digital replica of the state and temporal evolution of Earth systems. It integrates multiple models along with observations, and connects them with analysis, AI, and visualization tools. Together, these enable users to explore the current state of the Earth system, predict future conditions, and run hypothetical scenarios to understand how the system would evolve under various assumptions. To establish a sustainable, extensible, and expandable ESDT solution, a formal software architecture is needed. Through the collaboration across multiple NASA centers, the teams partnered with the Apache Software Foundation to establish the professional open-source Integrated Digital Exploratory Analysis System (IDEAS) framework for digital twins. IDEAS is designed with three basic concepts. 

• Design to Exploit– ESDT architect needs to be aware of existing freely available provisioned data, model, and visualization services that the ESDT could benefit from. In computer science, we instructed our students the simple Don’t Repeat Yourself (DRY) principle. The is especially true for ESDT, because ESDT is aiming for operational big data solutions, not just a quick demonstration. If an organization is already actively making their curated data, analysis, or model outputs available, the ESDT should find ways to exploit them. Software reuse is another form of exploitation. Why reinvent? Exploiting existing services and software reuse would significantly reduce future operation costs. An ESDT should also be prepared to be exploited by other ESDTs. It will be discussed in the Design to Expand principle. 
• Design to Extend– An ESDT needs to be extensible to support new measurements, models (both numeric and AI-based), and interfaces. As we are connecting our digital assets, we will encounter gaps and limitations, both in data and technology. As we identify new climate phenomena or scientific needs, the ESDT needs to be prepared for these changes in technology and requirements. 
• Design to Expand– It would be unrealistic to expect a single ESDT is capable to replicate the complex Earth System that is equipped with all the past and present observations, to drive all possible global and regional numerical models, and process all AI capabilities concerning instruments, data, and predictions. We could try to acquire all publicly available data, but an ESDT should also be able to integrate private, commercial data. The Babel-like ESDT would require significant computing and store, and well as staff (science and technical) to operate. This is the motivation behind the open-source IDEAS framework, to encourage collaboration to set up common public-facing architecture. Imagine being able to orchestrate a federation of ESDTs using exploration tools like a Jupyter notebook without having to setting up a local ESDT instance. We think federated ESTDs is the answer to making actionable science available to Disadvantaged Communities. 

The presentation will discuss IDEAS architecture, its latest progress, and its growing portfolio of digital twins for the Earth including air quality, wildfire, hydrology, and coastal zone.

How to cite: Huang, T.:  Integrated Digital Exploratory Analysis System (IDEAS) – An Open-Source Software Framework for Digital Twins, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2622, https://doi.org/10.5194/egusphere-egu25-2622, 2025.

Environmental digital twins face significant interdisciplinary challenges in their development and operation, particularly in managing complex environmental data and facilitating effective communication among diverse stakeholders. While these virtual representations of environmental systems offer powerful capabilities for monitoring and decision-making, they often struggle to bridge the communication gap between technical experts, decision-makers, and end-users.

Data storytelling is the practice of narrating messages derived from data to address specific needs and visually communicating these messages to an audience in an ordered manner that is easily understandable. Interestingly, digital twins share a similar objective: both aim to simplify and communicate complex data through intuitive and meaningful narratives.

Building on this shared characteristic, we propose an approach that adapts data storytelling techniques to the creation of digital twins. This abstract focuses on how data storytelling can enhance the creation and communication of digital twin data through visual formats tailored to specific audiences, addressing their unique needs to support monitoring, decision-making, and actionable insights.

This innovative integration of data storytelling and environmental digital twins establishes a comprehensive approach to address three key challenges:

• Documenting and structuring the development process  from data to communication to incorporate stakeholder needs and communication requirements from the outset.
• Facilitating collaboration among interdisciplinary teams through shared narrative frameworks.
• Ensuring environmental insights are effectively translated into actionable knowledge.

We present a methodology that leverages data storytelling techniques to enhance the accessibility and impact of environmental digital twins, ultimately improving their effectiveness in environmental monitoring, decision-making, and stakeholder engagement.

How to cite: EL outa, F. and Dechambenoit, G.: What if Data story telling was the corner stone for environmental digital twins?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3702, https://doi.org/10.5194/egusphere-egu25-3702, 2025.

Machine learning models have emerged as powerful tools for simulating Earth system processes. Following their successful application in capturing atmospheric evolution for medium-range weather forecasts, attention has increasingly shifted towards other components of the Earth system, such as the marine and land environments. This interest is further driven by the potential to enhance forecasting capabilities beyond the medium range. Machine learning frameworks offer remarkable flexibility in integrating these model components to achieve a coherent Earth system representation. At one end of the spectrum, model components can be trained jointly within a unified framework optimised using a shared loss function. At the other end, components may be developed independently and coupled by exchanging physically relevant information at multiple interfaces, mirroring the traditional coupling strategies employed in numerical models. In this presentation, we will examine the advantages and challenges of these approaches, with a particular emphasis on coupling the atmospheric, land, and marine components within the deterministic AIFS model, the machine learning-based forecast system developed at ECMWF. Furthermore, we will compare the coupling strategies of data-driven models with those of traditional numerical models, highlighting their strengths and limitations.

How to cite: Zampieri, L., Cook, H., Furner, R., Hahner, S., Pinault, F., Raoult, B., Raoult, N., Santa Cruz, M., and Chantry, M.: Coupling approaches for data-driven Earth system models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4499, https://doi.org/10.5194/egusphere-egu25-4499, 2025.

Our presentation aims to describe the development and operationalisation of the Destination Earth (DestinE) Extremes Digital Twin (DT), including the On-Demand component, a system designed to improve the prediction and management of extreme weather events in Europe. The system leverages high-resolution weather models using information from Extreme Detection (EDF) and Triggering (DTF) Frameworks, as well as ECMWF ensemble, incorporating impact-specific models for hydrology, air quality, renewable energy, and more. A key component is a configuration lookup table prioritising end-user needs and available resources. The system incorporates various masking techniques (ACCORD models configurations, geographical, capacity, event type) to refine forecasts. The presentation describes the system's architecture, data sources, and workflow, emphasising the integration of multiple models and data sources, and the use of cutting-edge technologies such as GPUs and machine learning for enhanced forecasting and efficient resource utilisation. Pilot regions are used for testing and operationalisation, with a phased approach planned for broader deployment. The project addresses challenges in forecasting accuracy, communication of uncertainty, and the integration of forecasts into decision-making processes across various sectors.

How to cite: Randriamampianina, R. and the DestinE Extremes Digital Twin Team: Digital twin for weather-induced extremes, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5110, https://doi.org/10.5194/egusphere-egu25-5110, 2025.

How to cite: Purohit, K., Sinha, M., Panda, A., Banerjee, S., and S Nanjundiah, R.: Enhancing the Skill of Medium Range Forecasts with a Machine Learning Based Multi-Model Super-Ensemble (MMSE), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5909, https://doi.org/10.5194/egusphere-egu25-5909, 2025.

How to cite: Ebel, P., Magnusson, L., and Schneider, R.: Global post-processing of ERA5 precipitation product via graph-based neural networks , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6965, https://doi.org/10.5194/egusphere-egu25-6965, 2025.

Developing region-specific radar quantitative precipitation estimation (QPE) products for South China (SC) is crucial due to its unique climate and complex terrain over there. Deep learning (DL) has emerged as a promising avenue for radar QPE, especially graph neural networks (GNNs). Many studies have tested the DL models in radar QPE, but virtually no studies have evaluated the performance of DL models in different precipitation intensities, types, or organizations. Moreover, limited attention has been given to whether DL-based methods can mitigate radar QPE errors caused by orographic influences in complex terrains, such as those in SC.

This study investigates the advantages of DL methods for QPE tasks in South China, utilizing nearly three years of hourly gauge data as labels and ground-based radar reflectivity as inputs. Firstly, multi-layer perceptron (MLP), Convolutional Neural Networks (CNNs), and GNNs with similar architectures are constructed and compared to traditional Z-R relationships considering precipitation types. DL methods outperform traditional Z-R relationships and GNNs perform the best. More importantly, this study conducts a systematic evaluation of the proposed GNN. For extreme precipitation (>30 mm/h), GNN achieves the smallest MAE, highlighting its potential for hazardous event estimation. It also demonstrates stable performance for stratiform and organized precipitation, with minimal bias and standard deviation. However, GNN is less effective for isolated precipitation, whereas CNNs are a better choice due to their ability to estimate scattered rainfall accurately. Last but not least, the Z-R relationship shows systematic spatial biases, overestimating precipitation in coastal plains and underestimating it in inland high-altitude regions. DL methods alleviate these terrain-induced biases by incorporating spatial information. Overall, this study highlights the advantages of DL methods across different precipitation scenarios and demonstrates their ability to mitigate systematic biases from complex terrain.

How to cite: Zhu, K. and Xu, W.: Deep Learning for Radar Quantitative Precipitation Estimation over Complex Terrain in Southern China, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8155, https://doi.org/10.5194/egusphere-egu25-8155, 2025.

Artificial Intelligence-based Numerical Weather Prediction (AI-NWP) models have recently emerged as powerful tools for weather forecasting, offering computational efficiency and high accuracy. This study explores the extreme weather events simulated by the Artificial Intelligence Forecasting System (AIFS), initialised with conditions derived from kilometer-scale storyline experiments using the IFS-FESOM model ─ where the atmospheric circulation is constrained to observations. We present two case studies: the 2023 South Asian humid heatwave and the 2024 Storm Boris. These two events are reproduced in the present climate, but also simulated if they were to unfold in pre-industrial and +2K future climates, effectively creating AI-driven storylines. The methodology we employ offers a complementary framework, where the use of AI-driven ensembles provides a scalable and rapid way to assess the potential uncertainty and variability associated with such events, by enabling us to explore a broader range of plausible outcomes at very low computational costs. By combining the strengths of physics-based modelling with the efficiency and flexibility of AI-driven simulations, this dual approach offers a pathway to operationalise ensemble-based extreme weather storylines.

How to cite: John, A., Rackow, T., Koldunov, N., Beyer, S., Sanchez Benitez, A., Gößling, H., Athanase, M., and Jung, T.: Exploring AI-Driven Event-based Storylines, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8169, https://doi.org/10.5194/egusphere-egu25-8169, 2025.

How to cite: Walt, E., Bruinsma, W., Schmeits, M., Gavves, E., and Coumou, D.:  Xaurora: Advancing subseasonal-to-seasonal forecasting by fine-tuning foundation weather models with spectral consistency , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9956, https://doi.org/10.5194/egusphere-egu25-9956, 2025.

How to cite: Montazeri, S., Stoicescu, M., Hinojo Comellas, O., Puechmaille, D., and Schick, M.: MLOps on DestinE Data Lake – Towards Reproducible AI on Edge Services, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10817, https://doi.org/10.5194/egusphere-egu25-10817, 2025.

How to cite: Furner, R., Adewoyin, R., Santa Cruz, M., Hahner, S., Keeley, S., Mogensen, K., and Zampieri, L.: Developing a data-driven global ocean model at ECMWF , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11883, https://doi.org/10.5194/egusphere-egu25-11883, 2025.

Systematic model errors significantly limit the predictability horizon and practical utility of the current state-of-the-art forecasting systems. Even though accounting for these systematic model errors is increasingly viewed as a fundamental challenge in the field of numerical weather prediction, estimation and correction of the predictable component of the model error has received relatively little attention. Modern implementations of weak-constraint 4D-Var are an exception here and a promising avenue within the variational data assimilation framework, showing encouraging results. Weak-constraint 4D-Var can be viewed as an online hybrid data assimilation and machine learning approach which gradually learns about model errors from partial and imperfect observations, allowing to improve the state estimation. We propose a natural extension of this approach by applying deep learning techniques to further develop the concept of online model error estimation and correction.

In this talk, we will present recent progress in developing a hybrid model for the ECMWF Integrated Forecasting System (IFS). This system augments the state-of-the-art physics-based model with a statistical model implemented via a neural network, providing flow dependent model error corrections. While the statistical model can be pre-trained offline, we demonstrate that by extending the 4D-Var control vector to include the parameters of the neural network, i.e. the model of model error, we can further improve its predictive capability. We will discuss the impact of applying the flow dependent model error corrections in the medium range forecasts on the forecast quality.

How to cite: Chrust, M., Farchi, A., Bonavita, M., Bocquet, M., and Laloyaux, P.: Development of an offline and online hybrid model for the Integrated Forecasting System, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11988, https://doi.org/10.5194/egusphere-egu25-11988, 2025.

The Horizon Europe project “Biodiversity Digital Twin for Advanced Modelling, Simulation and Prediction Capabilities” (biodt.eu) combines state-of-the-art supercomputing resources with FAIR data practices to address biodiversity grand challenges such as the loss of diversity at genetic, species and ecosystem levels driven by factors such as anthropogenic climate change, intensified land use and the spreading of alien invasive species.

Key tools for the integration and assessment of such information in BioDT are so-called prototype Digital Twins (pDTs) which provide digital replicas of particular phenomena of the biosphere. The data produced by pDTs and the computational workflows themselves are annotated with rich machine-interpretable metadata, notably schema.org and its Bioschemas and Croissant (high-level metadata format for machine learning datasets) extensions to enable discovery and interaction with the data for humans and machines. 

The crop wild relatives digital twin developed in BioDT for the identification of novel genetic resources in crop wild relatives (CWR) aims to support the development of mitigation strategies against food-related crises associated with climate change (doi:10.3897/rio.10.e125192). Underutilised crops and CWR are a valuable source of untapped genetic diversity for implementing such strategies because their greater genetic diversity in relation to crops makes adaptation to droughts, cold waves, heavy precipitation or other weather-induced extreme events possible. They are therefore of particular importance to guarantee food security.  

We developed a pilot study to migrate and functionally integrate the pDT CWR in the Destination Earth Data Lake (DEDL) to increase its availability and, more importantly, use both data provided in the data space and the near data processing features of DEDL to improve models and prediction. The framework makes use of web-based technologies such as RO-Crates and FAIR Signposting to facilitate reuse and repurposing of the data within and across different data spaces as well as orchestrated interplay of multiple digital twins (involving the Climate Change Adaptation Digital Twin). 

Building on this blueprint, we will showcase in this presentation the deployment of FAIR agrobiodiversity workflows in DEDL’s compute service Islet and the subsequent publication of results in a lightweight service framework (figure).

How to cite: Weiland, C., Bauer, D., Chala, D., Endresen, D., Grieb, J., Orwick Rydmark, M., and Zuquim, G.: Implementing FAIR Agrobiodiversity Workflows for the Destination Earth Data Lake, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12042, https://doi.org/10.5194/egusphere-egu25-12042, 2025.

Destination Earth (DestinE) is a flagship initiative led by the European Commission, implemented by EUMETSAT, ESA and ECMWF. It aims to create highly detailed Digital Twins (DTs) of the Earth, enabling precise simulations for a variety of uses. Currently, the initiative focuses on two primary Digital Twins:  the Weather Extremes Digital Twin (ExtremeDT) and the Climate Change Adaptation Digital Twin (ClimateDT). Over the coming years, the scope of DTs is set to expand, necessitating improved access to data and streamlined methods for working with it. This is where the Destination Earth Data Lake (DEDL) plays a pivotal role, offering comprehensive data discovery, access, and processing services tailored to the needs of DestinE users.

The DEDL operates on two key levels: ‘Data Discovery and Access’ and ‘Edge Services’. DEDL Discovery and Data Access services is provided by Harmonized Data Access (HDA) tool which provides a single, federated entry point to the services and data, including resources from existing datasets and complementary sources such as in-situ and socio-economic data. Notably, it also provides access to the unique datasets generated by DestinE’s DT’s. The services rely on use of the SpatioTemporal Asset Catalogs (STAC) standard which means:

• The search in the dataset is done according to the STAC protocol;
• The Federated Catalog search proxy component converts STAC queries into queries adapted to the underlying catalog and returns the results to the user in STAC format.

The cloud computing service is powered by the ISLET infrastructure, a distributed Infrastructure as a Service (IaaS) built on OpenStack. It allows users to manage virtual machines, s3 storage, and run advanced computations via a graphical user interface or command-line interface. A standout feature of ISLET is its proximity to data sources, operating near High-Performance Computing (HPC) facilities. This is achieved through data bridges, enabling efficient processing of large datasets, including those from Digital Twins, in conjunction with HPC systems.

The STACK environment supports application development using JupyterHub and DASK, with Python, and R languages. Users can create DASK clusters on selected infrastructure (sites) to process data directly where it resides, removing the need for extensive local setup and optimization.

Hook Services is a set of pre-defined workflows which could be used by users as a ready-to-use processors like: Sentinel-2: MAJA Atmospheric Correction; Sentinel-1: Terrain-corrected backscatter. It also enables workflow functions to generate on-demand higher-level products, such as temporal composites.

DEDL is a transformative initiative that revolutionizes how Earth Observation data is managed and utilized. By integrating innovative infrastructure (ISLET), data services (HDA), reliable processors (Hook Services), and user-friendly development tools (STACK), DEDL enables unprecedented levels of data harmonization, federation, and processing. Moreover, the DEDL plays a crucial role in empowering DestinE users by providing them with seamless access to vast datasets and advanced computational tools. It simplifies the process of data exploration, integration, and analysis, enabling researchers, policymakers, and developers to focus on innovation and decision-making rather than technical barriers. This cutting-edge system enhances climate research capabilities and supports sustainable development efforts on a scale previously unattainable.

How to cite: Grzybowski, P., Ziółkowski, M., Lambare, A., Reimer, C., and Schick, M.: How Destination Earth Data Lake support Destination Earth users, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12349, https://doi.org/10.5194/egusphere-egu25-12349, 2025.

Tropical deciduous ecosystems play a critical role in terrestrial ecological processes and the global carbon cycle, influencing seasonal climates through phenology-induced biophysical and biogeochemical feedbacks. Phenological processes for tropical deciduous ecosystems are complex with multiple intertwining climatic and physiological factors that co-shape the underlying dynamics. Here, we present a deep learning framework based on Temporal Fusion Transformer for predicting tropical deciduous phenology globally in real-time with high accuracy. The framework integrates long-term AVHRR-derived vegetation greenness data, high-resolution climate data from ERA-Land, and land surface features including physical and chemical properties to account for terrestrial spatial heterogeneities that affect phenological processes. Our preliminary results demonstrate the ability of the developed framework to accurately predict historical phenological dynamics across 35 growing seasons in the pan-tropical regions. Key phenological metrics, including the start, peak, and end of the growing season, are identified with high accuracy. We believe the framework provides a powerful tool for real-time predictions and reconstructions of phenological states for tropical deciduous ecosystems, especially in regions where human activities like deforestation and agriculture heavily influence the estimates of tropical carbon cycle potential. With insight into the potential phenological states, this framework may help inform sustainable land management practices in pan-tropical regions.

How to cite: Wu, M., Tagesson, T., Cai, Z., and Duan, Z.: Real-time Prediction of Global Tropical Deciduous Ecosystem Phenology with Deep Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13817, https://doi.org/10.5194/egusphere-egu25-13817, 2025.

Based on the TRMM dataset, this paper compares the applicability of the improved MCE (minimum circumscribed ellipse), MBR (minimum bounding rectangle), and DIA (direct indexing area) methods for rain cell fitting. These three methods can reflect the geometric characteristics of clouds and apply geometric parameters to estimate the real dimensions of rain cells. The MCE method shows a major advantage in identifying the circumference of rain cells. The circumference of rain cells identified by MCE in most samples is smaller than that identified by DIA and MBR, and more similar to the observed rain cells. The area of rain cells identified by MBR is relatively robust. For rain cells composed of many pixels (N > 20), the overall performance is better than that of MCE, but the contribution of MBR to the best identification results, which have the shortest circumference and the smallest area, is less than that of MCE. The DIA method is best suited to small rain cells with a circumference of less than 100 km and an area of less than 120 km², but the overall performance is mediocre. The MCE method tends to achieve the highest success at any angle, whereas there are fewer “best identification” results from DIA or MBR and more of the worst ones in the along-track direction and cross-track direction. Through this comprehensive comparison, we conclude that MCE can obtain the best fitting results with the shortest circumference and the smallest area on behalf of the high filling effect for all sizes of rain cells.

How to cite: Cai, H., Mao, Y., Zhu, X., Fu, Y., and Zhou, R.: Comparison of the Minimum Bounding Rectangle and Minimum Circumscribed Ellipse of Rain Cells from TRMM, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14658, https://doi.org/10.5194/egusphere-egu25-14658, 2025.

We propose a simple yet effective framework for real-time surface ozone forecasting using deep learning. The framework highlights three key modules: independent channel encoders, frequency information extraction, and fine-tuning, all of which consistently enhance model performance. This unified model is built well to autonomously capture different spatial and temporal patterns of ozone concentrations, with an averaged RMSE of 8 ppb for day 1 forecasting. The performance of day 4 forecasting is slightly lower. We find that chemistry becomes less important than meteorology over time, indicating their different roles in short-term and long-term forecasting. Most high ozone episodes can be simulated, though capturing extremely high ozone values remains a challenge. Observations from China are trained and tested to demonstrate our model.

How to cite: Liu, Z.: A Unified Model of Forecasting Ozone by Deep Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14783, https://doi.org/10.5194/egusphere-egu25-14783, 2025.

Accurate rainfall prediction is essential not only for water resource management but also for forecasting and mitigating the impacts of climate change-driven weather events such as floods and droughts. Due to the high spatiotemporal variability of complex meteorological phenomena like rainfall, effective prediction necessitates in high-quality data collection, model application, and uncertainty analysis. Unlike existing studies that focus primarily on developing deep learning models to improve rainfall prediction accuracy, this study evaluates the uncertainty of rainfall predictions using pre-existing deep learning models, U-Net and ConvLSTM, with artificially generated elliptical rainfall data. Artificial rainfall data were designed with four temporal patterns: constant, gradually increasing, gradually decreasing, and time-varying. These patterns were applied in horizontal, vertical, and diagonal movements to evaluate the models' ability to handle spatiotemporal complexity. The results indicate that both deep learning models exhibited spatial smoothing issues on rainfall predictions over time. However, the U-Net model demonstrated superior spatiotemporal performance compared to ConvLSTM. While this study focuses solely on deep learning models for rainfall prediction, future research will consider factors such as data complexity and loss functions to conduct a comprehensive evaluation of prediction uncertainty. This work is expected to contribute to the development of methodologies for rainfall modeling using deep learning approaches.

Funding: This research was supported by Disaster-Safety Platform Technology Development Program of the National Research Foundation of Korea(NRF) funded by the Ministry of Science and ICT. (No. 2022M3D7A1090338)

How to cite: Kim, Y. and Lee, G.: Uncertainty Evaluation of Deep Learning Models Using an Artificial Rainfall, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15162, https://doi.org/10.5194/egusphere-egu25-15162, 2025.

Numerical Weather Prediction (NWP) models, which integrate physical equations forward in time, are the traditional tools for simulating atmospheric processes and forecasting weather in modern meteorology. With recent advancements in deep learning, Neural Weather Models (NeWMs) have emerged as competent medium-range NWP emulators with reported performances that compare favorably to state-of-the-art NWP models. However, they are commonly trained on reanalysis with limited spatial resolution (e.g., 0.25° horizontal grid spacing) and thus smooth out key features associated with a number of weather phenomena. For example, tropical cyclones—among the most impactful weather events due to their devastating effects on human activities—are challenging to forecast, as extrema like wind gusts, which serve as proxies for tropical cyclone intensity, are smoothed in deterministic forecasts at 0.25° resolution. To address this, we use our best global observational estimate of wind gusts and minimum sea level pressure to train models that post-process NeWM outputs and enable accurate and reliable forecasts of TC intensity. We present a tracking-independent post-processing algorithm and show that even naïve, linear models extract useful information from NeWM model outputs beyond what is present in the initial conditions used to roll out NeWM predictions. We explore how the NeWM’s spatial context may further improve the forecast through masking and convolutional architectures. Our post-processing framework thus presents a step towards democratization of tropical cyclone intensity forecasting, given the reduction in computational requirements for producing global weather forecasts with NeWMs compared to traditional NWP approaches and the algorithmic simplicity of the tracking-independent approach.

How to cite: Gomez, M., Beucler, T., Berne, A., and Poulain--Auzéau, L.: Post-Processing Neural Weather Model Outputs for Tropical Cyclone Intensity Forecasts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15669, https://doi.org/10.5194/egusphere-egu25-15669, 2025.

The Climate Change Adaptation Digital Twin (Climate DT) is part of the Destination Earth (DestinE) initiative, developing Digital Twins of Earth to increase resilience against environmental changes. More specifically, Climate DT provides capabilities supporting climate change adaptation at regional and national levels at multi-decadal time scales. We present here an overview of Climate DT, highlighting the added value for the users and discussing the transition of the system towards the operations.  

The development of Climate DT has started in Phase 1 of DestinE, during which the first prototype of the new climate information system has been developed. A key innovation of Phase 1 was the introduction of a generic state vector (GSV), which is evolved by the Earth system models (ESMs) and streamed to applications from climate adaptation impact sectors.  This has created a basis for a pioneering climate information system that enables (i) provision of global climate information at an unprecedented granularity, (ii) scaling the system across a number applications that have access to all the data they need, (iii) user-centric approach with new ways of co-design and opportunities for enhancing interactivity. In Phase 2, which started in May 2024, our focus is on operationalizing Climate DT to deliver high-quality climate and impact-sector information regularly while incorporating new interactive features.

The operational model of the Climate DT is built around three storm- and eddy resolving ESMs; ICON, IFS-NEMO and IFS-FESOM. The operational framework utilizes a DevOps-like cycle, including three set-ups: d-suite for development, e-suite for testing the operational set-up and o-suite for operating the system. The o-suite simulations will provide data covering both past (1990-2020) and future periods (2020-2050) with a 5 km global grid. Additionally, capabilities for special simulations are developed, including story-line simulations for future periods of extremes as well as what-if scenario simulations enabling a new level of interactivity.

The added value of Climate DT to users is demonstrated through four impact sector applications. These applications operate on the streamed GSV as part of the operational framework, and they are improved in co-design with key users. The impact sector applications cover societally relevant climate change adaptation domains, including wind energy management, disaster risk management (with regards to wildfires and floods), as well as agriculture and water management. Climate DT output, including high-resolution climate simulations, storyline simulations, user-relevant indicators and impact assessments are made available to users via DestinE Service Portal.

How to cite: Kontkanen, J., Acosta, M., Bretonnière, P.-A., Castrillo, M., Davini, P., Doblas-Reyes, F., Früh, B., von Hardenberg, J., Jung, T., Järvinen, H., Klocke, D., Koldunov, N., Manninen, P., Milinski, S., Mäkelä, J., Naraynappa, D., Polade, S., Sandu, I., Sievi-Korte, O., and Thober, S.: Climate Adaptation Digital Twin – building an operational climate information system to support decision-making, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15811, https://doi.org/10.5194/egusphere-egu25-15811, 2025.

The rapid growth of weather and climate datasets is increasing the pressure on data centres and hinders scientific analysis and data distribution. For example, kilometre-scale weather and climate models can generate 20 gigabytes of data per second when run operationally, making it generally infeasible to store all output unless advanced compression is applied. 

To address this challenge, novel lossy compression techniques, including recently so-called neural compressors which learn smaller representations of climate data, have been proposed with compression factors beyond 100x. However, if applied without care, lossy compression can remove valuable information from a dataset for often unknown downstream applications. It is therefore important to validate that the compression process does not alter scientific conclusions drawn from the data. Whether the compression error is tolerable is often easier to assess for domain experts and rarely well defined. 

Here, we address this challenge by presenting a benchmark suite for lossy compression of climate data (atmosphere, ocean, and land). We are defining data sets that can be used to train neural compressors as well as corresponding evaluation methods. Compressors have to pass a set of tests for each data set while compressing into the smallest file size possible at a reasonable (de)compression speed. To ensure evaluation on a diverse set of inputs, the benchmark covers climate variables following various statistical distributions at medium to very high resolution in time (hourly to yearly) and space (~1 km to 150 km). Evaluation tests are for single and multi-variable compression of gridded data with stable or changing statistics, random data access or large archives, in medium to very large datasets.

To provide references towards what compression levels can be achieved with current state of the art lossy compressors, we also evaluate a set of baseline compressors (SZ3, ZFP, Real Information) on our benchmark tasks. The benchmark is a quality check for new compressors towards a standardization of climate data compression, aiming to make compressors with high compression factors safe to use and widely supported.

How to cite: Reichelt, T., Tyree, J., Kloewer, M., Dueben, P., Lawrence, B., Hammerling, D., Baker, A., Faghih-Naini, S., and Stier, P.: ClimateBenchPress: A Benchmark for Compression of Climate Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15912, https://doi.org/10.5194/egusphere-egu25-15912, 2025.

Weather forecast downscaling, the problem of recovering accurate local predictions given a lower resolution forecast,  is commonly used in operational NWP pipelines. Its purpose is to recover some of the sub-grid processes that could not be represented by the underlying numerical model due to a limited resolution. This misrepresentation provokes statistical mismatches between the observation data gathered from stations and the nearest grid point in the numerical simulation.

Using a downscaling model typically requires making a compromise between spatial consistency and statistical calibration. Traditionally, these models are trained to target a traditional verification metric. Consequently, they suffer from the double penalty issue and fail to correctly model spatial correlation structures by becoming overly smooth. This is detrimental to downstream modeling tasks such as power grid management, which require a good assessment of spatially-correlated phenomena. 

Recently, the finer details of the atmospheric state have successfully been recovered using generative models such as denoising diffusion [2-4]. We propose a similar strategy for in situ downscaling by introducing a flow matching [1] model for that task. A cross-attention transformer [5] backbone allows us to build an internal representation for the gridded numerical forecast as well as the in situ downscaled forecast. 

Our model avoids the numerical instability and mode collapse issues related to Generative Adversarial Networks. It produces well-calibrated forecasts that better represent the spatial correlations between the stations when compared to non-generative alternatives. Our model makes no assumptions about the underlying forecast, and thus can be thought of in two ways. It can be considered a hybrid NWP/AI model, where we first run a numerical simulation and then downscale it. It can also be considered a supplementary forecasting product in a full machine learning pipeline.

Using our flow matching weather forecast downscaling model, we run experiments on the EUPPBench post-processing dataset to predict surface temperature and wind speed. Particular care is given to evaluating the model, where we assess both the marginal performance (via the CRPS, reliability histogram, and spread-error ratio) and the joint performance (via the Energy Score, local Variogram Score and forecast spatial frequency content). The accurate representation of extreme events is evaluated using Brier scores. Further experiments discuss the pitfalls of fitting the Energy Score directly without a generative model.

[1] Lipman, Y. et al. (2023) ‘Flow Matching for Generative Modeling’. arXiv. Available at: https://doi.org/10.48550/arXiv.2210.02747.

[2] Couairon, G. et al. (2024) ‘ArchesWeather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecasting’. arXiv. Available at: https://doi.org/10.48550/arXiv.2412.12971.

[3] Price, I. et al. (2023) ‘GenCast: Diffusion-based ensemble forecasting for medium-range weather’. arXiv. Available at: https://doi.org/10.48550/arXiv.2312.15796.

[4] Lang, S. and Chantry, M. (2024) ‘Enter the ensembles’, AIFS Blog, 21 June. Available at: https://www.ecmwf.int/en/about/media-centre/aifs-blog/2024/enter-ensembles (Accessed: 15 January 2025).

[5] Vaswani, A. et al. (2017) ‘Attention is All you Need’, in Advances in Neural Information Processing Systems. Curran Associates, Inc. Available at: https://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html (Accessed: 17 May 2022).

How to cite: Landry, D., Charantonis, A., and Monteleoni, C.: Flow matching for in situ, spatially consistent weather forecast downscaling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16981, https://doi.org/10.5194/egusphere-egu25-16981, 2025.

With the increasing availability and demand of remote sensing data from Earth observation satellites, the accuracy of weather prediction models can be improved substantially. Satellite products, such as Land-Surface Temperature (LST), however, suffer from missing data, either caused by clouds that cover the ground, by missing spatial coverage of the mission, or by outages of the sensors. Such spatial data gaps in LST products impose strict limitations when aiming to process the data further with, e.g., numerical weather prediction models assuming spatial continuity with gapless input data. We therefore propose a gap-filling algorithm based on a masked autoencoder that only receives a small percentage from a 32x32 LST snapshot and learns to reconstruct the missing patches. We use the spatial domain defined by the Land Atmosphere Feedback Initiative (LAFI) over central Europe and operate on geostationary LST data from the Copernicus Global Land Service in June 2023 at 5 km resolution. Our approach indicates considerable potential when filling spatial gaps in LST products, however, we emphasize one critical aspect. The LST estimates below clouds cannot be expected to be realistic and would require a sophisticated atmospheric correction. To mitigate this limitation, we aim to incorporate microwave data in future that penetrates clouds and therefore could help to estimate LST below clouds. In its current formulation, our algorithm can be used to fill gaps in LST products as if there were no clouds. We will show the potential and limitations of the autoencoder-based gap-filling algorithm for several showcases across Europe. 

How to cite: Karlbauer, M., Hellwig, F. M., Jagdhuber, T., and Butz, M. V.: Spatial Gap Filling in a Geostationary Land-Surface Temperature Product with a Masked Autoencoder, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17080, https://doi.org/10.5194/egusphere-egu25-17080, 2025.

The Children’s Climate Risk Index (CCRI) was first released in 2021, providing a comprehensive, global view of children’s exposure and vulnerability to the impacts of climate change. The CCRI is a composite index that aims to rank countries where children are exposed to climate and environmental hazards. The CCRI 2.0 builds on the previous index by integrating two pillars; Pillar 1 focusing on climate hazards and Pillar 2 on inherent vulnerabilities to WASH, health, education and other relevant dimensions. 

We highlight our contributions to CCRI 2.0, using a cluster methodology for quantifying children’s exposure to climate risks including riverine and coastal flooding, tropical storms, heatwaves, and drought. Using unsupervised learning, we allow for a data-driven approach to provide an interpretation of the ranking of children’s exposure to climate risks on a global scale between countries, as well as at the sub-national and local levels. It complements the previous method of constructing the synthesized index, which involved calculating the simple average of multiple indicators. We further discuss our techniques in tackling the challenges of multisource data processing, analysis, and visualization of geospatial data for user insight.

How to cite: Kim, D. and Doerksen, K.: Quantifying Children’s Exposure to Climate Risks using Unsupervised Learning with Multi-Source Geospatial Datasets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17875, https://doi.org/10.5194/egusphere-egu25-17875, 2025.

Sampling from climate models to generate ensembles of predictions is computationally expensive (Hawkins et al., 2015). Climate model ensembles are used to understand probabilities of climatic events and identify internal variability in climate models. In the short term, model uncertainty and inter-annual variability dominate uncertainty in climate predictions (Smith et al., 2019). A typical approach to address these uncertainties is to use large ensembles of non-learned, physical numerical global circulation models (GCM) (Eade et al., 2014). These ensembles allow for statistical analysis of distributions and determination of internal variability in the climate model.

Our approach demonstrates that we can efficiently learn to emulate a GCM. We use ensembles generated by the IPSL submission to the Decadal Climate Prediction Project (DCPP). The dataset ranges from 1960-2016 and produces 10-member, 10-year forecast ensembles for each year. On this dataset, we train a modified version of ArchesWeatherGen, a Swin Transformer based on PanguWeather that can be used in a generative way using flow matching (Couairon et al. 2024). The model was modified to predict additional climatic variables (e.g. air temperature, specific humidity, ocean potential temperature at depth, sea surface temperature, sea level pressure) at a monthly temporal resolution. Once trained, the model probabilistically generates ensemble members rapidly which can be auto-regressively rolled out. We show that they are physically reliable via evaluation methods that assess physical processes derived from the variables represented in the machine learning model, such as by evaluating it on El Niño/La Niña events. This model demonstrates that machine learning can enhance climate models by expanding ensemble sizes to improve our understanding of climatic processes. We aim to output physically realizable month-to-month trajectories to estimate future climate and its uncertainties across various domains, including land, ocean, and atmospheric processes.

Couairon, G., Singh, R., Charantonis, A., Lessig, C., & Monteleoni, C. (2024). ArchesWeather & ArchesWeatherGen: a deterministic and generative model for efficient ML weather forecasting. arXiv preprint arXiv:2412.12971.

Eade, Rosie, Doug Smith, Adam Scaife, Emily Wallace, Nick Dunstone, Leon Hermanson, et Niall Robinson. « Do Seasonal-to-Decadal Climate Predictions Underestimate the Predictability of the Real World? » Geophysical Research Letters 41, no 15 (2014): 5620‑28. https://doi.org/10.1002/2014GL061146.

Hawkins, Ed, Robin S. Smith, Jonathan M. Gregory, et David A. Stainforth. « Irreducible Uncertainty in Near-Term Climate Projections ». Climate Dynamics 46, no 11 (1 juin 2016): 3807‑19. https://doi.org/10.1007/s00382-015-2806-8.

Smith, D. M., R. Eade, A. A. Scaife, L.-P. Caron, G. Danabasoglu, T. M. DelSole, T. Delworth, et al. « Robust Skill of Decadal Climate Predictions ». Npj Climate and Atmospheric Science 2, no 1 (17 mai 2019): 1‑10. https://doi.org/10.1038/s41612-019-0071-y.

How to cite: Clyne, G., Couairon, G., Mignot, J., Gastineau, G., Charantonis, A., and Monteleoni, C.: ArchesClimate: Ensemble Generation for Decadal Prediction using Flow Matching, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18447, https://doi.org/10.5194/egusphere-egu25-18447, 2025.

Data-driven weather prediction models based on deep learning have been on the rise for several years and have outperformed traditional physics-based numerical models in various benchmark forecasting scores. However, a significant challenge remains: accurately predicting extreme events on a local scale, such as thunderstorms and wind gusts. Previous models struggle in this area, as they were primarily developed for medium-range forecasting and operate at relatively coarse spatio-temporal resolutions. However, the capability of weather models to predict extreme events at a local level is essential for preventing severe consequences for communities, ecosystems, and the financial and material losses they entail. Recently, task-agnostic foundation models, trained on extensive and diverse datasets using self-supervised methods, have demonstrated remarkable skill and robustness, especially in their ability to generalize to rare extreme events. 

The RAINA project aims to develop a foundation model for the atmosphere, with an emphasis on delivering reliable, high-resolution forecasts of extreme wind and precipitation events. In partnership with the EU Horizon-funded WeatherGenerator project, which aims to create advanced digital twins for Destination Earth, RAINA will extend the pioneering AtmoRep model (Lessig et al., 2023) by employing a multi-modal learning approach.
The foundation model seeks to develop a comprehensive, statistically robust, and multi-scale understanding of atmospheric dynamics by incorporating a wide range of meteorological datasets from both models and observations. Innovative deep learning methods, including diffusion models and test-time adaptation, will be investigated to facilitate short-range forecasts of temperature, wind, and precipitation at kilometer-scale resolution over Germany.

In a first demonstrator, short-range forecasts are generated using the AtmoRep model and subsequently refined with the CorrDiff downscaling approach (Mardani et al., 2024) that combines a generative diffusion model with a residual approach. This two-step strategy delivers high-resolution forecasts with a maximum lead time of six hours while disentangling uncertainties inherent in the forecasting and downscaling processes, a separation that can enhance training quality when properly applied. By using ERA5 and COSMO REA2 reanalysis data, the approach enhances the precision of high-resolution forecasts over Germany. 
Initial results from the first demonstrator will be presented in a poster, along with the overall timeline and key milestones of the RAINA project.

How to cite: Pavel, E., Langguth, M., Schultz, M. G., Lessig, C., Hollborn, S., Keller, J., Potthast, R., Seegebrecht, B., Wahl, S., Gall, J., Allaham, A., Shams Eddin, M. H., and Luise, I.: RAINA - High-resolution nowcasting of precipitation and wind extremes with a foundation model for the atmosphere, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19021, https://doi.org/10.5194/egusphere-egu25-19021, 2025.

How to cite: Delmas, C., Lambaré, A., and Le Carvennec, A.: Destination Earth Data Lake user story on Danube Delta water reservoir , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19238, https://doi.org/10.5194/egusphere-egu25-19238, 2025.

High-resolution machine learning faces the challenge of balancing local computation with large physical context windows. GPU memory limitations and the slow training process when distributing the model across multiple GPUs further complicate this task. 

We present a transformer model for high-resolution climate-related tasks that uses neural operators within a multi-grid architecture. This approach allows resolution independence, large physical context windows, and the handling of discontinuities such as coastlines.

The model is spatially flexible, supporting both regional and global training schemes. It is also independent of the number of input variables, allowing training to be scaled to large numbers of input variables.

We demonstrate the ability of the model to scale, both spatially and in terms of variables. The model forms the foundation of an approach to learn from the rich and diverse climate data available, enabling high-resolution downscaling, infilling, and predictions.

How to cite: Witte, M., Meuer, J., and Christopher, K.: A Spatial Multi-Grid Neural Operator-Transformer Mdoel for High-Resolution Climate Modeling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19942, https://doi.org/10.5194/egusphere-egu25-19942, 2025.

Digital twins are an exciting and rapidly developing research area, with the potential to provide a step change in the way we understand our evolving environment and its impact on sensitive systems.

The TWInning Capability for the Natural Environment (TWINE) programme is being co-delivered by the Met Office and Natural Environment Research Council (NERC) to explore the potential of this technology for transforming environmental science and across priority areas including climate change adaptation and mitigation, biodiversity and ecosystems, and natural hazards and their mitigation.

Through the TWINE programme, NERC and the Met Office have funded six digital twin pilot projects across a range of applications, including harmful algal blooms, flooding and coastal overtopping, optimising data collected by ocean gliders and aircraft, and multi-objective land-use decisions.

We will introduce the TWINE programme, giving a brief overview of the projects which are advancing our understanding of how we can harness the potential of digital twinning technology. These include cross-sector challenges such as: risk to natural resources, Net Zero targets, and addressing the science-to-policy lag.

How to cite: Mendes, J., Pope, E., Jones, Z., Cottrell, A., Eastman, M., Wiggs, J., Findley, H., Heard, A., Hallett, P., Vanvyve, E., Vandaele, R., Williams, H., Miltiadou, M., Gibson, F., Dale, K., Angus-Smyth, A., Gardner, S., and Tailby, S.: TWINE: TWInning capability for the Natural Environment, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20187, https://doi.org/10.5194/egusphere-egu25-20187, 2025.

Uncrewed aerial vehicles (UAVs) have been identified as an important tool supporting more detailed remote sensing applications compared to satellite-based platforms, from agriculture to forest monitoring and sensing mountains. The insights the UAV can offer, due to its flexibility, high precision and sensor variety, are far beyond the previous approaches to measure the health of forests, yield in field crops, and even rockfall risk. The flexibility however also poses a problem, flight conditions, sensor types, flight height and angles all affect the generalization of developed approaches using UAVs. These supervised approaches also rely on large amounts of human-labelled datasets. A pathway to reduce high-label requirements is to utilize unsupervised training with Vision Transformers (ViTs). Pretrained Vision Transformers on large datasets generalize well to unseen data, with only supervised few samples required to specify the application. However, these models are often trained on massive web-scraped RGB datasets. Furthermore, RGB-ViTs miss the infrared domain to handle crucial vegetation information. Finally, UAV imagery is exclusively from the aerial perspective, this is missing in existing pretraining datasets.

We present an openly available, pre-trained Vision Transformer specifically for UAV multispectral imagery across various domains. Furthermore, various downstream applications such as canopy height modelling and semantic segmentation are evaluated and compared against RGB baselines. The main contribution is the openly available training dataset, and the pre-trained models, with recipes for finetuning a task-specific head.

The dataset is built around multispectral image contributions from the ICAERUS Drone Data Analytics Library and an additional database search on Zenodo and Data in Brief (Table 1.). This is followed by a quality check after, including radiometric calibration, and spectral alignment. Furthermore, all data is quantized into 16-bit float and sliced into smaller 224x224 chips with four channels (Green, Red, Red Edge and NIR).  A summary of included datasets is presented in Table 1. DINOv2-s and DINOv2-b were chosen for the architecture as there is much available documentation and provide a state-of-the-art vision foundation model. The training was done in minibatches of size 32, for 6 days on two V100 GPUs.

Early experiments suggest that the pre-trained models outperform existing DINOv2-s and DINOv2-b pre-trained foundation models in both the clarity of the features, as well as tuned on UAV-specific tasks (canopy height modelling, and semantic segmentation).

Table 1. Included datasets for pretraining, total size on disk is 399GB

How to cite: Doornbos, J. and Babur, Ö.: Features from Multispectral Drone Data: Curating, training and distributing Transformers for all, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1534, https://doi.org/10.5194/egusphere-egu25-1534, 2025.

There is a significant growth in development and utilization of foundation models for geospatial applications. These models are trained on large scale unlabeled data and commonly evaluated on downstream tasks using labeled datasets. While this approach provides a platform to assess the performance of the model for specific downstream tasks, there has been limited effort to quantify the characteristics of the foundation model after pre-training. 

Explainable AI (XAI) approaches aim to increase the accuracy and transparency of AI models and to make their results interpretable. In the case of geospatial foundation models, it is essential to assess if the model learns the spectral, spatial and temporal properties of geospatial data, and how this learning impacts the accuracy of model predictions. 

To this end, we introduce a new global XAI benchmark for geospatial foundation models using multispectral remote sensing imagery. This benchmark contains separate tasks that allows the user to test a foundation model’s properties in the embedding space, and demonstrate whether the model has learned spectral, spatial and temporal features. The spectral task consists of a set of chips with homogeneous spatial patterns from all major land cover classes. The spatial task consists of the same data used for spectral taks but regular spatial patterns are replaced with heterogeneous features representative of their true distribution. Finally, the temporal task includes a set of chips with time series imagery of pre- and post-event for disturbances such as wildfire and flood. 

In this presentation, we will demonstrate the results of using this benchmark to evaluate the properties of multiple geospatial foundation models.

How to cite: Alemohammad, H., Khallaghi, S., Godwin, D., Balogun, R., Roy, S., and Ramachandran, R.: An Explainable AI (XAI) Benchmark for Geospatial Foundation Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3302, https://doi.org/10.5194/egusphere-egu25-3302, 2025.

NASA has collected—and continues to amass—petabytes of scientific data, ranging from the vastness of galaxies to the intricacies of cellular biology. These ever-expanding datasets provide unparalleled opportunities for discovery but pose significant challenges for managing data and extracting meaningful insights. Artificial intelligence (AI) and machine learning (ML) are emerging as transformative tools for addressing these issues. However, state-of-the-art deep neural networks often require large volumes of labeled training data, which are costly and time-intensive to generate. AI foundation models (FMs) offer a promising alternative by leveraging self-supervised learning to identify patterns within data. These FMs enable diverse applications with reduced dependence on compute resources and labeled datasets.

NASA’s Office of the Chief Science Data Officer has formulated a "5+1" strategy to develop AI foundation models for science. This strategy emphasizes creating foundation models (FMs) pre-trained using flagship datasets from each of NASA’s science divisions while building a science-specific language model to support cross-divisional applications. Key achievements include the release of INDUS, an encoder language model trained on scientific publications and technical documents; two versions of the Prithvi Geospatial model for environmental monitoring applications; and the Prithvi Weather and Climate model, designed to reconstruct atmospheric states from incomplete data and forecast future states. Additionally, a heliophysics foundation model for space weather applications is under development and is scheduled for release by mid-2025.

To encourage NASA’s research and application communities to use these FMs in their work and to support NASA’s new Earth Science to Action Strategy, the Earth Science Division has developed additional research and application solicitations to further enhance these FMs and to build applications and tools leveraging these FMs. These announcements are available in NASA’s Research Opportunities in Space and Earth Science (ROSES 2025).

NASA has forged strategic partnerships with private sector organizations, academia, and other entities grounded in open science principles to build these models. Each model is designed around a specific set of scientific use cases to ensure relevance and practical impact. All models and associated use case notebooks are shared openly. 

This presentation will provide an overview of the foundation models released to date, the workflows used in their design and development, and the roadmap for future models. It will also highlight upcoming workshops aimed at equipping the broader scientific community to effectively integrate these models into their research.

How to cite: Ramachandran, R., Lee, T., and Murphy, K.: AI Foundation Models for Science: Current Initiatives, Workflow, and Future Roadmap, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3328, https://doi.org/10.5194/egusphere-egu25-3328, 2025.

Foundation models have shown substantial potential in enhancing Earth observation by providing high accuracy while minimizing the need for manual annotation. However, the combined application of multiple foundation models for processing ground-truthing data remains largely underexplored. This study introduces a new approach for ground-level land use classification and ground-truthing by integrating a vision foundation model, the Segment Anything Model (SAM), with a general-purpose large language model (GPT-4o). Using high-resolution thermal and RGB imagery captured from human-eye height with a handheld camera, the proposed method generates object-level land use classifications and surface temperature profiles. Data collection was conducted in the Lusatia region of Germany, covering diverse land use types. SAM was utilized to segment complex landscape structures into meaningful elements such as roads, water bodies, and trees, followed by GPT-4o, which classified these segments into custom-defined land use categories. At a broad level (7 classification types), the workflow achieved approximately 80% accuracy, with high F1 scores for categories such as Road (0.89), Vegetation (0.82), and Built Structure (0.81). At a finer level (28 classification types), the method attained around 64% accuracy, effectively classifying detailed sub-classes such as Asphalt-Concrete Road (F1 = 0.85), Brick Road (F1 = 0.86), Tree (F1 = 0.74), and Arable Land (F1 = 0.68). By overlaying thermal imagery with classified segments, the method revealed distinct microclimatic patterns across land use types, with agricultural land showing the lowest surface temperatures (p < 0.001). The proposed workflow underscores the potential of combining SAM and GPT-4o to deliver robust ground-truthing data using portable cameras, advancing AI-enabled environmental monitoring.

How to cite: Mengsuwan, K. and Ryo, M.: Ground-level land surface classification and thermal analysis using foundation models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7220, https://doi.org/10.5194/egusphere-egu25-7220, 2025.

The detection of environmental changes caused by natural disasters is critical for rapid response and effective management. In this study, we present a methodology for unsupervised change detection that leverges optical Sentinel-2 [1] and Synthetic Aperture Radar (SAR) Sentinel-1 [2] accessed through public SpatioTemporal Asset Catalogs (STAC) [3] along with Earth Observation (EO) foundation model, namely, Clay [4]. The analysis was conducted independently for each dataset to capitalize on the unique properties of these satellite sensors. Sentinel-1 offers robust surface texture sensitivity with its all-weather, day-and-night imaging capability, while Sentinel-2 provides detailed spectral and spatial information critical for vegetation and land-use analysis.

Clay foundation model, a large-scale pretrained Vision Transformer trained on EO data from various missions (Sentinel-1, Sentinel-2, Landsat, Planet, NAIP, LINZ, and MODIS), was used to extract spatially and spectrally rich embedding from Sentinel-1 and Sentinel-2 images. The model takes as an input the satellite imagery along with information about location and time and outputs mathematical representations of a given area at a certain time on Earth’s surface. The images were fed to the model as patches of size 256×256 along with the timestap of the scene, the spatial location and other metadata of the input image to estimate the embeddings that can be rearranged to be of size (1024×32×32). These embeddings were then analyzed using pixel-wise distance metrics to quantify changes between pre- and post-even imagery and the resulting distance image was then spatially interpolated to the size of the input image.

The approach was validated on satellite imagery of the Valencia region in Spain, an area significantly impacted by recent flooding on the 29th October 2024. For Sentinel-1, the method effectively highlighted surface water changes and structure affected by the floods in two scenes acquired on the 7th October and the 12th November 2024, while Sentinel-2 data captured variations in vegetation areas that was impacted by the floods using two scenes acquired on the 1st October and the 10 November 2024. By analyzing the datasets independently, this framework demonstrates the complementary insights offered by radar and optical imagery in assessing disaster impacts.

This study highlights the potential of leveraging open satellite data available via STAC catalogs and EO foundation models for unsupervised change detection in disaster monitoring, enabling rapid response without relying on specialized models tailored to specific regions. Unlike traditional approaches that require retraining for new areas due to geographical variability, this methodology is both scalable and adaptable, providing a generalizable framework for environmental monitoring, disaster response, and resilience planning. The results emphasize the value of integrating multi-sensor satellite imagery to enhance understanding of disaster impacts, facilitating more informed and timely decision-making.

References:

[1] https://earth-search.aws.element84.com/v1

[2] https://planetarycomputer.microsoft.com/api/stac/v1

[3] https://stacspec.org/en

[4] https://clay-foundation.github.io/model/index.html

How to cite: Albughdadi, M., Antonacci, M., Baousis, V., Fornari, F., Kaprol, T., and Pisa, C.: Unsupervised Change Detection Using Sentinel-1 and Sentinel-2 Imagery with the Clay Foundation Model: A Case Study of Flood-Affected Areas in Valencia Spain, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8272, https://doi.org/10.5194/egusphere-egu25-8272, 2025.

With the biodiversity crisis and land use intensification, macroecological questions related to biodiversity assessment and conservation are becoming increasingly pressing. Such questions require global datasets such as satellite imagery. Traditional methods using satellite data rely heavily on supervised learning and annotated datasets, which are limited and difficult to generalize across geographical scales. In recent years, self-supervised learning (SSL) has opened the doors to learning expressive representations of massive datasets without annotations,  thus revolutionizing the analysis of remote sensing imagery. However, currently available datasets for pre-training such models have a skewed geographical distribution, focusing on cities and agricultural areas while failing to adequately represent regions of high ecological interest, such as rainforests or polar latitudes.

We propose a new Sentinel 2A (10m resolution) multiband dataset, globally distributed on a regular grid across the landmass(250k locations). At each location, the dataset captures four different seasons determined based on the local EVI-curve and includes NDVI index, which is widely used in ecological applications. Our temporal sampling is specifically designed to align with plant phenology rather than ad-hoc calendar dates. We use this data to pre-train Momentum Contrast and Seasonal Contrast SSL models that have shown similar performance on commonly-used benchmarks and advanced performance on macroecological downstream tasks, such as species distribution modelling. We anticipate that the dataset and model will be valuable for macroecological applications, such as deep species distribution modeling or large-scale biodiversity assessments.

How to cite: Plekhanova, E., Robert, D., Dollinger, J., Brun, P., Wegner, J. D., and Zimmermann, N. E.: SeCo-Eco: Global multiband seasonal pre-training dataset and self-supervised model for ecological applications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8460, https://doi.org/10.5194/egusphere-egu25-8460, 2025.

How to cite: Kluczek, M., Czerkawski, M., and Bojanowski, J. S.: Developing Global Embeddings from Sentinel-1 and Sentinel-2 Data to Enhance Earth Observation Analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9632, https://doi.org/10.5194/egusphere-egu25-9632, 2025.

 Aurora is an important manifestation of solar-terrestrial physical processes. The aurora activities have rapid changes in spatial and intensity distribution during a substorm, especially the expansion phase. In this work, a newly developed aurora evolution model is built including the aurora images prediction and the substorm expansion duration prediction. The aurora images prediction model is trained based on the Convolutional Long Short-Term Memory network, using the aurora images captured by the ultraviolet imager on the Polar satellite during the substorm expansion phases. Given the images after the onset, the model can predict the following aurora images sequences during the substorm expansion phase. However, the images prediction model works well only for 30-45 minutes, which is close to the substorm expansion phase duration. Considering this, the expansion phase duration prediction model is trained using the solar wind and interplanetary magnetic field data. Using the traditional machine learning method, the duration is predicted by inputting these physics parameters.

How to cite: Lu, Y., Jiang, J., Zhong, J., and Zou, Z.: Aurora Evolution Model During the Substorm Expansion Phase using Machine Learning based method , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12338, https://doi.org/10.5194/egusphere-egu25-12338, 2025.

AI foundation models have already demonstrated their usefulness in harnessing their potential in a wide range of science application domains. They derive their power from the large volumes of data, along with the computational methods used to exploit them using unprecedented amounts of compute power. We are inundated with data but have managed to exploit only a small fraction of the available data.  The Earth System Grid Federation (ESGF) is hosting nearly 16 PB of data collection from the Coupled Model Intercomparison Project (CMIP6), expected to grow to 5 - 10 more in the CMIP7 era. The NASA Earth Observation Data and Information System (EOSDIS) archive is expected to exceed 600 PB by 2030. 

AI-enabled solutions will require integrating multimodal data while being cognizant of the energy footprint introduced by the data and the computational methods. Currently, the energy consumption of transformer-based foundation models scale with the amount of data and corresponding model sizes. This impediment needs to be mitigated by developing data-efficient methods that lead to energy efficiency as well across all scales. There is little guidance in the research community on developing a computational plan for the optimal use of the resources for developing foundation models using multimodal scientific data. The benchmarks based on LLM scaling are still insufficient for vision transformers (ViTs), commonly adopted for geoscientific applications. We need a suite of community benchmarks based on ViT backbones and other methods at different scales to understand energy efficient methods for different classes of science problems.

Relatively few studies have focused on the issue of data efficiency for training science foundation models. We have adopted a smart sampling approach to extract the most informative samples is an effective means of significantly reducing the training data. We trained two ViT models, one with all available MODIS data over the ocean and another using an intelligently-sampled subset. We applied the models to classify clouds over the ocean.  Our preliminary results indicate that reasonably accurate models can be trained with only a fraction of total training data. Improvements in reduction of data translate directly into improvements for energy efficiency. 

How to cite: Anantharaj, V., Kurihana, T., Padovani, G., and Fiore, S.:  Data efficiency: The master key for unlocking energy efficiency, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14068, https://doi.org/10.5194/egusphere-egu25-14068, 2025.

Weather Foundation Models in general represent a significant advancement in computational weather prediction by leveraging data-driven techniques to improve accuracy and speed. This study evaluates the Prithvi WxC foundation model for weather and climate by systematically assessing its adherence to fundamental physical constraints governing atmospheric processes. While traditional model validation primarily focuses on error statistics of modeled fields, this work takes a more comprehensive approach, incorporating a series of process-based tests to ensure the model's consistency with key atmospheric principles.

We first test the model's compliance with conservation of mass, ensuring that it respects the fundamental principle that mass is neither created nor destroyed within the atmosphere. Next, we examine the model's representation of geostrophic balance, critical for large-scale flow, by evaluating the relationship between the pressure gradient and the Coriolis force. The hypsometric equation is also applied to assess the vertical consistency of the model’s simulations, verifying that changes in pressure are appropriately related to temperature and height. To further evaluate large-scale flow dynamics, we analyze the model’s consistency with thermal wind relations, ensuring that temperature gradients are correctly reflected in the vertical wind profile. Finally, we tested our model on radiative and convective parameterization by comparing its performance of convection to established methods in conventional weather models, testing its ability to properly simulate convective processes. 

The results of these tests highlight the Prithvi WxC model's strengths and areas for improvement in terms of physical consistency. By adhering to these atmospheric principles, the findings of this study offer valuable insights into how the model can be refined, enhancing its potential applications in both weather forecasting and climate research.

How to cite: Kumar, A., Roy, S., Nair, U., Maskey, M., and Ramachandran, R.: Comprehensive Validation of the Prithvi WxC FM Through Atmospheric Process Analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14422, https://doi.org/10.5194/egusphere-egu25-14422, 2025.

Few hundred million to billion parameters of autoregressive transformer-based weather foundation models (FMs) have demonstrated generalizabilities for various downstream applications such as regional forecasting to downscaling. They occasionally outperform traditional physics-based models for medium-range forecasting skills as well as enable significantly faster execution speeds. These ML approaches are designed for timeframes ranging from hourly to up to 10 days, and sub-seasonal forecasting, defined as a range spanning two weeks to two months, often receives less attention for their downstream tasks due to the inherent challenges in predicting the chaotic nature of atmospheric systems. However, the sub-seasonal to seasonal forecast has socio-economic impacts influencing actions from seasonal extreme weather events and economic activities. While community standards for benchmarking studies have been conducted for the medium-range forecasts, the benchmarking of sub-seasonal forecasts still needs further efforts. In this study, we are aiming to fine-tune foundation models to predict sub-seasonal forecasts for various variables to conduct comprehensive benchmarking for weather foundation models. Particularly, to reduce the complexity of tasks, our fine-tune task forecasts two-week averaged atmospheric variables with a forecasting lead-time of two weeks. For this task, we resample the community standard dataset, WeatherBench, for the two-week averaged dataset. We primarily work with the Oak Ridge Base Foundation Model for Earth System Predictability (ORBIT), and extend the benchmarking to other FMs across Aurora, ClimaX, and Prithvi WxC models. Our initial fine-tuning task uses a 100 million parameters ORBIT model to predict geopotential height at 200 hPa with two-week lead time, a key indicator for extreme precipitation in Central Southeast Asia. The preliminary results demonstrate that the fine-tuned ORBIT predicts realistic spatial distributions achieving an MSE of 24.32 m when evaluated against the 2018 data. The comprehensive sub-seasonal forecasting benchmarking can highlight the potential of weather FMs whether they capture underlying principles of atmospheric dynamics, thereby enabling their performance to be extended to longer forecast lead-times. 

How to cite: Kurihana, T., Anantharaj, V., and Ashfaq, M.: Fine-tuning Foundation Models for Benchmarking Prediction Skills for Sub-seasonal Forecasting , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14848, https://doi.org/10.5194/egusphere-egu25-14848, 2025.

Recent advancements in AI and Self-Supervised Learning (SSL) have revolutionized large-scale computer vision models, enabling exceptional performance on downstream tasks in remote sensing with minimal labelled data. These models are pre-trained on large amounts of unlabelled data and, then, fine-tuned on specific downstream Earth Observation (EO) applications [1-3]. The scarcity of large-scale labelled datasets and the technical challenges of annotating the vast volumes of data collected by satellites pose significant barriers to achieving high accuracy in many important downstream tasks, which require extensive labeling at a large scale to be effective. Furthermore, the dynamic nature of Earth adds complexity, as labels at a particular moment in time are not enough. 

Consequently, SSL and Foundation Models offer a powerful solution to these challenges. By pre-training a Foundation Model on extensive unlabelled data, only a small amount of labelled data is required for supervised fine-tuning on downstream tasks. This approach reduces the need for labelled EO data. This enables the development of general-purpose EO Foundation Models capable of solving a diverse range of problems. 

In this work, we train the EO Foundation Model PhilEO Version 1.0 [2] on the dataset MajorTOM [6]. The model is trained on Sentinel-2 data, i.e. 23TB, Core-S2L2A. We scale up the pre-training data from less than 1TB in the dataset PhilEO-Globe [7] to 23TB. The model is trained using a combination of reconstruction and auxiliary task losses, including the Mean Squared Error (MSE) for geo-location longitude and latitude prediction. The architecture is a modified U-Net. The training is conducted on the Leonardo Davinci-1 supercomputer. 

In this work, we also extend the capabilities of the evaluation framework for EO Foundation Models we recently introduced in [2]. We develop PhilEO-Bench++. To take advantage of multi-level features and of a U-Net-like architecture, for fine-tuning on downstream tasks, we use the decoder UPerNet [4]. Furthermore, to strengthen the evaluation of EO Foundation Models, we also perform confidence quantification and assessment [5] on both classification and regression tasks, including on land cover semantic segmentation and building density pixel-wise regression. 

Experiments on the PhilEO-Bench downstream tasks of building density estimation, road segmentation and land cover mapping demonstrate the effectiveness of our model. For building density regression, for n-shots n=50 and n=100, the PhilEO model trained on MajorTOM achieves the MSEs 0.0191 and 0.0058, respectively.

References:  

[1] D. Szwarcman, et al., “Prithvi-EO-2.0: A Versatile Multi-Temporal Foundation Model for Earth Observation Applications,” arXiv:2412.02732, 2024. 

[2] C. Fibaek, et al., “PhilEO Bench: Evaluating Geo-Spatial Foundation Models,” in Proceedings IGARSS, 2024. 

[3] N. Dionelis, et al., “Evaluating and Benchmarking Foundation Models for Earth Observation and Geospatial AI,” arXiv:2406.18295, 2024. 

[4] T. Xiao, et al., “Unified Perceptual Parsing for Scene Understanding,” 2018.

[5] N. Dionelis and N. Longepe, “Fine-Tuning Foundation Models with Confidence Assessment for enhanced Semantic segmentation,” 2024. 

[6] A. Francis and M. Czerkawski, “MajorTOM: Expandable Datasets for Earth Observation,” IGARSS, 2024. 

[7] B. Le Saux, et al., “The PhilEO Geospatial Foundation Model Suite,” EGU, 2024.

How to cite: Dionelis, N., Musto, R., Paoletti, G., Bosmans, J., Naylor, P., Sarti, S., Di Matteo, F., Cascarano, G., Fibaek, C., and Longepe, N.: Generalist Geospatial Foundation Model PhilEO on Satellite Sentinel-2 MajorTOM Multi-Spectral Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16635, https://doi.org/10.5194/egusphere-egu25-16635, 2025.

How to cite: Vinge, R., Marszalek, M. L., Schneider, J., and Albrecht, C. M.: Earth Observation embeddings at the test: A novel benchmark to evaluate (neural) compression for satellite imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16756, https://doi.org/10.5194/egusphere-egu25-16756, 2025.

Food insecurity is typically modelled using inter-regional data comprising economical, geophysical and social variables. Such datasets are often of varying granularity, with each variable corresponding to a certain granularity level (e.g., GDP is a national variable, while disaster displacement can be local or regional). Additionally, each level shows a specific causal relation of its variables.  Since countries affected by food insecurity are usually underdeveloped, collecting such variables is a challenging task, leading to highly-incomplete datasets. To deal with the multi-level complexity and incomplete nature of the data, we propose to build a hierarchical causal graph (HCG) structure of the variables, that can then be injected in different imputation methods. Specifically, we propose to classify the variables at different granularity levels, and use causal graph discovery to learn a causal graph at each level. We test the proposed approach for imputing food insecurity using a dataset of 300+ economical, geophysical and social variables for more than 70 countries.

How to cite: Rodrigo-Bonet, E., Cerda, J., Ronco, M., and Camps-Valls, G.: Hierarchical Causal Graph-Based Methods for Imputing Food Insecurity, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17133, https://doi.org/10.5194/egusphere-egu25-17133, 2025.

AI foundation models hold considerable promise for leveraging the vast and diverse datasets available in atmospheric and geoscientific research. These models have the potential to advance scientific discovery by capturing complex spatial and temporal relationships inherent in earth system processes. However, the development and deployment of such models is often hindered by limited computational resources.

Accurate reconstruction of fine-scale atmospheric features from coarse-resolution data is a critical challenge in geoscientific modeling, as well as a benchmark for understanding the performance of climate-related models. High-resolution atmospheric data are essential for capturing localized phenomena, such as convective systems, topographic effects, and land-atmosphere interactions, that influence weather patterns and climate processes. However, the generation and storage of high-resolution datasets are computationally expensive, necessitating methods that can infer fine-scale structures from lower-resolution observations.

The primary objective of this study is to validate PrithviWxC [4], a ViT-based foundation model [1], for the task of downscaled image reconstruction. While Prithvi WxC was trained on 160 atmospheric variables from the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) dataset [2], we implement a smaller version of the initial model, with 21 million parameters, and pretrained on the same dataset with a set of six variables. 

The evaluation process involves the assessment of the model's capacity to reconstruct fine-scale features through the process of downscaling atmospheric data. In this procedure, inputs at a high spatial resolution of 1 km from [5] are first coarsened to 25 km resolution, the same as European Centre of Medium-range Weather Forecasts Reanalysis v.5 (ERA5) [3], and subsequently upscaled to recover the original fine-grained structure. This process serves as a benchmark for assessing the model's capacity to learn and preserve spatial details during resolution transformations, which is an essential requirement for geoscientific modeling tasks.

We fine-tune the model for downscaled image reconstruction on a set of 30784 128x128 patches, and validate its output, produced after learning on a limited temporal period, on tiles coming from the ERA5 dataset, which encompass all seasonality. In particular, we aim at highlighting the model's ability to generalize to data domains beyond its pretraining distribution, demonstrating its adaptability and the transferability of knowledge embedded within ViT architectures. By applying PrithviWxC to a knowledge domain that is distinct from its original training context, we demonstrate the potential for cross-domain learning in geoscientific applications.

REFERENCES

[1] Dosovitskiy, Alexey. "An image is worth 16x16 words: Transformers for image recognition at scale." arXiv preprint arXiv:2010.11929 (2020).

[2] Gelaro, Ronald, et al. "The modern-era retrospective analysis for research and applications, version 2 (MERRA-2)." Journal of climate 30.14 (2017): 5419-5454.

[3] Hersbach, Hans, et al. "The ERA5 global reanalysis." Quarterly Journal of the Royal Meteorological Society 146.730 (2020): 1999-2049.

[4] Schmude, Johannes, et al. "Prithvi wxc: Foundation model for weather and climate." arXiv preprint arXiv:2409.13598 (2024).

[5] Wedi, Nils P., et al. "A baseline for global weather and climate simulations at 1 km resolution." Journal of Advances in Modeling Earth Systems 12.11 (2020): e2020MS002192.

How to cite: Padovani, G., Kumar, A., Kurihana, T., Fiore, S., and Anantharaj, V.: PrithviWxC Foundation Model Validation on Weather Downscaling for Cross Domain Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17360, https://doi.org/10.5194/egusphere-egu25-17360, 2025.

Foundation Models are emerging as a transformative paradigm in Earth observation, offering powerful solutions to the challenges of processing and understanding satellite imagery at scale. The scarcity of large-scale labeled datasets and the technical challenges of annotating the vast volumes of data collected by satellites pose significant barriers to achieving high accuracy in many important downstream tasks. Furthermore, the dynamic nature of Earth adds complexity, as labels tied to a specific geographical region at a particular moment in time are insufficient to capture the evolving characteristics of the environment. Self-supervised learning techniques have emerged as a promising solution, enabling models to learn rich representations from unlabeled data while requiring minimal supervised fine-tuning for specific applications.
In this work, we present GeoDINO, a novel foundation model that adapts the DINO self-supervised learning architecture to handle multi-spectral Sentinel-2 data. While the original DINO framework has shown remarkable success in computer vision tasks through its teacher-student architecture and self-distillation approach, we extend it significantly for Earth observation applications. Our key innovation lies in the addition of multiple supervised auxiliary tasks: after the encoder generates representations, we attach specialized MLPs designed to predict various geospatial attributes including climate zones, permanent water bodies and geographical coordinates. Both the teacher and student networks are trained to predict these auxiliary labels, with the teacher network being updated through Exponential Moving Average (EMA) of the student's weights. This modification enables our model to learn not only from the self-supervised distillation process but also from the rich spatial and temporal information inherent in satellite imagery.
We are currently training GeoDINO on MajorTOM, a comprehensive Sentinel-2 dataset comprising 23TB of Core-S2L2A data, exploiting the Leonardo Davinci-1 Supercomputer. Furthermore, to validate our approach, we are also training the model on FastTOM and TinyTOM, two subsets of MajorTOM. Finally, the model will be evaluated within the PhilEO Bench framework to assess its performance on different tasks, including land cover classification, change detection, and building density estimation. Looking ahead, we plan to transition to the DINOv2 architecture to further enhance our model's capabilities. Through this research, we aim to demonstrate how self-supervised learning techniques, when properly adapted for Earth observation data, can address the fundamental challenges of data scarcity and temporal dynamics in remote sensing applications. The development of GeoDINO represents a step toward more efficient and adaptable Earth observation systems that can leverage the vast amounts of available satellite data while minimizing the need for extensive labeled datasets.
References:  
[1] M. Caron, et al., “Emerging Properties in Self-Supervised Vision Transformers”, arXiv:2104.14294, 2021
[2] C. Fibaek, et al., “PhilEO Bench: Evaluating Geo-Spatial Foundation Models,” in Proceedings IGARSS, 2024. 
[3] N. Dionelis, et al., “Evaluating and Benchmarking Foundation Models for Earth Observation and Geospatial AI,” arXiv:2406.18295, 2024. 
[4] N. Dionelis and N. Longepe, “Fine-Tuning Foundation Models with Confidence Assessment for enhanced Semantic segmentation,” 2024. 
[5] A. Francis and M. Czerkawski, “MajorTOM: Expandable Datasets for Earth Observation,” IGARSS, 2024. 
[6] B. Le Saux, et al., “The PhilEO Geospatial Foundation Model Suite,” EGU, 2024.

How to cite: Musto, R., Paoletti, G., Dionelis, N., Sarti, S., Di Matteo, F., Bosmans, J., Naylor, P., Donato Cascarano, G., Fibaek, C., and Longépé, N.: GeoDINO: A Vision Foundation Model for Earth Observation Leveraging DINO Architecture and Sentinel-2 Multi-Spectral Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18029, https://doi.org/10.5194/egusphere-egu25-18029, 2025.

Earth observation (EO) yields large-scale, multimodal datasets collected from various satellite missions, DEMs, land-use data, and textual metadata. Foundation models like 4M (Massively Multimodal Masked Modeling) can learn a joint embedding space that bridges modality gaps, mitigates missing data issues, and facilitates partial spatio-temporal alignment [1]. However, directly training such foundation models on the vast, high-dimensional original EO datasets is not only computationally intensive but also imposes substantial demands on storage resources.

To address this, one can leverage VQ-VAE (Vector Quantized-Variational AutoEncoder) as neural compressors to transform high-dimension multimodal inputs into a few discrete indices, significantly reducing data volume while preserving critical information. By inverting the tokenization process, we can reconstruct the original high-dimensional data with minimal quality loss, aided by adversarial and perceptual losses that enhance reconstruction fidelity.

Traditional VQ-based approaches, however, face challenges such as inefficient codebook utilization and limited latent space representation. To overcome these, we propose scaling strategies that complement 4M’s tokenizer-based architecture. By expanding the codebook size, latent dimensions, and network depth, our method captures the complexity of EO modalities more effectively. Specifically, we employ spherical quantization techniques like Grouped Spherical Quantization (GSQ) to address limitations in traditional approaches [2]. GSQ constrains codebook vectors to a spherical surface, stabilizing training, preventing code collapse, and promoting uniform codebook usage. Unlike standard VQ, GSQ uses spherical initialization and normalization to maintain consistent distances among codebook entries, ensuring robust latent space coverage even under extreme compression or large codebooks. From our empirical and ablation studies, alternative methods like LFQ (Lookup-Free Quantization), FSQ (Finite Scalar Quantization), and RVQ (Residual Vector Quantizer) often exhibit limitations, such as tightly coupling the latent dimension to codebook size or relying on specialized training losses. In contrast, spherical-based techniques effectively decouple latent dimensions from codebook vocabulary, providing greater flexibility and scalability as data demands increase.

Our approach enables neural compressors to adapt to varying scales of compression and complexity without compromising performance. Comprehensive scalability experiments—examining large codebooks, deeper networks, and diverse compression ratios—assessed the generalizability of the proposed compression strategies and demonstrated their effectiveness on high-dimensional, large-scale EO data with minimal information loss. By integrating advanced compression techniques with scalable architectures, this framework establishes a robust foundation for addressing multimodal challenges in EO research that significantly reduces the difficulty of training foundation models on multimodal high-dimensional EO data.

References

[1] Mizrahi, D., Bachmann, R., Kar, O. F., Yeo, T., Gao, M., Dehghan, A., & Zamir, A. (2023). 4M: Massively Multimodal Masked Modeling (Version 1). arXiv. https://doi.org/10.48550/ARXIV.2312.06647

[2] Wang, J., Qin, Z., Zhang, Y., Hu, V. T., Ommer, B., Briq, R., & Kesselheim, S. (2024). Scaling Image Tokenizers with Grouped Spherical Quantization (Version 2). arXiv. https://doi.org/10.48550/ARXIV.2412.02632

Acknowledgments

This work is performed in the Embed2Scale (Earth Observation & Weather Data Federation With AI Embeddings) project, funded by the EU’s Horizon Europe program under Grant Agreement number 101131841.

How to cite: Scheurer, E., Wang, J., Sedona, R., Maurogiovanni, S., Blumenstiel, B., Jakubik, J., Fraccaro, P., Brunschwiler, T., Kesselheim, S., and Cavallaro, G.: Scalable Efficient Compression in Large-Scale Earth Observation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19016, https://doi.org/10.5194/egusphere-egu25-19016, 2025.

Hyperspectral imagery (HSI) provides rich spectral information that is the basis for applications such as mineral mapping, trace gas identification, and precision agriculture. Yet, the development of HSI Foundation Models (FMs) is less advanced compared to multi-spectral remote sensing modalities.

In this study, we leverage the SpectralEarth dataset [1] to explore practical aspects of training robust HSI FMs. In particular, we shed light on the role of:

• the impact of model architecture (transformers vs. convolutional networks),
• self-supervised learning methods (contrastive vs. masked autoencoders),
• model size & training data volume,
• and the resulting computational requirements.

Through extensive experiments, this study aims to provide concrete guidelines for the development and effective application of FMs in the HSI domain. Moreover, we report on findings to identify downstream applications where hyperspectral imagery has an edge over multi-spectral photos [2], and where such an advantage is less likely to expect.

References

[1] Braham, Nassim Ait Ali, et al. "SpectralEarth: Training Hyperspectral Foundation Models at Scale." arXiv preprint arXiv:2408.08447 (2024)

[2] Bangalore, Ranjini, et al. "Hyperspectral foundation model trained by spectral reconstruction for greenhouse gas emission estimation", annual meeting of the American Geophysical Union (2024)

How to cite: Albrecht, C., Gonzalez, R., Braham, N. A. A., Bangalore, R., and Brunschwiler, T.: A Practical Guide to Hyperspectral Foundation Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19131, https://doi.org/10.5194/egusphere-egu25-19131, 2025.

Understanding the global carbon budget with its carbon sources and sinks is scientifically important and economically relevant. In particular, vegetation and soils are major and highly dynamic carbon pools in the Earth System and a substantial part of the terrestrial carbon budget is influenced by land use changes, vegetation dynamics, and soil processes.

Recent advances in Foundation Models (FMs) are transforming AI, enabling remarkable generalization and zero-shot learning capabilities. Within the Helmholtz Foundation Model Initiative, we are developing the 3D-ABC FM, a tool targeting the accurate mapping of above- and below-ground carbon stocks in vegetation and soils at high spatial resolution. 3D-ABC aims to provide a seamless understanding of terrestrial carbon distribution by integrating multimodal remote sensing, climate, and elevation datasets, and addressing complex challenges such as multi-dimensionality and multi-resolution in FMs. Our unique 3D-ABC partnership brings together key capacities from the domains remote sensing, carbon monitoring, AI, and high-performance computing to take on such FM development.

The 3D-ABC FM integrates large-scale remote sensing data, including multispectral satellite imagery from the Harmonized Landsat-Sentinel-2 (HLS) dataset, TanDEM-X InSAR coherence data, and 3D lidar data from space (GEDI, ICESat 1&2), aircraft, and ground-based platforms. We also aim to incorporate ERA-5 Land climate reanalysis information, GLO-30 digital elevation data, as well as local lidar and field data on vegetation, soils, and carbon flux parameters. High-resolution forest models will be used to benchmark carbon fluxes.

To accommodate the diverse data modalities assembled for 3D-ABC and to address eight selected downstream tasks, the AI model employs an adaptive architecture, integrating a multi-modal input processor, an FM encoder, an adaptive fusion neck, and task-specific prediction heads. The multi-modal input processor handles data with varying spectral dimensions, automatically mapping inputs to a unified feature space. The FM encoder extracts generalized deep features from the normalized inputs, which are then integrated into universal feature representations through the adaptive fusion neck. This fusion enhances interactions across modalities. Finally, the universal features are decoded into various outputs tailored to the specific needs of downstream tasks. In the first FM training phase, a pretraining strategy leverages a masked autoencoder to train the multi-modal input processor, the encoder, and the fusion neck in an unsupervised manner, enabling the model to develop robust representation capabilities. In the second phase, by leveraging the principles of transfer learning, the pretrained model is fine-tuned using labeled data from various downstream tasks.

3D-ABC targets use of the JUWELS Booster and JUPITER high-performance computing (HPC) systems located at the Jülich Supercomputing Centre (JSC). The JUWELS Booster comprises 936 compute nodes, each equipped with four NVIDIA A100 GPUs. JUPITER, the first European exascale supercomputer, is currently being installed at JSC. Its Booster module will consist of ~6,000 compute nodes, each featuring four NVIDIA GH200 GPUs. To maximize efficient JUPITER utilization, 3D-ABC is leveraging the JUPITER Research and Early Access Program, which provides early access for code optimization and preparation to ensure FM applications are optimized and ready for deployment when the system becomes operational in 2025.

How to cite: Grosse, G., Ghamisi, P., Cavallaro, G., Herold, M., Huth, A., and Hajnsek, I. and the 3D-ABC Team: Towards a Foundation Model for Global Terrestrial 3D Above and Below Ground Carbon Stock Mapping (3D-ABC), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19378, https://doi.org/10.5194/egusphere-egu25-19378, 2025.

AI-based weather emulators have begun to rival the accuracy of traditional numerical solvers, for a fraction of the computational cost. The question of whether they can be reliably deployed in all use cases (e.g., for the forecast of extreme scenarios), however, is still open. We outline an ensembling strategy based on architectural variations of the Prithvi WxC foundation model (FM), highlighting the impact of each of these variations on physical accuracy and ability to capture the distributional extremes. A simple of ensemble of 100 models is sufficient to observe the complex mapping between configuration parameters and the forecast sensitivity of different atmospheric variables. We characterize some features of this mapping and connect them to the task of predicting various weather extremes.

How to cite: Bentivegna, E., Anantharaj, V., Schmude, J., Roy, S., Kumar, A., Lin, A., Shivanand, S., Papamarkou, T., Allmendinger, R., Maskey, M., and Ramachandran, R.: From architecture to atmospheric sensitivity: studying forecast uncertainty with Prithvi-WxC, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19821, https://doi.org/10.5194/egusphere-egu25-19821, 2025.

The growing interest in deep learning and large language models (LLMs) in recent years highlights their remarkable adaptability and ability to generalize, drawing researchers from a wide array of disciplines. Despite their promise, in many instances, these advancements have exposed a lack of transparency and rigor during development processes. Although this rapid pace of research undoubtedly offers numerous benefits, it has also led to an increasing prevalence of works conducted without rigor and in a superficial way. Code that is not accompanied by documentation and results that are not reproducible inevitably lead to confusion among researchers and an environment in which trust is not a fundamental aspect of the proposed work. The complexity of data manipulation, characterized by ad hoc transformations, exacerbates these issues by hindering the traceability of processes, and hyperparameter tuning introduces additional difficulties, requiring repeated experimentation that consumes excessive computational resources, especially for large models. 

To address these challenges, we introduce yProv4ML, a python library which provides an accessible option for tracking dataset and model statistics, hyperparameters, and energy metrics. It allows for the comparison of sets of experiments, and introduces a suite of directives to easily track the flow of information through provenance metadata. 

yProv4ML is a component of the yProv framework, a research project on multi-level provenance management which provides scientists with a rich software ecosystem consisting of a web service to manage track and analyze provenance documents. Leveraging the PROV-JSON standard for provenance artifact recording, yProv4ML ensures comprehensive documentation and reproducibility while facilitating a seamless integration process similar to well-known libraries such as MLFlow.

During the last year, yProv4ML was integrated in a variety of use cases in different domains (i.e., Climate Science, High Energy Physics and Earth Observation) in the context of the interTwin (https://www.intertwin.eu/) and ClimateEurope2 (https://climateurope2.eu/) EU projects, as well as the ICSC Italian National Project (https://www.supercomputing-icsc.it/en/icsc-home/). The collection of provenance data in these use cases not only helped facilitate the reproducibility of experiments, but also helped diagnose performance bottlenecks and ensure the reliability and integrity of results, all of which are critical to advancing the field of large-scale ML in a trustworthy manner.

How to cite: Fiore, S., Padovani, G., Kurihana, T., Fronza, M., and Anantharaj, V.: Enabling Seamless Provenance Collection in Large-Scale Machine Learning Tasks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19892, https://doi.org/10.5194/egusphere-egu25-19892, 2025.

Accurate hurricane intensity estimation is critical for disaster preparedness, yet remains challenging for weather models trained on coarse-resolution datasets. This study proposes a hybrid approach that integrates NASA-IBM's Prithvi WxC model with a deep learning-based Hurricane Intensity Estimation (HIE) model. While the Prithvi WxC model excels in global atmospheric predictions, its coarse-grained outputs can struggle with precise hurricane intensity estimation. To address this, the HIE model is triggered when it identifies a hurricane in the Prithvi model output, providing corrected intensity predictions based on high-resolution data.

A dataset was created for training and evaluation, consisting of 6,000 unique initial conditions from 1980 to 2024 that resulted in hurricanes across all major basins. Ground truth hurricane tracks and intensity data were obtained from the HURDAT database The training phase focused on hurricane cases from 1980 to 2000, building a foundational understanding of global hurricane characteristics. Subsequently, the model was fine-tuned with 2000–2020 data to account for basin-specific variations and improve regional accuracy. The remaining cases (2020–2024) are reserved for validation and assessment. The HIE model employs advanced deep learning techniques to refine key intensity metrics, such as maximum sustained wind speeds and central pressure. By addressing the limitations of Prithvi WxC's coarse-resolution training data, the HIE model achieves greater precision, leveraging fine-grained atmospheric and oceanographic features. This two-step framework, hurricane detection by Prithvi WxC followed by intensity refinement by the HIE model, capitalizes on the strengths of both models to deliver improved predictions.

This highlights the potential of combining foundation models like Prithvi WxC with specialized deep-learning frameworks to overcome existing limitations in hurricane intensity estimation. By incorporating diverse data sources and leveraging modern machine-learning techniques, this hybrid approach bridges the gap between coarse-grained global models and the need for precise regional forecasting.

How to cite: Roy, S., Kumar, A., Lal, R., Nair, U., Maskey, M., and Ramachandran, R.: Integrating Prithvi WxC with a Hurricane Intensity Estimation Model for Accurate Hurricane Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20521, https://doi.org/10.5194/egusphere-egu25-20521, 2025.

The classification of near-Earth plasma regions, i.e., distinguishing the region in which a spacecraft is located at any given time, is beneficial for both understanding the dynamics of the interaction between the Earth’s magnetosphere and the solar wind, and for modeling the characteristic boundaries separating these regions. We use measurements from the THEMIS B spacecraft between 2008 and 2010 (340 days in total) with a time resolution of one minute. The data include solar wind velocity and density, magnetic field magnitude, and standard deviation of magnetic field magnitude calculated over one-minute intervals. These data are used for manual labeling of four distinct plasma regions: solar wind, foreshock, magnetosheath, and magnetosphere. Ion energy flux data are used to classify the foreshock, if necessary. An automated classification of the respective regions based on measured plasma and magnetic field parameters is then achieved using either neural network or random forest classifiers. The performance of these classifiers is evaluated and compared. Generally, very high accuracy is achieved, but distinguishing between solar wind and foreshock remains an issue.

How to cite: Aghabozorgi Nafchi, M., Pi, G., Nemec, F., Tsai, T.-C., and Lee, K.-H.: Region identification in spacecraft data using supervised machine learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5082, https://doi.org/10.5194/egusphere-egu25-5082, 2025.

The Waves of HIgh frequency and Sounder for Probing Electron density by Relaxation
(WHISPER) instrument, is part of the Wave Experiment Consortium (WEC) of the ESA
CLUSTER II mission. WHISPER is designed to measure the electric field fluctuation and derive the electron density, i.e. the plasma density, a key parameter of scientific interest for
magnetospheric and near-Earth solar wind studies. The electron density is the WHISPER highest level product and is provided, among other products, to the scientific community through the CLUSTER Science Archive (CSA).
The instrument consists of a receiver, a transmitter, and a wave spectrum analyzer. It delivers both ambient (in natural mode) and active (in sounding mode) electric field spectra. The characteristic signatures of ambient plasma waves or active plasma resonances, combined with the spacecraft position, reveal the different magnetosphere regions. These spectral signatures are used to derive the electron density. Until recently, ad-hoc algorithms have been used to derive the electron density from WHISPER measurements, but at the cost of time-consuming manual steps. These algorithms are dependent on measurements provided by other instruments onboard CLUSTER, thus introducing dependencies and potential delays in the data production.

In this context, the goal of this work is to significantly reduce human intervention by fully
automating the WHISPER electron density derivation, exclusively using WHISPER data.
For this purpose, we develop a two-step derivation process, based on neural networks: first, the plasma region is identified with a Multi-Layer Perceptron classification algorithm; second, the electron density is derived using a Recurrent Neural Network, adapted to each plasma region. These networks have been trained with WHISPER spectra and electron density previously derived from ad-hoc algorithms. The resulting accuracy is up to 98% in some plasma regions. This derivation process has been implemented in a production pipeline, now routinely used to deliver WHISPER electron density to the CSA and dividing by 10 the human intervention. The pipeline has already delivered 3+ years of data and will be used to reprocess some of the archive focusing on the most complex plasma regions with recent improvements. This work will present the implemented methods and models for each region focusing on results and performance. 

How to cite: De Leon, E., Vandevoorde, M., Vallieres, X., and Henri, P.:  Automatic detection of the electron density from de WHISPER instrument onboard CLUSTER II, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6747, https://doi.org/10.5194/egusphere-egu25-6747, 2025.

The necessity to limit budget, size, weight and power consumption of the instruments placed on board space mission satellites results in several drawbacks, including the exclusion of dedicated instrumentation for the monitoring of the spacecraft environment. Understanding the environmental conditions of space missions is essential to correctly analyse their observations. Seldom the necessary interplanetary parameters, not measured in situ, can be gathered from nearby dedicated missions, however this is not always feasible. Other solutions envisage the application of machine learning models to estimate the missing parameters on the basis of those that are available on board the satellites. Despite the high performance of machine learning predictors, they come along with issues related to the model selection and training, the data pre-processing and the opaqueness of the outcomes returned to end-users. The application of tools developed in the explainable artificial intelligence (XAI) field can be considered to encode through symbolic knowledge the functional relationship between parameters observed in situ and correlated parameters for which measurements are lacking but useful. In this context, XAI methods in general, and symbolic knowledge extraction in particular, constitute a promising alternative to traditional machine learning models, enabling users to avoid the model selection and training phases and to obtain completely interpretable results. This presentation provides an overview on the application of symbolic knowledge-extraction techniques to perform rule induction from available in-situ data, aimed at carrying out a human-interpretable estimation and forecasting of missing platform parameters. Potentialities, drawbacks and challenges of this approach are discussed to highlight the direction from current results to future applications.

How to cite: Sabbatini, F. and Grimani, C.: Missing Interplanetary Data Estimation for Space Missions via Symbolic Rule Induction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7486, https://doi.org/10.5194/egusphere-egu25-7486, 2025.

Many machine learning models have provided significant results in predicting the geomagnetic activity quantified by Earth-measured geomagnetic indices. For instance, one such model is the SERENADE model that provides probabilistic forecasts of the Kp index, days ahead solely from solar imaging. It consists of three modules combining convolutional, recurrent, and linear neural network layers that first extract the important information contained in the input solar imagery and transform them into an intelligible forecast. To improve the performance of this model, we evaluate solar-imaging-adapted dimensionality reduction techniques that extract the features from the images and can therefore be used as the first layer of the forecast model. We use a solar imagery dataset formatted specifically for machine-learning research (SDOML). We applied the Principal Component Analysis method and trained AutoEncoders and Variational AutoEncoders (VAE) targeting several reduced dimensions. We consider the convolutional GoogLeNet method, which was pre-trained on the ImageNet dataset, as a baseline for our comparison. We analyze the information retained by the extracted features in terms of solar activity physical parameters and find high correlations between the latter and the the reduced representations of the images, with the VAE results standing out. In addition, we re-train the SERENADE model to predict the daily maximum of the Kp index two days in advance using the extracted features by the new dimensionality reduction methods as input to the model. We first use the same hyperparameters that were optimized for the GoogLeNet model and obtain more stable predictions using the dedicated solar imaging feature extractors than when using the baseline model, specifically in the VAE case. Furthermore, when fine-tuning SERENADE's hyperparameters to the VAE model, the predictive performance of the model was enhanced, notably during geomagnetic storms, which indicates that the use of adapted feature extractors could improve the geomagnetic activity forecasting.

How to cite: Tahtouh, M., Bernoux, G., and Brunet, A.: Evaluating Solar Imaging Feature Extraction Techniques for Enhancing Space Weather Prediction with Deep Learning Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8698, https://doi.org/10.5194/egusphere-egu25-8698, 2025.

Solar flares, sudden bursts of electromagnetic energy originating from magnetically active regions on the solar surface, pose significant risks to satellite infrastructure, communication systems, and power grids. Accurate forecasting of these events is crucial for advancing space weather prediction and safeguarding technological infrastructure. The interconnected nature of the Sun's atmospheric layers—from the corona to the lower photosphere—highlights the need for comprehensive data analysis techniques that leverage modern advancements in machine learning (ML) and physically informed models.

Traditional approaches have relied on features extracted from line-of-sight (LoS) magnetograms of solar active regions, historically linked to increased flare activity. However, recent studies employing LoS magnetogram time series have shown limited improvements, prompting the need for novel methodologies that integrate learning-based and physics-based insights.

To address this challenge, we present a deep learning-based framework for solar flare forecasting, leveraging the Solar Dynamics Observatory’s Helioseismic and Magnetic Imager (SDO/HMI) LoS magnetograms. Our model frames flare forecasting as a binary time series classification problem, aiming to distinguish active regions likely to produce M- or X-class flares within a 24-hour window. The approach integrates a Convolutional Neural Network (CNN) autoencoder for feature extraction and a Long Short-Term Memory (LSTM) binary classifier for flare activity prediction, achieving a 90% test accuracy.

By leveraging advanced ML techniques, this methodology demonstrates the potential of data-driven models in heliophysics. Our results highlight the transformative role of AI-powered science in advancing solar flare prediction and contributing to the development of reliable early warning systems for space weather forecasting.

How to cite: Doria Rosales, E., Carbone, P. V., Falanga, P. M., Ciaramella, P. A., and Di Nardo, PhD. E.: Machine Learning for Space Weather: Solar Flare Forecasting Using SDO/HMI Magnetogram Time Series, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9116, https://doi.org/10.5194/egusphere-egu25-9116, 2025.

Space weather describes the dynamic conditions in near-Earth space, mostly driven by the variable interaction between the continuous flow of the solar wind and the Earth’s magnetic field.  Extreme space weather has the potential to disrupt or damage key infrastructure on which we rely, for example through the generation of large, anomalous Geomagnetically Induced Currents (GICs) in power networks and transformers.  Accurately forecasting a risk of large GICs would enable key actions to be taken to mitigate their impact.

Given the sparsity of direct GIC measurements, and their inherent specificity to the contemporaneous network properties and configuration, we turn to forecasting the driving factor: the changing ground magnetic field (R).  In this talk we discuss a recent model developed to forecast whether the rate of change of the ground magnetic field (R) will exceed specific, high thresholds in the United Kingdom.  The model uses a common space weather forecasting framework: an interval of data from the upstream solar wind is used to make a prediction as to future conditions at the Earth.  We will use this model as an example to discuss forecasting performance, particularly with respect to different magnetospheric driving and processes.  We demonstrate the use of techniques such as SHAP (Shapley Additive exPlanations) to investigate how and why the model is making the predictions that it does.  What physical processes can this model set up capture?  Where do we need to go in the future?

How to cite: Smith, A., Rae, J., Forsyth, C., Coxon, J., Walach, M.-T., Lao, C., Bloomfield, S., Reddy, S., Coughlan, M., Keesee, A., and Bentley, S.: Space Weather Forecasts of Ground Level Space Weather with Machine Learning: Performance, Limitations and Challenges, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9149, https://doi.org/10.5194/egusphere-egu25-9149, 2025.

The timely and precise prediction of coronal mass ejection (CME) arrival times and the characterization of near-Sun solar wind conditions are essential for space weather forecasting and planetary sciences. We develop a novel deep-learning framework that integrates imaging observations and physical parameters to predict CME arrival times with improved accuracy. Using time-series data from synchronized solar white-light and EUV observations of 156 geoeffective CME events (2000–2020), we train two models: Model A, a convolutional neural network (CNN) regression model, and Model B, an enhanced version incorporating 11 key physical parameters of CMEs and background solar wind. Model B achieves a minimum mean absolute error (MAE) of 5.12 hours, a 33% improvement over Model A. This demonstrates the value of combining observational and physical data in forecasting CME arrival times.

In addition, we explore the use of GONG/ADAPT magnetograms with a U-Net-based architecture to model solar wind conditions at 0.1 AU. The training labels are derived from the COCONUT coronal model, which offers a potential acceleration in generating initial driving conditions for heliophysical models like ICARUS. While preliminary, this approach highlights a pathway to streamline the modeling of near-Sun solar wind environments, further supporting interplanetary CME propagation studies.

Our results underscore the potential of machine learning when synergized with solar physics to advance predictions critical to heliophysics and planetary sciences.

How to cite: Li, Y., Yang, Y., Shen, F., Lin, R., Wang, H., and Poedts, S.: Integrating Machine Learning and Solar Physics for Enhanced Prediction of CME Arrival Times and Near-Sun Solar Wind Conditions, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9587, https://doi.org/10.5194/egusphere-egu25-9587, 2025.

Segmentation and characterization of solar coronal structures are essential for advancing our understanding of solar atmosphere and accurately identifying key regions, such as active regions and coronal holes, which are precursors to phenomena like solar flares and coronal mass ejections (CMEs). In this study, we investigate two complementary approaches to automate this process. First, we employ a previously presented deep learning-based U-Net architecture tailored for segmenting and characterizing solar coronal structures. Second, we develop a lightweight algorithm aimed at optimizing resource efficiency, consisting of classical computer vision techniques, which include thresholding and morphological filtering. The approach that best balances segmentation performance and computational efficiency will be selected for integration into a prototype designed to support future space exploration missions.

To characterize the segmented regions, we propose a set of carefully designed hand-crafted features to represent and characterize the resulting segmentations. These representations are analyzed using unsupervised clustering techniques, such as K-means and t-SNE, to distinguish solar coronal structures, including active regions, coronal holes and bright points.

Our dataset spans multiple layers of the solar atmosphere, incorporating HMI magnetograms (photosphere) and AIA wavelengths—94 Å (flaring regions), 171 Å (quiet Sun), 193 Å (coronal structures), and 304 Å (chromosphere). The performance of both segmentation approaches is thoroughly evaluated using metrics such as Dice score and Intersection over Union (IoU), with comparisons made against state-of-the-art methods.

Future work will focus on developing feature encoding techniques to better understand and predict solar phenomena, such as solar flare emissions, while investigating the impact of different feature extraction strategies on model performance.

References:

• Galvez, Richard, et al. "A machine-learning data set prepared from the NASA solar dynamics observatory mission." The Astrophysical Journal Supplement Series 242.1 (2019): 7.
• Šimon Mackovjak et al. “SCSS-Net: solar corona structures segmentation by deep learning”, Monthly Notices of the Royal Astronomical Society, Volume 508, Issue 3, December 2021, Pages 3111–3124, https://doi.org/10.1093/mnras/stab2536
• Gonidakis, Panagiotis & Sóñora-Mengana, Alexander & Jansen, Bart & Vandemeulebroucke, Jef. (2023). Handcrafted Features Can Boost Performance and Data-Efficiency for Deep Detection of Lung Nodules From CT Imaging. IEEE Access. PP. 1-1. 10.1109/ACCESS.2023.3331315. 

How to cite: Gonidakis, P., Carella, F., Miloshevich, G., and Poedts, S.: Efficient Segmentation and Clustering of Solar Coronal Structures: A Comparison of U-Net and Classical Computer Vision Techniques Using SDO Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9849, https://doi.org/10.5194/egusphere-egu25-9849, 2025.

Bursty bulk flows (BBFs) play a crucial role in transporting energy, mass, and magnetic flux in the Earth's magnetotail, particularly in the earthward direction. However, their impulsive nature and small spatial scale present significant challenges for in-situ observation, as only a limited number of spacecraft operate within the vast expanse of the magnetotail. Consequently, studying their statistical characteristics is a highly demanding task, and accurately predicting their behavior remains a distant goal. In this study, we analyze key characteristics of BBFs and apply regression-based models to predict their parameter behaviorUsing observational data from the THEMIS mission collected between 2007 and 2023, we conducted a feature analysis on parameters associated with BBFs evolution, including velocity, magnetic field, electric field, temperature, density, pressure, and specific entropy indices. Through statistical techniques, we identified parameters exhibiting predictable patterns during BBF events, distinguishing them from background conditions. Furthermore, we used XGBoost regression model, optimized for different parameter combinations, to forecast BBF duration, physical parameters’ average minimum, and peak intensity. This study also tested combinations of parameter predictions across instruments. When using observed background value in parameter combination, our models achieved Mean Absolute Percentage Errors of under 35% for critical variables, including Bz, Btotal, plasma pressure, and ion temperatures, and ion specific entropy and so on. Additionally, we observed BBF duration’s spatial distribution trends: it peaked at approximately X=-13Re, while decreasing with increasing Z distance from the plasma sheet, showing dawn-dusk asymmetry consistent with prior observations. This work highlights the potential of regression methods in forecasting BBFs characteristics and offers insights into their spatial behavior, supporting enhanced prediction capabilities in magnetospheric studies. Future research will aim to improve accuracy with enriched datasets.

How to cite: Feng, X., Yang, J., Bortnik, J., Wang, C.-P., and Liu, J.: Predicting characteristics of bursty bulk flows in Earth’s plasma sheet using machine learning techniques, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10383, https://doi.org/10.5194/egusphere-egu25-10383, 2025.

The rapid growth of high-dimensional data in solar physics presents significant challenges for analysis and interpretation, making it an excellent domain for the application of machine learning (ML) algorithms. Synoptic full-disk observations with the Solar Dynamics Observatory (SDO)  provide continuous observations of the solar magnetic activity over more than one solar cycle, facilitating the study of solar variability and space weather impacts. The Space-weather HMI Active Region Patches (SHARP) vector magnetic field (VMF) maps and parameters, based on Helioseismic and Magnetic Imager (HMI) full-disk observations, are developed to study the magnetic evolution of individual active regions and flare triggering mechanisms. We present a method for active region parametrization by combining empirical parameters and ML-extracted features. Time series of SHARP VMF maps are used as input for the Disentangled Variational Autoencoder (VAE), a Disentangled Representation Learning (DRL) algorithm that facilitates the extraction of a low-dimensional feature representation. The VAE model is used to encode generalized information about nonlinear dynamical systems, i.e., a solar active region, aiming to isolate distinct factors of variation in the data, allowing a clearer interpretation of physical processes. We demonstrate how the ML features can be used to identify and study the stages of the magnetic patches evolution. These are benchmarked with SHARP parameters, relating empirical and learned features. Furthermore, the empirical dataset enhanced with ML features can be used to analyze the development of individual active regions and searching for eruption precursors.

How to cite: Dineva, E., Miloshevich, G., Lapenta, G., Magdalenic Zhukov, J., and Poedts, S.: Parametrization of SHARP Vector Magnetic Field Using Disentangled Representation Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11565, https://doi.org/10.5194/egusphere-egu25-11565, 2025.

The solar wind, a stream of charged particles originating from the Sun, poses significant risks to technology and astronauts. It is driven by large structures on the solar surface like coronal holes and active regions, which can be identified in extreme ultra-violet (EUV) solar images several days before they become geoeffective. In this work, we propose to use a distributional regression algorithm to forecast the solar wind speed at the Lagrange 1 point from solar images. Instead of predicting a single value, this method models the entire conditional distribution as a function of input features. It allows computing the uncertainty of predictions and specifying the probability of the solar wind speed exceeding certain thresholds, which is especially useful for extreme event predictions like coronal mass ejections and high-speed solar wind streams. We employ a convolutional neural network to encode solar images from multiple wavelength channels into unstructured low-dimensional representations. Using a semi-structured distributional regression approach, we couple the deep learning encoder with structured physical input parameters, such as past solar wind properties and solar cycle information. Thereby, we incorporate physical knowledge into the model and enhance explainability. We predict the solar wind speed distributions with a one-hour cadence four days in advance. We train and evaluate our method using cross-validation on 15 years of data and compare it to current state-of-the-art models. We find that it provides an accurate forecast and especially models the heavy-tailed solar wind speed distribution well. We further show the advantages over standard regression approaches and how to use the predicted conditional quantiles to improve extreme event predictions, highlighting the potential for operational space weather forecasts.

How to cite: Collin, D., Shprits, Y., Hofmeister, S., Bianco, S., Klein, N., and Gallego, G.: Solar Wind Speed Forecasting From Solar Images Using Distributional Regression , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11587, https://doi.org/10.5194/egusphere-egu25-11587, 2025.

Artificial intelligence (AI), particularly machine learning (ML), is widely applied in fields such as medicine, autonomous driving, and manufacturing. Over time, ML has also seen increasing use in space and geosciences, where its algorithms hold the potential to enhance orbit prediction and orbit determination (OD) by utilizing measurement data. However, ML models like Artificial Neural Networks (ANNs) are limited to problems with abundant data and are often considered "black boxes", as their predictions lack interpretability in a scientifically meaningful way. To address these challenges, Raissi et al. 2018 introduced Physics Informed Neural Networks (PINNs), a specialized type of ANN. PINNs integrate the governing differential equations of a system into the learning process, imposing a physical constraint on the network's training and predictions. This approach allows effective training with small datasets, removing the reliance on large amounts of measurements. Additionally, PINNs can estimate unknown or poorly defined parameters within the differential equations, making them conceptually similar to classical OD algorithms like the Weighted Least Squares method. Building on this, Scorsoglio et al. 2023 successfully applied a variant of PINNs, called Physics Informed Extreme Learning Machines (PIELMs), for OD. In this study, a similar approach is employed for OD within the AI4POD (Artificial Intelligence for Precise Orbit Determination) software tool, focusing on resident space objects (RSOs) in low Earth orbit. Following this, we explore various methods, such as output perturbation, to determine the covariance matrix for the PINN-based OD approach. The covariance matrix provides an assessment of uncertainty in the predicted orbit and therefore being an essential tool in real space missions and collision avoidance. These methods are compared for their realism and effectiveness, both against each other and against the covariance matrix results from classical approaches. This study aims to evaluate whether the proposed methods can replicate and potentially improve upon traditional covariance estimation techniques.

How to cite: Dallinger, F., Aigner, B., Andert, T., Haser, B., Pätzold, M., and Hahn, M.: On Covariance Estimation in Physics Informed Neural Networks for Orbit Determination, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11680, https://doi.org/10.5194/egusphere-egu25-11680, 2025.

In recent years, the field of space situational awareness (SSA) has gained increasing attention, driven by the rapid rise in both active satellites and orbital debris. Therefore, being able to predict the orbit of a resident space object (RSO) as accurately as possible is more critical than ever in order to reduce collision risks and to preserve the orbital environment. However, incomplete knowledge of debris geometry, uncertain object characteristics, or simplified force models can cause prediction errors which exceed orders of several kilometers within just a few days, making it useless for reliable collision avoidance operations. Using modern Machine Learning (ML) algorithms can enhance prediction accuracy by addressing these challenges as recent studies have shown. In this context we present Artificial Intelligence for Precise Orbit Determination (AI4POD), a Python package that is designed to simplify the integration of ML-algorithms within the orbit prediction and determination process. AI4POD is structured as a comprehensive toolbox that includes a high-fidelity force model, various measurement functions, and classical orbit determination (OD) algorithms such as the batch least-squares estimation method. This integrated approach allows users to combine traditional orbit simulations with data-driven approaches to improve accuracy and to extend the predictability horizon. Based on this catalog, several approaches from artificial intelligence (AI) shall be tested in the future. Inspired by already proposed methodologies we are generating a training set of historical tracking data along with their corresponding orbit determinations using the AI4POD toolbox. Several machine learning algorithms will be explored to learn the nonlinear prediction errors, aiming to compensate for unmodeled or uncertain factors such as incomplete knowledge of satellite geometry or environmental conditions.

How to cite: Aigner, B., Dallinger, F., Andert, T., Haser, B., Pätzold, M., and Hahn, M.: AI-Enhanced Orbit Determination: The AI4POD Framework, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11790, https://doi.org/10.5194/egusphere-egu25-11790, 2025.

Imagers that observe emissions from the atmosphere are commonly used to study various ionospheric phenomena.  These phenomena include the auroral oval, equatorial plasma bubbles, and travelling ionospheric disturbances.  A difficulty in using imager observations is accurately and automatically retrieving the locations of interest from these images.  We present an automated method designed to identify the auroral luminosity boundaries from space-based imager data.   These boundaries are important for high-latitude studies that use statistical or machine learning approaches, as geographic and magnetic coordinate systems that do not account for changes in the polar cap or equatorward auroral oval boundaries will mix together data from regions experiencing different types of coupling with the magnetosphere.

The boundary identification method was originally developed for the Imager for Magnetopause-to-Aurora Global Exploration (IMAGE) observations, and has been further adapted for use in a wider variety of situations.  We will discuss the updated detection method and demonstrate the process on two different satellite data sets.  The updated detection method will be made publicly accessible through a new Python package, pyIntensityFeatures.

How to cite: Burrell, A., Chisham, G., Longden, N., and Zawdie, K.: Automated Identification of Auroral Luminosity Boundaries using pyIntensityFeatures, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12327, https://doi.org/10.5194/egusphere-egu25-12327, 2025.

How to cite: Stepanyuk, O., Pötzi, W., Kozarev, K., Dechev, M., and Miteva, R.: Hybrid AI Approaches for Solar Feature Recognition Using Ground-Based Instrument Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12654, https://doi.org/10.5194/egusphere-egu25-12654, 2025.

The identification and characterization of the coronal mass ejections (CMEs) and fast solar wind flows in the in situ data are important for understanding dynamics of these phenomena and consequently for space weather forecasting. In this study, we apply Self-Organizing Maps (SOMs) and clustering techniques to analyze in situ solar wind observations. SOMs (Kohonen, T, 1982) [1] an unsupervised learning technique, is employed to project high-dimensional interplanetary plasma parameters such as velocity, density, temperature, and magnetic field onto a lower-dimensional representation, preserving the topological structure of the data. Clustering algorithms, such as k-means, are then applied to the SOM output to distinguish between ICME events, fast and slow solar wind flows.
Our approach is validated using a few months long interval of the ACE and Wind in situ observations, with labeled CME intervals from Richardson and Cane [2] as a benchmark. This combination of SOMs and clustering provides a framework for automated identification of interplanetary plasma structures, important for space weather studies but also for operational services. 

[1] T. Kohonen, ‘Self-organized formation of topologically correct feature maps’, Biol. Cybern., vol. 43, no. 1, pp. 59–69, Jan. 1982, doi: 10.1007/BF00337288
[2] Richardson, Ian; Cane, Hilary, 2024, "Near-Earth Interplanetary Coronal Mass Ejections Since January 1996"https://doi.org/10.7910/DVN/C2MHTH

How to cite: Carella, F., Magdalenić, J., and Bemporad, A.: Identification of fast solar wind flows and CMEs in the in situ data using Self-Organizing Maps and clustering techniques, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13602, https://doi.org/10.5194/egusphere-egu25-13602, 2025.

The ionosphere has important impacts on many different systems, such as communications, thus modeling it is a crucial task. The influence of the ionosphere is closely linked to its electron density, but this is difficult to measure adequately. Because of this, modeling requires the use of additional correlated values, such as solar activity metrics. These measures do not capture enough to reproduce small-scale changes in electron density, so we have developed a technique to expand our input space to include sparse measurements of Total Electron Content (TEC), or the integral of electron density.

TEC data is measured more densely than electron density, although it is still not consistent spatially, with many gaps in measurement coverage. Despite this, it is collected very consistently throughout time so it presents itself as a good candidate for an input to an ionospheric model. Even so, TEC has not been used as an input to such models, especially Machine Learning (ML) models, as the irregular coverage of the measurements makes it difficult to deal with.

We have developed a technique to use transformer-like architectures to move from an irregular domain to a fixed size embedded domain to facilitate further usage of the TEC data. This approach has enabled us to use TEC as a direct input to electron density models, noticeably improving performance. Our technique also enables the use of a variety of irregular inputs all at once, enabling a wider range of possible model inputs. Lastly, as a byproduct of the process, we can use the inverse of our embedding technique (which is also how we train the model) to perform TEC map completion, where we can predict TEC values even where no measurements have been taken.

How to cite: Smith, L. and Cohen, M.: Using Transformers to Integrate Irregular Data for Improved Ionospheric Modeling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14036, https://doi.org/10.5194/egusphere-egu25-14036, 2025.

Numerical magnetohydrodynamic models (MHD) are often used to simulate the global interaction between the solar wind and the magnetosphere system. Increasingly, such MHD models require very computationally expensive, high numerical resolutions for realistic global magnetosphere simulations of multiscale turbulent plasma flows. To address this problem, we investigate and compare several ML-based approaches for subgrid-scale (SGS) parameterizations in the coarse-scale Grid Agnostic MHD for Extended Research Applications (GAMERA) model, starting with the Large-Eddy Simulation (LES) formalism. We use a 2D simulation of MHD turbulence in the Orszag-Tang vortex as a testbed to diagnose from benchmark high-resolution GAMERA  solutions the distributions of subgrid-scale (SGS) and large-scale (LS) fields, and model subgrid-scale (SGS) forcing that encapsulates induced feedbacks on the LS fields. 

How to cite: Kondrashov, D. and Sciola, A.: Subgrid-scale modeling of MHD turbulence, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14724, https://doi.org/10.5194/egusphere-egu25-14724, 2025.

We present a novel physics-informed machine learning (ML) model designed to forecast the background (ambient) solar wind up to five days in advance. Solar wind speed is a critical driver of geomagnetic activity, and inaccuracies in its prediction significantly contribute to large errors in forecasting the arrival times of coronal mass ejections (CMEs), which are typically off by at least 10 hours.

Predicting solar wind speed has historically been a challenging task, with even state-of-the-art models often failing to consistently outperform a simple 27-day persistence model. Operational physics-based (3D MHD) models, in particular, struggle to accurately forecast high-speed streams associated with co-rotating interaction regions. These regions arise from fast solar wind generated by coronal holes, which are not clearly captured in the magnetogram maps routinely used as inputs. While recent empirical and data-driven methods have shown relatively better performance, significant challenges remain.

Our approach integrates lessons from prior models into what we believe represents the current state-of-the-art. Specifically, we use GONG synoptic maps (magnetograms) and full-disk SDO EUV images as inputs to a neural network. This network estimates the optimal inner boundary condition for the radial solar wind velocity profile at 10 solar radii, which is then propagated to 1 AU using a simplified 1D hydrostatic model.

The key innovation lies in seamlessly integrating the physics-based model within the neural network, creating a true physics-informed ML framework.

We will present validation metrics to assess the model’s performance and discuss plans to make the forecast outputs available to the community 24/7 via the swx-trec.com portal.

How to cite: Camporeale, E. and Hu, A.: Ambient Solar Wind Speed Forecast with Physics-Informed Machine Learning , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14910, https://doi.org/10.5194/egusphere-egu25-14910, 2025.

Although the search for exoplanets currently incorporates various computational methods, it still heavily relies on manual analysis of light curves, a process that is both time-intensive and demanding. Our research in the EXOWORLD project addresses these challenges by integrating advanced machine learning techniques, including convolutional, into the transit search process, combining them with recurrent networks to create a fully integrated machine learning-based transit detection and characterization pipeline. This approach reimagines transit search as a pattern recognition task, employing self-learning algorithms to efficiently process vast amounts of astronomical data. We aim to explore and apply a range of machine learning methods, establishing a foundation for comparison not only among these methods but also against traditional transit search techniques. This comparison is expected to focus on potential improvements in efficiency, accuracy, and computational demands. Although still in the early stages, our research aims to significantly enhance exoplanet detection methods, streamlining the process and building a framework for making new discoveries through light curve analysis.

In this context, we present TRANSCENDENCE, our machine learning-based pipeline, which has demonstrated the ability to identify exoplanets larger than 2 Earth radii consitently. Moreover, the pipeline is capable of detecting smaller planets, albeit with lower detection probabilities. One of TRANSCENDENCE's key strengths lies in its remarkably low false positive rate, which ranges between 5% and 10% of all identified transits. By significantly reducing the need for manual intervention and minimizing false positives, this pipeline has the potential to strongly immprove the efficiency of exoplanet detection and characterization.

How to cite: Schmerling, H., Hribar, R., Grziwa, S., and Pätzold, M.: TRANSCENDENCE - A TRANSit Capture ENgine for DEtection and Neural network Characterization of Exoplanets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15985, https://doi.org/10.5194/egusphere-egu25-15985, 2025.

The Automatics in SpAce exploration (ASAP) project has as a goal the design and development of Machine Learning algorithms for the automation of operations to be implemented on the on-board processors of space missions. In the framework of ASAP a set of ML algorithms for on-board science operations of space missions have been developed/optimized on consumer-grade computing systems to be further selected for orting of existent ML models directly on an FPGA prototype. In more detail, algorithms pertaining to four main use cases have been considered: the autonomous triggering of special measurement modes and the selective downlink of plasma environment parameters; the advanced on-board data analysis of three-dimensional particle distribution functions; the on-board analysis of solar images; the on-board prediction capability of SEP related hazards. Here we describe the algorithms, their performances and requirements for the on-board implementation. ASAP has received funding from the EU’s HORIZON Research and Innovation Action (GA no.101082633)

How to cite: Torda, T., Alberti, T., Consolini, G., De Marco, R., Dineva, E., Ekelund, J., Gonidakis, P., Laurenza, M., Marcucci, M. F., Markidis, S., Miloshevich, G., Poedts, S., Sanò, B., and Chrysaphi, N.: Machine Learning Algorithms for Autonomous Space Mission Operations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16713, https://doi.org/10.5194/egusphere-egu25-16713, 2025.

Remote sensing observations, whether astronomical or within the solar system, are constrained by instrumental limitations, such as the point spread function in imaging. Ensuring the reliability of scientific analysis from such data requires robust deconvolution techniques. We present a spatio-temporal deconvolution method, to minimise the effect of an extended or complex-shaped point spread function, applicable to dynamic systems with various timescales. This approach enhances observational data by improving image contrast and resolving small-scale dynamic features.

Our method employs a deep neural network trained on state-of-the-art numerical simulations, enabling it to identify dynamic patterns in both spatial and temporal dimensions and to estimate and correct the degradation of intensity contrast. The resulting improvements in intensity representation and resolution facilitate more accurate analyses of small-scale features.

We apply this methodology to solar observations in the millimeter wavelength regime, recovering fine-scale structures critical for understanding the complex behaviour of the solar atmosphere, predict the generation of potentially harmful events, solar flares and the solar wind. By incorporating the temporal domain, our approach surpasses traditional 2D deconvolution techniques.

While initially developed for solar imaging, the method is versatile and can be adapted to various observational contexts across different wavelength regimes. This makes it a valuable tool for advancing future observational studies and expanding research capabilities.

How to cite: Eklund, H.: Spatio-temporal deconvolution method for enhanced image analysis, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17531, https://doi.org/10.5194/egusphere-egu25-17531, 2025.

Solar activity significantly influences the near-Earth environment, leading to magnetic storms and magnetospheric substorms that can impact both technological and human systems. Understanding the physical processes that govern the Sun-Earth relationship and developing models to forecast magnetic disturbances on Earth are therefore of critical importance. In this context, we present a preliminary work to model and forecast the dynamics of magnetic storms, as measured by the SYM-H geomagnetic index, using Physics-Informed Neural Networks (PINNs). This approach is applied to models based on deterministic ordinary differential equations (ODEs), such as those described by Burton et al. (1975) and others, which were proposed to describe the evolution of geomagnetic indices during magnetic storms. The findings and significance of this approach are discussed in the context of Earth's magnetospheric dynamics and the relevance of PINN techniques in space weather research.

How to cite: Lacal, M., Camporeale, E., Consolini, G., and Piersanti, M.: Modeling Magnetic Storms' Dynamics with Physics-Informed Neural Networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18050, https://doi.org/10.5194/egusphere-egu25-18050, 2025.

In this study, we develop an artificial intelligence (AI)-based solar surface flux transport (SFT) model. We predict global magnetic field distributions on the solar surface up to the next solar rotation (27.3 days) using deep-learning. Here we train and evaluate our deep-learning model, based on the Pix2PixCC architecture, using data sets of SDO/HMI, SOHO/MDI, and NSO/GONG synoptic maps with a resolution of 360 by 180 (longitude and sine-latitude) from 1996 to 2023. We present results of our model and compare them with those from the persistence model and the conventional SFT model, including the effects of differential rotation, meridional flow, and diffusion on the solar surface. Our AI-based SFT model generates magnetic field distributions for the next solar rotation, better than the conventional SFT model and the persistence model in the quantitative metrics such as RMSE, FSIM, and pixel-to-pixel CC. Our model successfully generates magnetic features, such as the diffusion of solar active regions and the motions of supergranules. Our model also generates small-scale magnetic features better than the conventional SFT models. Using synthetic input data with bipolar structures, we confirm that our model successfully reproduces differential rotation and meridional flow. Finally, we discuss the advantages and limitations of our model in view of magnetic field evolution and its potential applications.

How to cite: Jeong, H.-J., Jeon, M., Kim, D., Kim, Y., Baek, J.-H., Moon, Y.-J., and choi, S.: Prediction of Solar Surface Magnetic Fields Using an AI-based Surface Flux Transport Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18475, https://doi.org/10.5194/egusphere-egu25-18475, 2025.

Magnetic reconnection on the dayside magnetopause is considered the primary mechanism for transporting mass, momentum, and energy from the solar wind into the terrestrial magnetosphere. Several studies have demonstrated that the spatiotemporal dynamics of the dayside magnetic reconnection can be inferred remotely from the analysis of the time-energy dispersion of ions in the Earth’s cusps. Despite the immense number of in-situ cusp measurements acquired by numerous space-based instruments, it is still challenging to determine the overall cusp behavior owing to the intermittency of the measurement acquisition. To overcome this issue, this work implements a regression model of the three-dimensional (3-D) ion flux in the Earth’s Northern cusp based on deep learning techniques and numerous measurements of the cusp under varying solar wind conditions. For the training process, we have used solar wind parameters obtained from NASA's OMNI database as input and in-situ ion flux measurements acquired by the CIS/HIA instruments on board ESA’s multi-spacecraft Cluster mission during the period from 2001 to 2010 for supervised output. The model allows the reconstruction of the time-dependent, 3-D ion flux distribution within the cusp region, which serves to determine the boundaries of the high-altitude cusp, analyze its structural response to time-dependent solar wind conditions, and investigate the relationship between the cusp and dayside magnetic reconnection. The experiments under controlled input parameters show that our model is capable of reproducing  expected ion dispersion signatures as a response to variable solar wind conditions.

How to cite: Cucho-Padin, G., Sibeck, D., Da Silva, D., and Wang, X.: A Deep-learning-based Model of the Three-dimensional Ion Flux in the Earth’s Northern Cusp, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20652, https://doi.org/10.5194/egusphere-egu25-20652, 2025.

How to cite: Davoli, G., Mastrangelo, D., Cherchi, A., and Alessandri, A.: A new orographic drag parameterization package for the GLOBO model: implementation and evaluation , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-872, https://doi.org/10.5194/egusphere-egu25-872, 2025.

 Drought persistence is a critical factor in assessing water availability and its impacts on agriculture, ecosystems, and society. In this respect, poorly constrained soil properties in climate models such as field capacity – i.e. the maximum water a soil can retain after drainage of excess moisture – may strongly affect severity and persistence of simulated soil drought conditions. This study examines for the first time the regulating role of soil properties, particularly of field capacity, in shaping drought memory and its broader impacts. Using the CMIP6 multi-model ensemble and observations, we analyze drought dynamics across various phases of the hydrological cycle applying non-parametric standardized indices: Standardized Precipitation Index (precipitation deficits), Standardized Precipitation-Evapotranspiration Index (precipitation-evapotranspiration balance), Standardized Soil Moisture Index (soil moisture deficits), and Standardized Runoff Index (reduced runoff). Our analysis investigates the persistence between hydrological drought indicators, showing that soils with greater field capacity sustain drought conditions longer, emphasizing the importance of accurately modeling soil properties to capture drought persistence effectively. The historical CMIP6 simulations are compared with observational datasets, including GLEAM and CRU, to assess the deviation between model outputs and observed climate conditions. The future scenarios (SSP126, SSP245, SSP370, SSP585) are also examined, revealing significant regional differences in projected drought behavior depending on the degree of radiative-forcing increase during 21st century. High-emission scenarios project prolonged drought conditions due to increased temperatures and evapotranspiration feedback, while low-emission pathways are effective in preserving more stable hydrological dynamics. Our results show that, in water limited and transition areas such as the Euro-Mediterranean region, the persistence of droughts and its projected change considerably depend on the modeled field capacity. This study highlights the essential role of field capacity and other soil characteristics in regulating the variability and the persistence of drought events. By bridging historical validation with future projections, it provides a comprehensive understanding of drought dynamics and trends, also identifying observational constraints for the Earth System Models. These findings are crucial for refining predictions of agricultural and hydrological drought impacts and for guiding adaptation strategies in water-limited regions that are vulnerable to drought exacerbation under climate change.

How to cite: Possega, M., Di Carlo, E., Senigalliesi, V., and Alessandri, A.: Understanding Soil Modulation of Drought Persistence in CMIP6 Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1040, https://doi.org/10.5194/egusphere-egu25-1040, 2025.

The Arctic region has been warming at a rate significantly faster than the global average, leading to an accelerated decline in sea ice. This trend is expected to continue, potentially resulting in a "low-ice regime," which could make sea ice conditions more unpredictable. Anticipating changes in Arctic sea ice and climate states is therefore crucial for guiding various human activities, from natural resource management to risk assessment decisions. While global climate and Earth system models project continuous sea ice decline over decadal time scales, achieving reliable seasonal forecasts remains challenging. To address this, we apply dynamical downscaling with the state-of-the-art Regional Arctic System Model (RASM), which enables us to forecast Arctic sea ice on time scales ranging from weeks to six months. RASM is a fully coupled regional climate model that integrates components for the atmosphere, ocean, sea ice, and land, interconnected through the flux coupler of the Community Earth System Model. In our study, we simulate RASM at a horizontal resolution of 1/12 degree (approximately 9 km) for both the ocean and sea ice, with 45 vertical levels in the ocean and five thickness categories for sea ice. The atmosphere is configured on a 50-km grid with 40 vertical levels, dynamically downscaled from the NOAA/NCEP Climate Forecasting System version 2 (CFSv2) at 72-hour intervals for the upper half of the atmosphere. Monthly ensemble forecasts extending up to six months are generated using initial conditions derived from a fully-coupled RASM hindcast simulation without bias correction and assimilation. This presentation highlights results for September sea ice predictions initialized on April 1, May 1, June 1, July 1, August 1, and September 1, covering pan-Arctic and regional sea ice spatio-temporal conditions from 2012 to 2021. Specifically, we examine how lead time and initial conditions affect the quantitative skill of seasonal predictability for Arctic sea ice and demonstrate skillful predictions of September sea ice up to six months in advance. Overall, our study underscores that enhancing model physics and obtaining more realistic initial conditions are crucial for achieving skillful sub-seasonal to seasonal predictions.

How to cite: Lee, Y., Maslowski, W., Craig, A., Clement Kinney, J., and Osinski, R.: Sub-seasonal to Seasonal Arctic Summer Sea Ice Forecasts Using Dynamical Downscaling with the Regional Arctic System Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1545, https://doi.org/10.5194/egusphere-egu25-1545, 2025.

AI deep learning for climate science has attracted increasing attentions in recent years with rapidly expanded applications to many areas. In this talk, I will briefly present our recent progresses on using various deep learning methods for seasonal-to-multi-seasonal predictions of ENSO, the Indian Ocean Dipole (IOD), summer precipitation in China and East Africa, Arctic sea ice cover, ocean waves, as well as the bias correction and downscaling of dynamical model’s forecasts. The results suggest that many popular deep learning methods, such as convolutional neural networks, residual neural network, long-short term memory, ConvLSTM, multi-task learning, cycle-consistent generative adversarial networks and vision transformer, can be well applied to improve our understanding and predictions of climate. In addition, a brief introduction of AI large models for ensemble weather-subseasonal-seasonal-decadal forecasts, together with the perspective on the future development of AI methods, will also be presented.

How to cite: Luo, J.-J.: AI deep learning for climate forecasts, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1847, https://doi.org/10.5194/egusphere-egu25-1847, 2025.

Tropical Instability Waves (TIWs) play a crucial role in modulating Sea Surface Temperature (SST) variability in tropical oceans, yet their representation in current forecast systems remains challenging. This study investigates the relationship between TIWs and sub-seasonal SST predictability while evaluating the performance limitations of the Licoms Forecast System. Through comprehensive analysis of observational data and model outputs, we demonstrate that TIWs provide significant potential for enhancing sub-seasonal SST forecast skill through their regular wave patterns and predictable evolution characteristics. However, our findings reveal that the current Licoms forecast systems systematically underestimate both TIW intensity and wavelength. Critical examination of error sources indicates that these deficiencies primarily originate from initialization fields rather than model physics or dynamics. 

How to cite: Tianyan, L., Yongqiang, Y., and Weipeng, Z.: The role of Pacific Tropical Instability Wave in Sub-Seasonal SST predictability , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2210, https://doi.org/10.5194/egusphere-egu25-2210, 2025.

The El Niño/Southern Oscillation (ENSO) is the primary internal climatic driver shaping extreme events worldwide^1,2,3. Its intensity and frequency in response to greenhouse gas (GHG) warming has puzzled scientists for years, despite consensus among models about changes in average conditions^4-16. Recent research has shed light on changes not only in ENSO variability^5,7,8,10,13, but also in the occurrence of extreme^{5,6,11,12,13,14} and multi-year El Niño^4,15, and La Niña^9,11,16 events under GHG warming. Here, we investigate potential changes in ENSO predictability associated with changes in ENSO dynamics in the future by using long-range deep-learning forecasts trained on extensive large ensemble simulations of Earth System Models under historical forcings and the future high GHG emissions scenario. Our results show a remarkable increase in the predictability of ENSO events, ranging from 35% to 65% under the high GHG emissions scenario due to reduced ENSO irregularity, supported by a broad consensus among multi-models. Under GHG warming, an El Nino-like warming flattens the thermocline depth with upper ocean stratification. This flattening of the thermocline depth leads to an increased transition frequency between El Niño and La Niña events, driven by strengthened recharge-discharge oscillation with enhanced thermocline feedback and SST responses to zonal wind stress. As a result, ENSO complexity would reduce with increased regularity and reduced skewness, increasing ENSO predictability. These results imply that the future social and economic impacts of ENSO events may be more manageable, despite an expected increase in the frequency of extreme ENSO events.

How to cite: Yang, Y.-M., Park, J.-H., Lee, J.-Y., An, S.-I., Yeh, S.-W., Kug, J.-S., and Ham, Y.-G.: Increased multi-year ENSO predictability under greenhouse gas warming accounted by large ensemble simulations and deep learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3974, https://doi.org/10.5194/egusphere-egu25-3974, 2025.

Climate services are essential to support climate-sensitive decision making, enabling adaptation to climate change and variability, and mitigate the sources of anthropogenic climate change, while considering the values and contexts of those involved. The unregulated nature of climate services can lead to low market performance and lack of quality assurance. Best practices, guidance, and standards serve as a form of governance, ensuring quality, legitimacy, and relevance of climate services. The Climateurope2 project (www.climateurope2.eu) addresses this gap by engaging and supporting an equitable and diverse community of climate services to provide recommendations for their standardisation. Four components of climate services are identified (the decision context, the ecosystem of actors and co-production processes, the multiple knowledge systems involved, and the delivery and evaluation of these services) to facilitate analysis. This has resulted in the identification of nine key messages summarising the susceptibility for the climate services standardisation. The recommendations are shared with relevant standardisation bodies and actors as well as with climate services stakeholders and providers.

How to cite: Doblas-Reyes, F., Lera St Clair, A., Baldissera Pacchetti, M., Checchia, P., Cortekar, J., Klostermann, J. E. M., Krauß, W., Muñoz, A., Mysiak, J., Paz, J., Terrado, M., Villwock, A., Volarev, M., and Zorita, S.: Standardisation of equitable climate services by supporting a community of practice, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5233, https://doi.org/10.5194/egusphere-egu25-5233, 2025.

Flooding is the most frequent and damaging hydrometeorological disaster in Indonesia, with Java being especially vulnerable due to its dense population and rapid urbanization. This study aims to refine the Impact-Based Forecasting (IBF) model to improve flood hazard predictions and mitigation efforts. Using Global Precipitation Measurement (GPM-IMERG) rainfall data as the hazard component combined with vulnerability and capacity datasets from InaRISK, this research focuses on enhancing the precision and reliability of impact assessments.

Initial analyses highlight the potential of impact-based rainfall thresholds and assessment probabilistic impacts to refine the IBF model and reduce subjectivity in impact assessments. By linking calculated impact values and disaster magnitudes for the 2014 – 2023 period, this study shows a promising skill for significant and severe flood events, although improvements are needed for minor and minimal disaster classifications.

This research lays the groundwork for a more robust and scalable IBF model tailored to Java’s unique challenges. The findings aim to support BMKG’s operational needs, enabling the delivery of more actionable early warnings and targeted disaster preparedness measures. By addressing critical gaps in existing IBF systems, this study contributes to bridging the divide between hazard-impact forecasts and societal resilience, ultimately mitigating the impacts of floods in Indonesia.

How to cite: Purnama, D. R., Tett, S. F. B., Doherty, R., and Pramuwardani, I.: Impact-Based Forecasting Model for Flood Hazard Mitigation in Java, Indonesia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7271, https://doi.org/10.5194/egusphere-egu25-7271, 2025.

The concept of hydroclimate services is predominantly recognised as web portals dedicated to the dissemination of data to potential users. However, the scope of climate services extends beyond the sole provision of data. This communication presents a comprehensive ecosystem of tools and resources associated with the development of an updated national hydrological projection dataset in France. The ecosystem was brought to life through a close collaboration between scientists and water managers in two joint projects: Explore2 and LIFE Eau&Climat. Tools and resources were thus developped with and for water resource managers, and designed to enhance the comprehension of both the conceptual framework and the data itself, facilitating utilisation in accordance with best practices for climate change adaptation.

The project websites serve as gateways to the ecosystem and the tools: the Explore2 website contains interviews with the scientific contributors, and the LIFE Eau&Climat website is hosted by the national website dedicated to water managers. A summary of the joint final public event accompanies the replay of the one-day conference and debates on a dedicated website. A compendium of antecedent research projects on climate change impacts on hydrology has been collated to summarise the state of the art prior to the two projects. A MOOC has been developed in conjunction with scientists to facilitate the comprehension of the Explore2 project, its design, and its application in adaptation studies.

Moreover, the Explore2 dataverse (https://entrepot.recherche.data.gouv.fr/dataverse/explore2) brings together a variety of products in an organised and searchable way, including thematic scientific reports, GIS layers, and other key metadata. It also contains three types of station datasheets aimed at locally contextualising outputs: hydrological model performance datasheets, projection results datasheets, and uncertainty quantification datasheets. The MEANDRE interactive data visualisation tool (https://meandre.explore2.inrae.fr/) offers a guided tour of the salient take-home messages and a comprehensive exploration of the Explore2 hydrological projection dataset. This multi-model dataset (GCMs/RCMs/bias correction methods/hydrological models) is made available through the DRIAS-Eau portal (https://drias-eau.fr/), which functions as a water mirror of the established DRIAS-Climat portal. The utilisation of this dataset for local climate change impact studies is facilitated by a methodological guide written as an adventure gamebook (https://livreec.inrae.fr/) and based on real-life studies carried out by water managers during the LIFE Eau&Climat project. Furthermore, experiments of sonification of hydrological projections offer a novel approach to apprehending future changes (https://explore2enmusique.github.io/).

This ecosystem has been met with great anticipation and acclaim by local to national-scale water managers, paving the way for ongoing local prospective studies. These will be able to confront future resources with the ecological needs of aquatic environments and human water usage.

This work is funded by the EU LIFE Eau&Climat project (LIFE19 GIC/FR/001259).

How to cite: Vidal, J.-P., Sauquet, E., Héraut, L., Siauve, S., Evin, G., Soubeyroux, J.-M., Tocquer, F., Bornançin-Plantier, A., Magand, C., and Berel, M.: Hydroclimate services are more than just providing data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9933, https://doi.org/10.5194/egusphere-egu25-9933, 2025.

Drought warnings are vital to sectors like agriculture and water management, especially at the seasonal time scale. Identifying the sources of drought predictability in regions where a prediction system demonstrated potential for useful applications of the forecasts, represents an important step toward building confidence in the predictions and refining the seasonal predictions. To better identify higher forecast skill in this context, one possible approach is to focus on specific “windows of opportunity”. The approach aims to identify periods when persistent anomalies occurring in the ocean, the atmosphere or the land surface may positively precondition the predictive ability of the seasonal forecast. In the case of SPI3, a high potential for preconditioned predictive skill is identified in the Middle East region, as suggested by a robust relationship with large-scale climate modes. Building on these results, this study explores the contributions of individual years to the skill for the region during the autumn season and in the hindcast period 1993-2016. We used a Multi Model Ensemble (MME) of eight seasonal prediction systems (SPSs) provided by the Copernicus Climate Data Store (CDS) and observations from the Climate Research Unit (CRU) to calculate the SPI3 time series and the values of the Pacific and Indian teleconnection indices, the Oceanic Nino Index (ONI) and the Dipole Mode Index (DMI), respectively. A novel methodology is implemented to cluster the year-by-year MME contributions to the Pearson correlation coefficient (PCC) that are preconditioned by the large-scale teleconnections. 

Results indicate that years with extreme high or low values of ONI and DMI are the main contributors to the forecasting skill of the MME drought predictions over the Middle East. In particular, a window of opportunity is identified in four (out of 24) years that show significantly high contribution to overall skill. These years are robustly preconditioned by El Niño or La Niña events. Among the years with higher contributions, 1994 stands out as being more influenced by the DMI, thus driven primarily by SST anomalies in the Indian Ocean rather than the Pacific Ocean.  The methodological approach developed in this study successfully highlighted the potential windows of opportunity for seasonal prediction in the Middle East region, and could be applied extensively to develop early warnings for the coming seasons to serve agriculture and water management operations.

How to cite: Dal Monte, T., Alessandri, A., Cherchi, A., Donat, M., and Gaetani, M.: Windows of Opportunity for Seasonal Prediction of droughts: the case of the Middle East, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11600, https://doi.org/10.5194/egusphere-egu25-11600, 2025.

Seamless climate predictions combine information across different timescales to deliver information potentially useful for sectors like agriculture, energy, and public health. Seamless operational forecasts for periods spanning from sub-annual to multi-annual timescales are currently not available throughout the year. We show that this gap can be closed by using a well-established climate model analog method. The method consists in sampling model states from the CMIP6 transient simulation catalog based on their similarity with the observed sea surface temperature as a means of model initialization. 

Here we present the methodology and basic skill evaluation of the analog-based temperature and standardized precipitation index retrospective predictions with forecast times ranging from 3 months up to 4 years. We additionally compare these predictions with the non-initialized CMIP6 ensemble and with two operational benchmarks produced with state-of-the-art dynamical forecasts systems: one on seasonal timescales and the other on annual to multi-annual timescales.

The analog method provides skillful climate predictions across the different timescales, from seasons to several years, offering temperature and precipitation forecasts comparable to those from state-of-the-art initialized climate prediction systems, particularly at the annual to multi-annual timescales. However, unlike operational decadal prediction systems that provide only one or two initializations per year, the analog-based system can generate seamless predictions with monthly initializations, offering year-round climate information. Additionally, analog predictions are computationally inexpensive once the multi-model transient climate simulations have been completed. We argue that these predictions are a valuable complement to existing operational prediction systems and may improve regional climate adaptation and mitigation strategies. 

How to cite: Acosta Navarro, J. C., Aranyossy, A., De Luca, P., Donat, M. G., Hrast Essenfelder, A., Mahmood, R., Toreti, A., and Volpi, D.: Seamless seasonal to multi-annual climate predictions by constraining transient (CMIP6) climate model simulations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11821, https://doi.org/10.5194/egusphere-egu25-11821, 2025.

Vegetation plays a crucial role in the land surface water and energy balance modulating the interactions and feedback with climate at the regional to global scale. The availability of unprecedented Earth observation products covering recent decades (and extended up to real-time) are therefore of paramount importance to better represent the vegetation and its time evolution in the land surface models (LSMs) used for offline analysis/initialization and for the seasonal-to-decadal predictions. 

Here, we integrate realistic vegetation Leaf Area Index (LAI) variability from latest generation satellite campaigns, available through Copernicus Land Monitoring Service (CLMS), in three different LSMs that conducted the same coordinated set of offline land-only simulations forced by hourly atmospheric fields derived from the ERA5 atmospheric reanalysis. The experiment implementing realistic interannually-varying LAI (SENS) is compared with simulations utilizing a climatological LAI (CTRL) to quantify the vegetation feedback and the effects on the simulation of near-surface soil moisture.

The results show that the inter-annually varying LAI considerably affects the simulation of near-surface soil moisture anomalies in all three models and over the same water-limited regions, but surprisingly the effects diverge among models: compared with ESA-CCI observations, the near-surface soil moisture anomalies significantly improve in  one of the three LSMs (HTESSEL-LPJGuess) while the other two (ECLand and ISBA-CTRIP) display opposite effects with significant worsening of the anomaly correlation coefficients. It is found that the enhanced simulation of near-surface soil moisture is enabled by the positive feedback that is activated by the effective vegetation cover (EVC) parameterization, implemented only in HTESSEL-LPJGuess. The EVC parameterization works such that the effective fraction of the bare soil being covered by vegetation does vary with LAI following an exponential function constrained by available satellite observations. The increased (reduced) soil-moisture limitation during dry (wet) periods produces negative (positive) LAI and therefore EVC anomalies, which in turn generate a dominating positive feedback on the near-surface soil moisture of HTESSEL-LPJGuess by exposing more (less) bare soil to direct evaporation from the sub-surface layer. On the other hand, in the EC-Land and ISBA-CTRIP models, EVC is fixed in time as it cannot vary with LAI and so the positive feedback described cannot be activated. The only feedback on near-surface soil moisture anomalies that operates  in these two models is negative and comes from the reduced (increased) transpiration related to the negative (positive) LAI anomalies.

Simply prescribing observed vegetation data into LSMs does not guarantee the introduction of the correct coupling and feedback on climate. In this respect, this multi-model comparison experiment demonstrates the fundamental role of the inclusion of the underlying vegetation processes in LSMs. Ignoring the proper representation of the vegetation processes could lead to unrealistic (and even the opposite effects compared with observations) behaviour in reanalysis and climate predictions.

How to cite: Alessandri, A., Possega, M., Di Carlo, E., Cherchi, A., Boussetta, S., Balsamo, G., Ardilouze, C., Dayon, G., and van Oorschot, F.: Representation of Temporal Variations of Vegetation in Reanalysis and Climate Predictions: Diverging Soil-Moisture Response in Land Surface Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13385, https://doi.org/10.5194/egusphere-egu25-13385, 2025.

Despite growing awareness of climate change, many Indonesians remain climate-silent, posing a significant challenge to the country's efforts to mitigate its impacts. This study aims to analyze the factors contributing to climate silence in Indonesia, using psychological theories related to climate science denial. A rapid systematic review was conducted to gather evidence, revealing five key drivers of climate denial: limited cognitive abilities, ideological beliefs, sunk costs, perceived risks, and discredence. These barriers are further shaped by factors such as government policies, economic conditions, religious influences, and insufficient environmental education.
This skepticism towards climate change undermines adaptation and mitigation efforts by disrupting community engagement and participation. The findings highlight the importance of government support in addressing the root causes of climate skepticism. Employing the concept of inoculation through a misconception-based learning approach—integrated into religion and education—can help reshape mindsets. Enhancing public understanding of climate change is essential to fostering community involvement and support for effective climate mitigation initiatives.

Keywords: climate silence, climate denial, psychological drivers, Indonesia.

How to cite: Dewita, A. and Inayah, B.: Psychological Drivers of Climate Silence: A Challenge to Indonesia's Climate Action, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14900, https://doi.org/10.5194/egusphere-egu25-14900, 2025.

In recent years, alternating drought and extreme precipitation events have highlighted the need for subseasonal to seasonal forecasts of the terrestrial water cycle. In particular, predictions of the impacts of dry and wet extremes on subsurface water resources are crucial to provide stakeholders in agriculture, forestry, the water sector, and other fields with information supporting the sustainable use of these resources.

In this context, we release an experimental Water Resources Bulletin (https://adapter-projekt.de/bulletin/index.html) four times per year, offering probabilistic forecasts of the total subsurface water storage (TSS) anomaly at a 0.6 km resolution, from the surface down to 60 m depth, for the upcoming seven months across Germany. These seasonal forecasts are generated using the integrated, physics-based hydrological model ParFlow/CLM, forced by 50 ensemble members of the SEAS5 seasonal forecast from the European Centre for Medium-Range Weather Forecasts (ECMWF).

To evaluate our forecasts, we evaluated six 7-months probabilistic forecasts covering the vegetation period (March to September) for the years 2018 to 2023 with a reference long-term historical time series based on the same ParFlow/CLM setup. The forecast skill was assessed by comparing these seasonal forecasts to a climatology-based 10-member pseudo-forecast over the 2013–2023 period (using the leave-one-out method), extracted from the reference time series.

The monthly Continuous Ranked Probability Skill Score (CRPSS), which evaluates the ensemble distribution based on daily TSS data, indicates that the probabilistic forecast outperforms the climatology-based pseudo-forecast in most regions, except in 2018 and, to a lesser extent, in 2020 and 2022. This can be attributed to an under-representation of extremely dry members in the ensemble, combined with the memory effect of the initial conditions at increasing soil depths. For example, while March 2018 started with a slightly above-average TSS and experienced a strong meteorological drought leading to an agricultural drought, the initial TSS anomaly in March 2019 was already negative, with a less pronounced precipitation deficit during the vegetation period. This resulted in a much higher forecast skill, because of the memory effect accurately simulated with the physics-based model. Notably, the forecast skill only slightly decreases with increasing lead time, both for precipitation and TSS.

The analysis of the Relative Operating Characteristic Skill Score (ROCSS) for the lower quintile of the TSS distribution assesses whether negative TSS anomalies (i.e., droughts) are adequately represented within the probabilistic forecast ensemble. The results are consistent with those of the CRPSS, showing lower skill in 2018. Nevertheless, the ROCSS analysis overall indicates moderate to high skill for the probabilistic forecast, while the climatology-based pseudo-forecast demonstrates no skill. This confirms that the dry conditions experienced in central Europe in recent years were captured within the probabilistic forecast, underlining the added value of these forecasts and their usefulness in the experimental Water Resources Bulletin.

How to cite: Belleflamme, A., Hammoudeh, S., Goergen, K., and Kollet, S.: Assessment of the skill of seasonal probabilistic hydrological forecasts with ParFlow/CLM over central Europe, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15251, https://doi.org/10.5194/egusphere-egu25-15251, 2025.

Women face disproportionate impacts from climate change due to significant barriers to accessing education and protection. In Indonesia, women often lack access to essential resources and opportunities, particularly in urban informal settlements. However, women also hold a pivotal position in the community in advancing climate literacy. Despite progressive regulations supporting women’s rights, gaps in implementation persist, highlighting the need for targeted initiatives to enhance women’s understanding of climate issues and their capacity to lead resilience efforts. Indonesia has established strong policies for gender equality and climate action, such as Presidential Regulation No. 59/2017 and the National Action Plan for Climate Change Adaptation (RAN-API), which emphasize gender-responsive strategies. However, translating these policies into real-world actions remains a challenge, highlighting the need to better connect scientific research and community insights to effective governance and implementation. This study identifies a critical gap in urban climate literacy and proposes empowering women as a solution. By leveraging women’s social network in Indonesia, the project disseminates climate knowledge and fosters collective action. Key initiatives include training women in climate literacy, introducing sustainable practices such as urban gardening, and developing accessible educational tools like songs, games, and visual materials. These programs are designed to position women as trusted leaders within their communities. Structured monitoring and evaluation methods, including annual surveys and peer-led literacy programs, ensure continuous improvement and scalability. Preliminary findings demonstrate that women-led climate literacy initiatives significantly enhance community resilience and resource allocation. Empowered women influence their families and peers, creating a ripple effect that strengthens societal adaptability. This scalable model integrates women-centered initiatives into governance frameworks, building pathways for sustainable, inclusive development. By empowering women, we transform vulnerability into strength, paving the way for a resilient future.

How to cite: Rachmawati, A. and Dewita, A.: From Policy to Action: Empowering Women to Lead Climate Resilience in Indonesia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15484, https://doi.org/10.5194/egusphere-egu25-15484, 2025.

Timely and fast dissemination are some of the key factors for the effective climate information services in order to support effective climate action. Various mean of communication channel has been used by an authoritative agency to disseminate their climate information, including social media. Currently, social media become one of the most effective chanel to disseminate of weather and climate information. Social media is not only a powerfull tools to ensure the timely, fast, massive dissemination of weather and climate information, but it is also easy to use and access by the general public. Social media also can be optimized public outreach and public education in order to raising awareness and mobilizing an effective climate action. Through it’s real time response tools, social media also can be used to strengthen the engagement between meteorological and hydrological services with their users. Our research will describe the effectiveness of social media to disseminate weather and climate information in order to support climate action in Indonesia.

How to cite: Achmad Fachri, R. and Ezra Reynara, A.: The Use of Social Media on Weather and Climate Information Dissemination To Support Effective Climate Action, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15694, https://doi.org/10.5194/egusphere-egu25-15694, 2025.

Afforestation and greening initiatives are increasingly considered viable strategies for mitigating climate change, particularly in arid regions. In this study, we assess the climate impacts of large-scale afforestation in the Arabian Peninsula (AP). The afforestation is represented by replacing sandy bare soil with woody savanna vegetation, assumed to be naturally sustained by rainfall, in the absence of overgrazing. Using a 30-year regional climate model simulation, we prescribe afforestation within a circular area of 4.5° radius (approximately 71.9 million hectares) centered at 24.2°N, 44.3°E. The afforestation modifies surface characteristics, including darker albedo (0.25 vs. 0.38 for bare soil), a green fraction of 0.3, and a leaf area index (LAI) of 0.1.

Our results show that the afforestation slows down near-surface winds and due to darker surface, increases sensible heat flux, leading to enhanced warming of the atmosphere over vegetated areas. Despite these warming effects, the additional vegetation promotes higher rainfall due to increased moisture availability and reduction of subsidence. This study underscores the dual role of afforestation in modulating regional climate, serving as both a climate mitigation measure and a potential warming source, depending on regional conditions. These findings highlight the importance of considering water availability and local climate factors when designing greening policies for arid regions.

How to cite: Luong, T. M., Zampieri, M., and Hoteit, I.: The Role of Afforestation in Modulating Arid Climate, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19049, https://doi.org/10.5194/egusphere-egu25-19049, 2025.

Subseasonal to decadal predictions provide essential information that bridges the gap in timescales between weather forecasts and long-term climate projections. The science and practice of making such predictions using global climate models initialized with observational data has advanced considerably in recent years, and as a result operational subseasonal, seasonal and decadal prediction services are now a reality. Nonetheless, important remaining challenges must be overcome if these predictions are to more fully realize their potential value for society. This talk highlights five key challenges recommended as targets for focused international research; these are set against a backdrop of wider challenges encompassing climate modelling and services across time scales.

How to cite: Muñoz, Á. G., Merryfield, W. J., and Hudson, D.: Progresses and Challenges for Subseasonal to Interdecadal Prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19889, https://doi.org/10.5194/egusphere-egu25-19889, 2025.

The climate fluctuates on various timescales and in various patterns, giving rise to extreme events over the globe. Skillful predictions of such climate variations would therefore benefit society, and there have been substantial efforts. For the CMIP6 Decadal Climate Prediction Project (DCPP), we performed decadal predictions with ten ensemble members using the Model for Interdisciplinary Research on Climate version 6 (MIROC6). However, since models tend to underestimate signal-to-noise ratio in some sectors, such as the Atlantic, a large ensemble size appears to be required for skillful predictions of those variations. To better understand the predictability on timescales out to a season to a decade, we have prepared a set of initialized predictions using MIROC6 that consists of 10-year-long hindcasts starting every November between 1960-2021, with 50 ensemble members. Compared to the original 10-member ensemble hindcast, both seasonal and decadal prediction skills are broadly improved (e.g., SAT and SLP over southeast China and Scandinavia for the first winter, North and South Pacific SSTs for decadal prediction). Regarding the decadal prediction skill, the impact of initialization is seen up to lead year 7-10 for the North and eastern tropical Pacific Oceans.
Also, building on our experience with decadal climate predictions, we have been working on decadal carbon predictions in recent years. Our efforts on earth system predictions will be introduced as well.

How to cite: Kataoka, T., Tatebe, H., Koyama, H., and Mori, M.: A large ensemble of decadal predictions using MIROC6, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21425, https://doi.org/10.5194/egusphere-egu25-21425, 2025.

Artificial Intelligence and Machine Learning (AI/ML) are gaining increasing interest due to their capacity to increase precision and productivity in the current big data era. Machine learning has indicated its robustness in geosciences, particularly rock-type classification. Lithological classification in the traditional way has raised critical concerns and the need to curb the limitations it breeds, such as time consumption and subjective results. The gold mineralisation occurrence is structurally controlled in the Obuasi Gold district of Ghana. It exhibits complex patterns and relationships that may not be readily discernible through traditional methods, leading to missing out on discovering new resources or potential exploration targets. Consequently, this work attempts to create a predictive model by exploring the best machine-learning algorithms to predict rock types in the Obuasi Gold District using X-ray fluorescence (XRF) geochemical data. Here we established comparative predictive modelling using four supervised classification algorithms: Gradient Boosting (GBoost), Adaptive Boosting (AdaBoost), Support Vector Machine (SVM) and Random Forest (RF). The acquired XRF data was integrated with the model using the Google Collaboratory cloud-based platform. Results show that the performance evaluation of the models indicated SVM as the best algorithm for deployment with a Classification Accuracy (CA) of 0.902. Therefore, ML algorithms have been a great tool in rock-type classification, whereby SVM emerged as the best in the case of the Obuasi Gold District. However, it is encouraged to understand the geology of a particular area before employing the tool and the datasets must be balanced to yield good results and avoid model overfitting.

Keywords: Artificial intelligence; Machine learning algorithm; Support vector machine; Lithogeochemistry; Rock-type classification; Obuasi Gold District

How to cite: Muhammed, A. B., Sunkari, E. D., and Basit, A. W.: Application of Machine Learning Algorithms to Predict Rock Types Using Geochemical Data: A Case Study from the Obuasi Gold District, Ghana, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-275, https://doi.org/10.5194/egusphere-egu25-275, 2025.

Preserving clean water resources and efficiently treating wastewater is critical for ensuring human survival on Earth and in extraterrestrial environments. Major pollutants, including ammonium, heavy metals, industrial dyes, and chemicals, threaten limited clean water supplies and soil. Among the renowned absorbent materials, natural zeolite minerals have demonstrated their effectiveness for pollutant removal compared to clay minerals and synthetic equivalents like biochar, activated carbon, and MOFs, owing to their relatively extensive reserves and eco-friendly nature. Recently, investigating and optimizing pollutant removal rates from water without conducting laboratory experiments is getting more crucial, considering the time-consuming, expensive, and error-prone nature of laboratory testing due to human factors and potential calibration issues among the chosen analytical techniques.

This study aims to forecast the ammonium removal efficiency (% adsorption) and capacity (mg/g) of natural and modified zeolites from aqueous solutions using the regression ensemble LSBoost (MATLAB R2024b) machine learning (ML) algorithm, which is equivalent to XGBoost open-source library. A total of 527 experiments on 15 different zeolite compositions were gathered from a combination of 14 suitable moderately recent (≥ 2005) and highly referred studies to assess the performance of zeolite minerals on ammonium removal rates from aqueous solutions. The LSBoost algorithm achieved over 0.99 R² fitting for training and overall, 0.95 R² for prediction on the quarterly partitioned testing data for both efficiency and capacity. Throughout the improvement of the ML models using different random forest ML approaches, the number of predictors was successfully reduced to 8 based on importance rates among 31 different features in the initial dataset, with a negligible accuracy loss (<0.1 R²) on both training and testing. This research provides a valuable contribution to optimizing applicable experimental parameters in water treatment processes by effectively identifying the significance of predictors within a comprehensive data set. In addition to this, our model not only provides a robust predictive tool for optimizing zeolite performance in water treatment but also represents the first open-sourced web application in the literature to estimate the water treatment performance of zeolites.

How to cite: Akkaş, E., Ünal, B. C., and Ersoy, O.: AI-Based Prediction and Optimization of Ammonium Removal Efficiency and Capacity of Natural Zeolites Using LSBoost (XGBoost) for Sustainable Ecosystems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-809, https://doi.org/10.5194/egusphere-egu25-809, 2025.

In the context of CO₂ storage, cost-effective monitor methods are essential to ensure safe and long-term storage. This work explores the use of seismic time-lapse monitoring, combined with deep learning (DL) techniques, to assess potential leakage and migration pathways. The goal is to develop a cost-effective monitoring method while guaranteeing the safety of storage operations. To this end, we propose a Siamese Neural Network (SNN)-based framework to analyse shot gathers, designed to detect and localize changes within the storage complex. We aim to address the challenges of working with large seismic datasets, enabling the identification of significant events with high confidence, while avoiding the need for event-by-event processing. This framework can allow experts to rely on semi-automatic detections while ensuring human evaluation for interpreting and validating the results.

The proposed SNN architecture processes pairs of shot gather from baseline and monitor surveys in a cross-well configuration. It uses two identical neural networks with shared weights to encode the shot gathers into latent feature embeddings, which are then compared to identify similarities and detect changes. By transforming the data into a shared latent space, the model focuses on capturing relevant patterns while filtering out irrelevant variations, ensuring robust and accurate comparisons. When the SNN detects changes between the baseline and the monitor surveys, it highlights the regions where these changes occur. This approach is particularly effective for identifying subtle but important changes in seismic data, such as those caused by CO₂ migration, which alters the velocity and density of the subsurface. Even in noisy data, the SNN can detect these variations, thanks to its ability to learn features that are highly sensitive to small but meaningful changes. The SNN architecture is scalable and can be adaptable to various seismic monitoring tasks, requiring minimal preprocessing. The proposed framework harnesses the power of deep learning to provide insights into the dynamics of the storage complex, with a focus on identifying changes in time-lapse seismic data related to localized variations. The proposed migration detection tool offers a cost-effective and reliable solution to the modern challenges of gas storage monitoring. This study aims to enable operators to identify and address problems promptly, thereby minimising the impact of potential leakages.

How to cite: Pantaleo, G. and Pipan, M.: Assessment of a deep learning framework for time-lapse seismic monitoring, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-882, https://doi.org/10.5194/egusphere-egu25-882, 2025.

Joint inversion is an essential technique in potential field data processing. The current methodology largely relies on the geology model of anomalous bodies, especially for deep, complex structures. Inspired by the excellent nonlinear mapping capability of the image semantic segmentation model and the advantages of supervised learning, a regressive, end-to-end, encoder-decoder structural, convolutional neural network with a double-branch structure called PFInvNet(Potential Field Inversion Neural Network) is proposed for joint 3D  inversion of physical properties from gravity and magnetic data. Its input is a four-channel dataset consisting of gravity and magnetic anomalies and their vertical gradients, and its output is a 3D matrix representing the spatial distribution of the remnant density and the magnetic susceptibility, which are predicted independently through the double-branch structure of the decoders and then concatenated in the final layer. For network training, a large amount of precisely labeled sample is exceedingly demanding; thus, forward modeling becomes a prerequisite approach. Two discretized forward modeling algorithms for gravity and magnetic anomalies of 3D homogeneous arbitrary-shaped bodies based on surface integrals are deduced and verified with analytic solutions of the sphere model. Furthermore, the neural network needs to learn from the anomalies generated by various forms of abnormal bodies with different physical properties. Therefore, different sizes and quantities of cuboids are randomly distributed in the model space to simulate different forms of abnormal bodies. The label represents the combined spatial distribution of remanent density and magnetic susceptibility for the cuboids, encompassing both spatial location information and physical properties information. With the help of the Marching Cubes(MC) algorithm, the surface of the cuboids can be easily extracted and divided into a triangular surface mesh. The surface mesh is then used to calculate the gravity and magnetic anomalies synchronously through the forward modeling algorithms. The anomalies are concatenated in the channel direction as a sample. A set of optimal network parameters has been determined, including the weight initialization method, the gradient calculation methods, the loss function, the training hyperparameters, the regularization method, and the normalization method. The PFInvNet is trained with 500 and 10000 pairs of samples and labels, respectively. The analysis and comparison of training results prove that PFInvNet has two crucial features: one is that the branch structure enables independent prediction of magnetic susceptibility and remanent density; the other is efficient anti-overfitting ability and efficient solution-finding ability .The prediction error of small samples is very close to that of large samples and is also not obviously enhanced by the noise-contaminated data , demonstrating the strong generalization and robustness of the network. Finally, the network is tested with magnetic and gravity anomalies of the Victoria Land Basin in the western Ross Sea through transfer learning and retraining, and definite 3D distributions of apparent remnant density and apparent magnetic susceptibility have been obtained and can be checked with geological evidences.

How to cite: Jiang, W., Zhao, Y., Gao, J., Ge, S., and Xie, Z.: 3D property inversion of gravity and magnetic data based on a double branch regressive CNN trained by synchronous forward modeling ：a case study of the Western Ross Sea, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1141, https://doi.org/10.5194/egusphere-egu25-1141, 2025.

Shale oil and shale gas are important unconventional resources. Organic matter serves as the primary source of shale oil and gas generation, and high TOC values typically indicate better oil and gas reservoir conditions and production. Therefore, an accurate TOC prediction model is conducive to low-cost evaluation of reservoir hydrocarbon potential and improvement of development efficiency. However, geochemical experimental measurements are costly, and the data obtained is discrete. It is unable to meet the requirements for fine-scale assessment of shale reservoirs. The multiple regression method and ΔlogR method, when directly applied to shale reservoirs, often result in significant errors. In this study, we propose a composite model for accurate TOC prediction in shale reservoirs based on data enhancement and empirically driven. We first address the issue of poorly characterized logging responses and discrete experimental data. The features and quantities of the dataset are enhanced by introducing reconstruction curves and generative adversarial networks (GAN). The validity of the synthesized data is then verified by plotting the data density. In the empirically-driven module, we optimize a density-gamma modified method on traditional ΔlogR method according to the characteristics of shale reservoirs. The modified ΔlogR method will be integrated into the GWO-SVR model as an empirically driven subject in the form of a fitness function. Above, a composite model with both empirical and data-driven components is constructed. We use the Dongying Depression in China as an example for model experiments. The composite model was generalized to wells X and Y. The R² (coefficient of determination) was 0.95 and 0.97, the RMSE (Root Mean Square Error) was 0.31 and 0.29, and the MAE (Mean Absolute Error) was less than 0.3, which indicated a high degree of consistency between the model predictions and the experimental values. Further controlled experiments revealed that the composite model predicted better than the ΔlogR method and the GWO-SVR model alone. Finally, we also performed SHAP interpretability analysis on the model. By revealing the decision-making mechanism inside the model, we verified the rationality of the empirical drive and enhanced the credibility of the model. This provides strong technical support and decision-making basis for the subsequent oil and gas exploration and development work.

How to cite: Hong, Y., Deng, S., Li, Z., and Wei, Z.: TOC Intelligent Prediction Model in Shale Reservoir: Integrating Data Enhancement with Empirically Driven Algorithm, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2156, https://doi.org/10.5194/egusphere-egu25-2156, 2025.

We present a novel approach to gravity forward modeling using conditional neural operators that establishes a forward generative model from the basin models and hyperparameters (reference basement depth, etc.) to gravity anomaly. Our methodology introduces an innovative adaptive embedding mechanism where scalar hyperparameters are first embedded into a 32-dimensional space and then adaptively expanded to match the dimensions of the basin depth model, enabling effective fusion with basin depth model data. Subsequently, Fourier Convolution Layers are employed to transform the fused data into gravity anomalies. The model demonstrates superior performance compared to existing convolutional neural networks on the test dataset, showcasing improved accuracy in capturing complex geological structures and their gravity responses. A key advantage of our architectural design is that it not only preserves the super-resolution capability of conventional neural operators but also enables controlled generation through different hyperparameters. This dual capability allows for both resolution-flexible modeling and parameter-controlled generation, while training on low-resolution data and producing high-resolution outputs, significantly reducing training data requirements and computational costs. The model's adaptive architecture effectively bridges the resolution gap between training and application scenarios, offering a practical solution for real-world geological surveys. Our results suggest that this approach could substantially improve the accessibility and applicability of gravity forward modeling in various geological settings, particularly in regions with limited high-resolution training data.

How to cite: Kang, R., Geng, M., Yang, Q., and Vega, F.: A Conditional Neural Operator Approach for Resolution-Flexible and Parameter-Controlled Gravity Forward Modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2242, https://doi.org/10.5194/egusphere-egu25-2242, 2025.

The combination of big data and AI/ML technologies shows tremendous promise within the domain of disaster management. So that benefits of AI/ML can be realized, and risks can be adverted, internationally agreed upon standards are an important mechanism. These can provide guidance on how to apply (and develop policy around) data collection and preprocessing, model training and evaluation, and operational implementation. In conjunction, they can cultivate interoperability and harmonization of AI-based systems. At the United Nations, the Global Initiative on Resilience to Natural Hazards through AI Solutions brings together experts from different disciplines (geosciences, disaster risk management, computer sciences) and sectors (government, research, NGO) to analyze use cases and lay the groundwork for such standards. Through proof-of-concept projects (e.g., HEU-funded MedEWSa), these standards can be further refined. Finally, through education and capacity building activities, the Global Initiative can help to democratize the responsible use of AI for this domain.

How to cite: Kuglitsch, M. and Xoplaki, E.: International standards for responsible AI in disaster management, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2478, https://doi.org/10.5194/egusphere-egu25-2478, 2025.

Geological discontinuities significantly influence rock mass behaviour. Understanding the origin, setting, and properties of discontinuities is of major relevance, especially in boreholes. Traditionally, manual interpretation of borehole logs is done by geologists, a process that is time-consuming, costly, and subject to variability based on the interpreter's expertise. Recent advancements in artificial intelligence have made it feasible to use machine learning models and automatically detect and differentiate various features in digital images. In this study, we employ a state-of-the-art semantic segmentation model to tackle domain-specific challenges, enabling the identification of discontinuity types (e.g., natural faults, fault zones) and rock mass behaviour features (e.g., breakouts, induced cracks). We applied the SegFormer semantic segmentation model, which integrates a hierarchically structured transformer encoder with a multilayer perceptron (MLP). The borehole data used in this study was collected from the Mont Terri underground rock laboratory. Specifically, we labelled several high-resolution optical logs from one borehole and divided the dataset into training and testing subsets. The borehole considered is an experimental borehole designed to investigate the spatial and temporal evolution of damage around an underground opening in faulted clay shale. Our strategy achieved robust and accurate segmentation results on borehole images. Following segmentation, post-processing techniques were employed to extract critical information such as the total length of induced cracks and the total area of breakouts, as well as their locations and frequencies. The experimental results demonstrate high performance, with the pixel accuracy of 96 % in under three minutes for a 10-meter borehole. Our study lays the groundwork for future research by introducing a powerful tool for extracting geological structures and demonstrating the potential of AI models in geological analysis. By reducing processing time and increasing consistency in the identification, mapping, and classification of geological features, our approach can reveal spatial and temporal patterns associated with the evolution of rock masses.

How to cite: Wang, R., Ziegler, M., Volpi, M., and Manconi, A.: Advances in the identification of geological discontinuities in boreholes with deep learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4298, https://doi.org/10.5194/egusphere-egu25-4298, 2025.

Meteorological data are vast and complex, and their rapid and accurate retrieval is essential for forecasting operations. However, traditional systems have struggled with limited search accuracy and inefficient processing speeds, hindering effective forecast support. To address these challenges, this study developed an AI-based system capable of performing speech recognition, URL search, extreme value detection, and local forecast error analysis. In speech recognition, the Whisper-large model achieved a character error rate (CER) of 3.19%, with GPU memory usage reduced by 15.7% and inference time by 38.18%, enabling real-time processing and scalability in multi-GPU environments. The URL search systems translated natural language inputs into SQL queries and URLs, achieving a Mean Reciprocal Rank (MRR) of 0.92, thereby enhancing data retrieval precision. The extreme value detection systems utilized GPT-4-based template augmentation to expand training data by approximately 111%, significantly improving detection performance and search accuracy. For local forecast error analysis, a prototype chatbot was implemented using prompt engineering and a Text-to-SQL model, allowing for the automated identification of inconsistencies in local forecasts and streamlining the analysis process. These systems have substantially enhanced operational workflows across meteorological tasks, facilitating rapid data retrieval through voice commands, precise responses to complex queries, and real-time analytical support. Future research will focus on further refining these technologies to tackle a wider range of meteorological challenges and integrate them into global forecasting systems for enhanced accuracy and reliability.

How to cite: Kim, B., Shin, H., Cho, A., Park, J., Lee, H., Park, C., Joe, J., Choo, J., and Seo, M.: AI-Enhanced Meteorological Data Retrieval Systems for Improved Forecast Operations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4404, https://doi.org/10.5194/egusphere-egu25-4404, 2025.

The shear wave (S-wave) velocity is a key basis for shale reservoir development, particularly for fracability evaluation. Additionally, S-wave velocity also plays a significant role in prestack seismic inversion and amplitude versus offset (AVO) analysis. However, the actual logging data often lack S-wave velocity data, so it is of significant importance for S-wave velocity prediction. We propose a rapid and precise prediction method for the S-wave velocity in shale reservoirs based on class activation maps (CAM) model combined with physically constrained two-dimensional Convolutional Neural Network (2D-CNN). High sensitivity curves related to S-wave velocity are selected as the foundation. Meanwhile, based on the petrophysical theory of pore medium, the petrophysical model of complex multi-mineral components is established. The dispersion effect is reduced to a certain extent and the results are used to constrain the model. The Adam optimization algorithm is used to construct a 2D-CNN model under the constraint of petrophysical model. The CAM is obtained by replacing the global average pooling (GAP) layer with a fully connected layer, which in turn leads to interpretable results. Then, the model is applied to wells A, B1, and B2 in the southern Songliao Basin. Afterwards, comparisons are made with unconstrained model and petrophysical model. The results show that the correlation coefficients and relative errors in the three test wells are 0.96 and 2.14%, 0.97 and 2.35%, and 0.97 and 2.9%, respectively. The higher prediction accuracy and generalization ability of the new method is confirmed. Finally, we present the defined C-factor as a means of evaluating the extent of concern regarding CAMs in regression problems. The C-factor confirms that the focus of 2D-CNN can be significantly enhanced by incorporating the petrophysical model, thereby imposing physical constraints on the 2D-CNN. In addition, we establish the SHAP model to assist in proving the importance of constraints.

Keywords: S-wave velocity prediction; Physically constrained 2D-CNN; Petrophysical model; Class activation mapping technique; Explainable results

How to cite: Li, Z., Deng, S., and Hong, Y.: S-wave velocity prediction of shale reservoirs based on explainable physically-data driven model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4699, https://doi.org/10.5194/egusphere-egu25-4699, 2025.

Accurate multivariate parametrization of subsurface properties is essential for subsurface characterization and inversion tasks. Deep generative models, such as variational autoencoders (VAEs) and generative adversarial networks (GANs), are known to efficiently parametrize complex facies patterns. Nonetheless, the inherent complexity of multivariate modeling poses significant limitations to their applicability when considering multiple subsurface properties simultaneously. Presently, diffusion models (DM) offer state-of-the-art performance and outperform GANs and VAEs in several tasks of image generation. In addition, training is much more stable compared to the training of GANs. In this work, we consider score-based DMs in multivariate geological modeling, specifically for the parametrization of categorical (facies) and continuous (acoustic impedance – I­_P) distributions, focusing on a synthetic scenario of sand channel bodies in a shale background. We benchmark modeling performance against results obtained by GAN and VAE networks previously proposed in literature for multivariate modeling. As for the GAN and VAE models, the DM was trained with a training dataset of 3000 samples, consisting of facies realizations and co-located I­_P geostatistical realizations. Overall, the trained DM shows significant improvements in modeling accuracy, for all evaluation metrics considered in this study, except for the sand-to-shale ratio, where the values are comparable to those of the GAN and VAE. In particular, the DM is 26% more accurate at reproducing the average (nonstationary) facies distribution and up to 90% more accurate at reproducing the I_P marginal distributions for both sand and shale classes. Higher accuracy is also found in the reproduction of the facies-to-I­_P joint distribution, whereas the spatial I­_P distributions generated by the DM honour the two-point statistics of the training samples. The iterative generative process in DMs generally makes these networks more computationally demanding than VAEs and GANs. However, we demonstrate that with appropriate network design and training parametrization, the DM can generate realizations with significantly fewer sampling iterations while maintaining accuracy comparable to these benchmarking networks. Finally, since the proposed DM parametrizes the joint prior probability density function with a Gaussian latent space, it is straightforward to perform inversion. In addition to improved modeling accuracy, the mapping between the latent and image representations preserves a better topology than that of GANs, overcoming the well-known limitation of the latter for inference tasks, particularly for gradient-based inversion.

How to cite: Miele, R. and Linde, N.: Multivariate generative modelling of subsurface properties with diffusion models , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7000, https://doi.org/10.5194/egusphere-egu25-7000, 2025.

Stratigraphic correlation of well log data is a fundamental step in geosciences. It involves correlating stratigraphic units across multiple wells to build a comprehensive understanding of subsurface geology. Currently, stratigraphic correlation is predominantly performed “manually” by geoscientists. The process is labor-intensive and time-consuming, and interpretations may vary among interpreters due to differences in expertise, experience, and perspective.

Recent advancements in the application of the Dynamic Time Warping (DTW) algorithm have demonstrated its potential to automate and enhance the stratigraphic correlation of well logs. DTW can generate multiple correlation scenarios highlighting different interpretations of subsurface continuity. Thus, the aim of this work is to explore the potential of DTW as a supporting tool in the standard workflows of geoscientists and test it on well log data from IODP Expedition 381 from the Gulf of Corinth. We automatically correlate lithostratigraphic subunits within a 700 m thick stratigraphic unit across two wells, using Natural Gamma Ray (NGR) and Magnetic Susceptibility (MAGS) logs. We selected this dataset because it illustrates the evolution of geological interpretations over time. Between the first version of the IODP data interpretations and a second version published a few years later, significant differences in interpretation were proposed. These differences highlight the critical role of geological expertise in refining subsurface data interpretations and correlations.

The automatic correlations interpreted by DTW showed a minimal average absolute difference with the most recent and updated published correlation, making the human and the machine correlation almost identical. By applying DTW to this dataset, we demonstrate it would have been possible to identify discrepancies and challenges in the interpretations of subunits at the initial stages after data acquisition. This approach could have flagged potential issues even before the IODP data were made available on the public site. Such early identification highlights the potential of DTW as a valuable tool for providing immediate feedback and guiding more accurate stratigraphic interpretations faster.

While DTW significantly reduces the time required for the correlation phase, the time investment needed for data formatting upstream should not be underestimated. Future work on larger datasets will be crucial to better quantify and validate the overall time savings provided by DTW, as well as to optimize the preparatory steps to ensure efficiency in broader applications.

In conclusion, we show that DTW can offer innovative approaches to enhance geological investigations and speed up interpretations. More generally, we consider this work illustrates how data science methods can be leveraged to assist geologists in routine tasks, with our Corinth case study highlighting both the promises and current limitations of digital transformation in well correlations.  

How to cite: Christ, A.-B., Hamieh, Z., Armitage, J., Divies, R., Rohais, S., Mattioni, L., and Bouziat, A.: Dynamic Time Warping algorithm: A geoscience aware AI for automatic interpretation in lithostratigraphy? Insights from an application to the Gulf of Corinth (Greece), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9590, https://doi.org/10.5194/egusphere-egu25-9590, 2025.

In the age of big data, artificial intelligence (AI) is transforming Earth sciences by enabling efficient analysis and visualisation of complex datasets and fostering innovative approaches to solve age-old geoscientific challenges. Kalpa, a Python-based free and cross platform software, represents a pioneering step in this direction. Built with versatility at its core, Kalpa seamlessly integrates AI and machine learning workflows into geoscience applications, offering  customization through its Python plugin architecture. What sets Kalpa apart is its ease of use, even for non-experts. Its intuitive interface lowers the learning curve, enabling a broader audience—including researchers, professionals, and enthusiasts—to leverage advanced geospatial and AI tools without requiring extensive technical expertise.

Kalpa's capabilities span advanced 3D visualization, geospatial data processing, and machine learning model development. It supports global and regional raster and vector dataset visualisation and processing, allowing for interactive analysis in both geographic and cartesian coordinates. With tools to process satellite imagery, geological and geophysical data, uncrewed aerial vehicle (UAV) data, and digital geological maps, Kalpa caters to a wide range of applications, from mineral exploration to natural hazard forecasting. Its machine learning integration supports supervised and unsupervised algorithms for applications such as lithological mapping, mineral prospectivity mapping, land cover and land usage studies, agricultural productivity mapping and natural disaster management. In this study, we demonstrate Kalpa’s transformative potential through three case studies:

• Lithological Mapping in Ladakh, India: Utilizing LANDSAT and SRTM data, we produced accurate lithological maps for this geologically complex region.
• Copper Prospectivity Mapping in Northwest India: Combining remote sensing, geophysical, and geological data, Kalpa predicted copper mineralization zones, with all known deposits falling within areas of predicted probabilities exceeding 0.70.
• Landslide Susceptibility Mapping in Uttarakhand, India: Using remote sensing datasets, Kalpa identified high-risk landslide zones, supporting disaster management efforts.

Kalpa’s user-friendly interface, robust machine learning integration, and publication-ready export capabilities position it as a powerful tool for advancing geoscience research and practical applications. By bridging the gap between domain expertise and cutting-edge AI methodologies, Kalpa empowers Earth scientists, environmental researchers, and GIS professionals to analyze, model, and predict with unprecedented efficiency and precision. This software marks a new frontier in the application of AI to Earth sciences, enabling multidisciplinary research and fostering innovative solutions to pressing geoscientific challenges.

How to cite: Sukhbir, S., Singh, S. P., Singh, U., Kumar, M., and Goyal, T.: Kalpa: Empowering Artificial Intelligence-Driven Geospatial Analysis for Multidisciplinary Applications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9671, https://doi.org/10.5194/egusphere-egu25-9671, 2025.

Artificial intelligence (AI) is transforming landscape archaeology by enabling the automated analysis of high-resolution datasets, such as airborne laser scanning (ALS). The Automatic Detection of Archaeological Features (ADAF) tool is an example of the potential of AI to streamline the identification of subtle surface features and demonstrate their value in uncovering and understanding archaeological landscapes. By improving the detection of archaeological sites, the ADAF plays a crucial role in the research, management and preservation of cultural heritage.

ADAF uses advanced AI models, including convolutional neural networks (CNNs) for semantic segmentation and object detection, to detect features in ALS datasets. The tool has been trained on a large archive of ALS data from Ireland and processes visualised inputs to detect patterns indicative of archaeological structures. The workflow integrates pre-processing with the Relief Visualisation Toolbox, inference with trained AI models and post-processing to refine the results to ensure reliable outputs with minimal false positives.

Designed with accessibility in mind, ADAF features an intuitive user interface that removes the barriers traditionally associated with AI-driven analyses. Users can process ALS data and export GIS-compatible results without the need for specialised knowledge, making the tool suitable for a wide audience. This approach democratises the use of AI in landscape archaeology and extends its utility to professionals and researchers in the field.

Tests with Irish ALS datasets have shown that ADAF is able to detect both known and previously unrecognised archaeological features in the landscape, while enhancing the spatial accuracy of identified sites. By automating complex data analysis, ADAF underlines the efficiency, precision and scalability of AI in landscape archaeology. In addition, the tool contributes to the preservation of cultural heritage by identifying sites that would otherwise remain undiscovered and enabling their preservation and integration into cultural heritage management strategies.

ADAF represents a significant advance in the application of AI in landscape archaeology, providing a powerful and accessible solution for surface feature recognition. Its development underlines the transformative potential of AI to revolutionise the study and interpretation of archaeological landscapes.

How to cite: Coz, N., Kokalj, Ž., Curran, S., Corns, A., Kocev, D., Kostovska, A., Davis, S., and O'Keeffe, J.: Advancing Landscape Archaeology with AI-driven insights from Airborne Laser Scanning data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9968, https://doi.org/10.5194/egusphere-egu25-9968, 2025.

Borehole images (BHI) are crucial for resource exploration, providing detailed fracture analysis at millimeter-scale resolution. However, their interpretation is typically carried out manually, a process that is time-consuming, costly, and subject to significant uncertainty due to interpreter bias and variability. Current state-of-the-art AI methods for automated or semi-automated fracture analysis of BHI often rely on field data for training, using manual interpretations as labels. This approach inherently embeds both aleatoric (data-related) and epistemic (manual) uncertainties, which may undermine the reliability and adaptability of these methods. This study proposes an alternative, synthetic data-driven approach to train a set of two deep neural networks (DNNs) connected in sequence. These DNNs are designed to replicate the primary cognitive tasks involved in manual interpretation: the segmentation of the BHI to identify potential edge zones and the tracing of sinusoids over these edges to approximate their best-fitting 2D representation. By utilizing synthetic data, we are able to systematically assess the sensitivity of both networks and explore various training strategies, including curriculum learning (CL) and self-attention mechanisms. Our proposed solution is designed for post-hoc human-machine collaboration, where the model supports but does not replace human expertise. This framework enables the possibility of a multi-level uncertainty assessment—at the human, machine, and human-machine interface levels—opening the door to new ways of understanding and quantifying the sources of uncertainty in BHI analysis. Additionally, the synthetic data-driven approach ensures the generalizability and scalability of the method, as demonstrated by its successful application to low-resolution logging-while-drilling (LWD) and high-resolution fullbore formation microimager (FMI) datasets from multiple global locations. By combining advanced AI techniques with geoscientific knowledge, this study outlines a potential pathway toward more robust, ethical, and sustainable fracture analysis workflows. Beyond the traditional benefits of reduced cost and time, the approach may provide a scientifically grounded framework for exploring the benefits of human-machine collaboration and uncertainty quantification in geoscience practices. If adopted, this framework could significantly advance the field of BHI analysis, offering new tools for resource exploration in hydrocarbon and geothermal applications.

How to cite: Molossi, A., Roncoroni, G., and Pipan, M.: NeuroFit: a robust and scalable synthetic data-driven deep learning solution for automated borehole image analysis at LWD and wireline resolution, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10472, https://doi.org/10.5194/egusphere-egu25-10472, 2025.

Since 2014, ESA Sentinel missions have been producing an ever-growing amount of data for Earth Observation. This creates the opportunity to monitor changes in high temporal and spatial resolution, however, interpreting this huge quantity of data is challenging. In recent years, the rapid advancement and widespread adoption of applied Artificial Intelligence (AI) methods made it feasible to create deep learning models for specific Earth Observation applications. Combining Sentinel datasets with the appropriate amount of ground truth, robust pre-trained models can be created and applied to produce good-quality thematic maps for different years and large areas. Due to responsibilities related to the European Union’s Common Agriculture Policy (CAP) and the motivation for regional yield estimation, crop classification is one of the most frequently studied remote sensing problems these days. Several papers investigate the possible methods to construct robust and generally functioning models to map the spatial distribution of crops as accurately as possible.

In this study, we present the development of a modular, pre-trained deep learning model designed specifically for crop type mapping. The model is tailored to classify the most prevalent crops in Hungary, including winter and autumn cereals, corn, sunflower, alfalfa, rapeseed, grasslands, and other significant types. For pre-training, we leverage country-wide Sentinel-1 Synthetic Aperture Radar (SAR) data such as Sigma Naught or polarimetric descriptors from H-A-alpha decomposition, collected during the 2021–2024 time period. This dataset comprises annual time series of Sentinel-1 pixels at a spatial resolution of 20 meters. Our approach builds upon prior findings that Sentinel-1-based crop type classification performs comparably to methods using Sentinel-2 optical data. However, Sentinel-1 has the added advantage of producing consistent and regular time series, as it is unaffected by atmospheric conditions such as cloud cover.

The proposed model employs a hybrid methodology integrating self-supervised and fully supervised learning paradigms. This modular architecture allows for seamless integration of task-specific classifiers, which can be fine-tuned using supervised learning to address both current and future classification requirements effectively. The fully supervised component is supported by an extensive ground truth dataset covering over 60% of Hungary’s total land area. This proprietary dataset is derived from the national agricultural subsidy database, providing detailed and accurate annotations. The abundance and quality of labeled data enable the construction of robust, highly generalizable models, ensuring reliable performance across diverse classification tasks. This methodology offers great potential to advance national-scale operational tasks, such as early land cover prediction and annual crop type mapping.

How to cite: Richter-Cserey, M., Simon, M., Magyar-Santen, G., Pacskó, V., and Kristóf, D.: Exploring Pretraining Possibilities of Crop Classifiaction Models Using Large-Scale Sentinel-1 Datasets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11754, https://doi.org/10.5194/egusphere-egu25-11754, 2025.

Advanced constitutive models, here represented by the hypoplastic clay model, are powerful tools which provide engineers with reliable responses in various practical applications. However, the model calibration is not an easy task. Calibration of these models can be addressed with several approaches, which are generally distinguished as stochastic or deterministic approaches. In general, these approaches extract information from the experimental data and the subsequent optimisation process finds the best combination of parameters to fit the desired constraints. . The deterministic approach was integrated and combined in development of the online automated calibration tool ExCalibre. This paper presents a Machine Learning approach for automated calibration of the Hypoplastic Clay model. By using pairs of input experimental data and calibrated results performed by ExCalibre as training data, a Deep Neural Networks (DNNs) model is constructed to recognise how the experimental data can be used to derive the asymptotic state parameters such as the slope and the interception of the Normal Compression Line (NCL), or the critical friction angle, and the optimised stiffness parameters. The training and testing data comprise of In-house protocols and User-upload data over 3 years of launching the ExCalibre, and synthetic data with small distortion to prevent overfitting. Finally, investigations on how the DNNs model recognises the asymptotic patterns, as well as its calibration results will be presented.

How to cite: Do, P. and Kadlicek, T.: Calibration of the Hypoplastic Clay model with a deep neutral network, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12065, https://doi.org/10.5194/egusphere-egu25-12065, 2025.

Accurate estimation of Particle Size Distribution (PSD) in rock avalanche deposits is essential for understanding the fragmentation processes and spatial distribution characteristics during mass movement. However, traditional methods, such as physical sieving or visual field estimation, are time-consuming, labor-intensive, and impractical for large-scale field measurements. To address these limitations, this study presents an automated PSD estimation framework that combines UAV imagery and deep learning-based segmentation. A synthetic dataset was used to train the segmentation model, improving its robustness across different scenarios. Image resolution adjustments were applied to improve detection accuracy for small and overlapping particles. Additionally, Fourier analysis was utilized to reconstruct smooth and continuous particle contours, to effectively handle overlapping particles. The reconstructed 2D outlines were further used to estimate 3D particle volumes through the shape-volume model based on laboratory and literature data. Projection correction was applied to mitigate image distortions to ensure precise volume predictions. The proposed approach overcomes the limitations of traditional methods dealing with complex particle distributions in real field environments. The results demonstrate the effectiveness of the proposed method for large-scale particle detection and volume estimation, providing new insights into rock avalanche fragmentation dynamics.

Keywords: Rock avalanche; Particle size distribution (PSD); deep learning; UAV Imagery

How to cite: Lin, R., Jaboyedoff, M., Derron, M.-H., and Lu, T.: Automated Particle Size Distribution Estimation of Rock Avalanches using Deep Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12242, https://doi.org/10.5194/egusphere-egu25-12242, 2025.

Hazelnuts are a significant crop, with global production exceeding 1.25 million tons by 2023 (INC, 2023). Quality is threatened by biotic agents, including insects causing the cimiciato defect. This defect, from insect bites during fruit growth, results in off-flavor, tissue alterations, and lipid oxidation (De Benedetta et al., 2023). Damage can be external or internal (hidden cimiciato). Industrial quality standards often exceed official regulations, making effective selection crucial. Traditional visual inspection is time-consuming and subjective. Non-destructive methods like NIR and NMR have potential but limited applicability. Deep Learning (DL) has revolutionized image classification, proving effective in agriculture, including disease and pest management (Mohanty et al., 2016; Dhaka et al., 2021; Meena et al., 2023). This study explores Deep Learning (DL) for automated detection of the cimiciato defect in hazelnuts using X-ray radiographs. Cimiciato, caused by insect feeding, degrades hazelnut quality, requiring product selection. Traditional methods are time-consuming and subjective. We propose a Convolutional Neural Network (CNN) model trained on X-ray images to classify hazelnuts as healthy or infected. Results demonstrate the model's effectiveness, offering a non-destructive, automated quality control solution.
Radiographs were acquired using a cone-beam micro-tomograph. Each hazelnut was positioned on a rotating stage. A CNN model was used for classification. CNNs effectively extract features from images. Convolutional layers apply filters to identify features; pooling layers reduce data dimensionality; fully connected layers combine features for classification. The Inception-ResNet-V2 architecture was chosen, combining Inception modules and residual connections (Szegedy et al., 2017). The model was trained (128 image batch size, 0.001 learning rate, 30 epochs), comparing SGD, ADAM, and RMSP optimizers. Images were pre-processed: resizing, pixel normalization, and data augmentation. 
The test dataset evaluated the trained network. SGD, ADAM, and RMSP yielded similar results. Confusion matrices visualize performance. ADAM performed best, but all achieved good results, especially for cimiciato detection. 

Keywords: Cimiciato defect, Hazelnut, Deep Learning, X-ray radiography, CNN

How to cite: Napolitano, A. G., Mele, G., Gargiulo, L., Giaccone, M., and Vitale, A.: Automatic Detection of Cimiciato Defect in Hazelnuts Using Deep Learning and X-ray Radiography, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12675, https://doi.org/10.5194/egusphere-egu25-12675, 2025.

Artificial Intelligence (AI) is revolutionizing geosciences, aligning perfectly with the themes of session GI2.4 by enabling the analysis of complex, multidimensional datasets and delivering actionable insights at unprecedented scales. This study presents two innovative AI-driven frameworks addressing critical challenges in soil moisture prediction and geomagnetic disturbance forecasting. Both approaches leverage decentralized networks to achieve scalability, foster collaboration, and enhance model performance through continuous refinement by a distributed community of contributors.

For soil moisture prediction, our multi-stream base model integrates data from Sentinel-2 (high-resolution spectral imagery), SMAP L4 (volumetric water content), ERA5 (meteorological variables), and SRTM (elevation data). The model predicts surface and rootzone soil moisture with six-hour lead times, achieving RMSE values of 0.1087 m³/m³ and 0.1183 m³/m³, respectively, across diverse Köppen-Geiger climate zones. By utilizing a decentralized network, contributors perform inference on 100 km² global regions, generating predictions evaluated against SMAP data using Root Mean Square Error (RMSE) and R² metrics. This system ensures robust model performance while addressing the spatial and temporal gaps inherent in traditional observational networks. These advances have significant implications for agriculture, hydrology, and climate modeling, enabling better water resource management, crop planning, and drought mitigation strategies. 

In geomagnetic disturbance forecasting, our GeoMagModel leverages Prophet, a time-series forecasting library, to predict the Disturbance Storm Time (Dst) index, a key indicator of geomagnetic activity. The model achieves an RMSE of 6.37 for December 2024 datasets, effectively capturing both trend shifts and weekly seasonality. The decentralized community enhances predictive accuracy by dynamically integrating historical and real-time Dst data, which is validated by benchmark predictions of the Kyoto World Data Center’s hourly records. This approach provides near-real-time forecasts critical for safeguarding power grids, satellite systems, and other infrastructure vulnerable to space weather events.

By integrating machine learning with decentralized computing and state-of-the-art data sources, these frameworks offer scalable solutions to longstanding challenges in geophysical monitoring. The decentralized network not only improves scalability but also incentivizes the geoscience community to refine baseline models, fostering innovation and enabling systems to outpace state-of-the-art benchmarks. The implications of this work extend beyond immediate applications, paving the way for hybrid models that combine AI-driven predictions with physical process-based simulations. This fusion has the potential to improve understanding and resilience in critical domains such as water resource planning, disaster mitigation, and space weather forecasting. By addressing the limitations of traditional observation systems and delivering actionable insights at scale, these AI-driven frameworks represent a paradigm shift in how we approach and solve complex geoscientific problems.

How to cite: Moraga, G., Hristopoulos, S., and Pearson Kramer, N.: Harnessing AI and Decentralized Networks for Next-Generation Geophysical Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13315, https://doi.org/10.5194/egusphere-egu25-13315, 2025.

Small satellite platforms are increasingly used for Earth observation due to their cost-effectiveness and flexibility. However, their limited payload size often results in reduced spatial resolution of captured images. In our work, we address this challenge by proposing an advanced multi-image super-resolution (MISR) approach tailored for small satellite applications.

It integrates:

• Sub-pixel image registration and on curvelet transform-based interpolation to preserve high-frequency details while reducing artifacts;
• A novel hybrid method called SP-MISR (Subpixel Multi-Image Super-Resolution), which leverages Convolutional Neural Networks (CNNs) for local detail analysis and Transformers for global spatial relationships.

Our experimental results demonstrate that this combined approach  significantly improves image sharpness, preserves fine details, and reduces artifacts, outperforming traditional super-resolution techniques. Moreover, SP-MISR exhibits robustness in processing noisy and distorted images, making it particularly suitable for the constrained imaging systems of small satellites.

Future developments will focus on improving computational efficiency, reducing interpolation errors, and extending the method to multi-spectral imaging and interplanetary missions, by exploring explore pure deep learning techniques.

This work highlights the potential of integrating traditional and deep learning methodologies to enhance image quality, thus expanding the scientific and operational capabilities of small satellite missions.

How to cite: De Martino, C., Della Corte, V., Inno, L., Cozzolino, F., Ruggiero, G., Da Deppo, V., Zuppella, P., Moualla, L., and Venafra, S.: Advanced Super-Resolution Techniques for Optical Payloads in Earth Observation: Combining Traditional and Deep Learning Methods, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15949, https://doi.org/10.5194/egusphere-egu25-15949, 2025.

Soil carbonates are critical players in the global carbon cycle and have a profound influence on soil health and agricultural productivity. Their quantification is also central to carbon sequestration efforts, where accurate measurement of soil carbonates can inform strategies for reducing atmospheric CO₂. However, conventional methods of carbonate analysis in soil---while effective---are often slow, costly, and labor-intensive ⁽¹⁾.

In this study, we introduce Shifted Excitation Raman Difference Spectroscopy (SERDS) as a rapid, non-destructive alternative, further enhanced by advanced preprocessing techniques and machine learning algorithms. Specifically, we employ Asymmetric Least Squares (ALS) for background correction, Standard Normal Variate (SNV) for normalization, and Savitzky–Golay filtering for smoothing. Unlike conventional Raman spectroscopy, SERDS effectively eliminates background fluorescence and reduces overlapping peaks, resulting in clearer spectral signatures ⁽²⁾. 

We employed Partial Least Squares Regression (PLSR) and eXtreme Gradient Boosting (XGBoost) to predict the inorganic carbon content from the carbonate vibrational modes in conventional Raman and SERDS spectra, benchmarked against total inorganic carbon (TIC) measurements from coulometric titration. Our results show that switching to dual-laser SERDS substantially boosted model performance. For PLSR, the coefficient of determination (R²) improved from 0.8 to 0.88 (an increase of about 10.5%), and the root-mean-square error (RMSE) declined from 0.29 to 0.22 (26% decrease). The XGBoost model exhibited an even greater increase, with R² increasing from 0.63 to 0.93 (approximately 49% improvement) and RMSE dropping from 0.39 to 0.16 (59% reduction).

Figure 1: Left: All SERDS data of soil samples showing the main carbonate peak; Right: XGBoost model prediction of soil inorganic carbon using SERDS data.

These findings underscore the potential of SERDS to replace conventional methods for carbonate quantification, offering reduced cost, faster analysis, and essentially no sample preparation. Furthermore, by providing highly accurate carbonate measurements, this methodology can be pivotal for carbon sequestration assessments and large-scale soil management practices, helping to advance both environmental sustainability and agricultural productivity.

References:

1) Barra I, Haefele SM, Sakrabani R, Kebede F. Soil spectroscopy with the use of chemometrics, machine learning and pre-processing techniques in soil diagnosis: Recent advances–A review. TrAC Trends in Analytical Chemistry. 2021 Feb 1;135:116166.

2) Orlando A, Franceschini F, Muscas C, Pidkova S, Bartoli M, Rovere M, Tagliaferro A. A comprehensive review on Raman spectroscopy applications. Chemosensors. 2021 Sep 13;9(9):262.

How to cite: Poursorkh, Z., Solomatova, N., and Grant, E.: Quantifying Soil Inorganic Matter: Integrating Shifted Excitation Raman Difference Spectroscopy (SERDS) with Machine Learning for Enhanced Analysis of Carbonates, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16705, https://doi.org/10.5194/egusphere-egu25-16705, 2025.

Seismic exploration heavily relies on the accurate processing of seismic data, as high-quality reconstructed data is essential for reliable imaging and interpretation. In recent years, data-driven approaches have shown great promise in seismic data processing. However, supervised learning methods require large amounts of labeled data, while generative models, such as GANs, often encounter issues like mode collapse and instability. On the other hand, generative diffusion models, leveraging principles from nonequilibrium thermodynamics and Markov processes, have emerged as powerful tools for capturing complex data distributions.

Despite these advantages, Denoising Diffusion Probabilistic Models (DDPM) purely generate data distributions from the latent space with the reliance on random noise, making it inadequate for seismic data reconstruction where the goal is to accurately recover missing traces. Thus, DDPM is  lacks interpretability in seismic data restoration, and may disrupt the structured patterns crucial for interpolating seismic signals. Furthermore, we view the reverse process that starts from noise as unnecessary and inefficient for reconstruction task.

To address these challenges, we propose a novel Conditional Residual Diffusion Model (CRDM) that enhances both certainty and interpretability by incorporating residual diffusion and conditional constraints derived from observed seismic data (Fig.1). This approach better aligns with the inherent structure of seismic signals, enabling more accurate and interpretable reconstruction. The model is grounded in DDPM, with mathematical derivations for loss functions, conditional probability distributions, and reverse inference steps, ensuring both theoretical rigor and practical applicability.

Additionally, Our CRDM utilizes a shallow U-Net architecture featuring one down-sampling and one up-sampling layer integrated with Multi-Head Self-Attention (MHSA), which significantly enhances the model's efficiency and effectiveness. Experimental results (Fig.2) demonstrate that CRDM outperforms DDPM, denoising convolutional neural network (DnCNN), and fast projection onto convex sets (FPOCS), achieving a 15.1% improvement in reconstruction SNR and reducing computation time by 139 times compared to DDPM. Notably, CRDM achieves optimal results in a few diffusion steps, whereas DDPM typically requires thousands of steps.

The innovative approach generates data through residuals for determinism, while guiding the processing with noise for diversity. This not only enhances the interpretability and efficiency of seismic data reconstruction, but also positions the model as a promising tool for advancing data-driven seismic processing through flexible coefficient adjustment. Therefore, we believe this model has great potential for broader applications in geophysical data analysis, offering significant value in accurately depicting complex geological structures and providing more effective guidance for petroleum exploration.

Fig.1 The framework of CRDM. The model consists of two stage: (a) the training stage with forward diffusion process; (b) the sampling stage for seismic data reconstruction.

Fig.2 Reconstruction results and residuals of the 1994 BP seismic data with 50% irregular missing traces. (a) Complete data, reconstruction using (b) FPOCS (SNR=10.5dB), (c) DnCNN (SNR=15.6dB), (d) DDPM(SNR=18.4dB), (e) CRDM(SNR=21.2dB), and (f) observed seismic data with 50% missing traces. (g-j) display the residuals corresponding to reconstructions (b-e), respectively. The red box highlights a zoomed-in region, which is shown in detail in (k-t).

How to cite: Gong, X. and chen, S.: Enhancing Seismic Data Reconstruction with a Conditionally Constrained Residual Diffusion Model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16835, https://doi.org/10.5194/egusphere-egu25-16835, 2025.

Recent trends in climate change and global warming have amplified the occurrence of severe convective initiation (CI) and other extreme weather events, underscoring the importance of high-frequency remote sensing. Modern satellite constellations such as EUMETSAT, GOES, Himawari, and GK2A now offer global monitoring intervals of 10 minutes and regional updates as frequent as 1–2 minutes. However, these capabilities are not universally accessible—many developing countries lack in-orbit assets or historical high-frequency data archives.

This study presents a zero-shot video frame interpolation (VFI) approach to generate high-frequency (1–10 minute) satellite imagery from legacy or sparsely sampled observations. Leveraging a flexible Many-to-Many Splatting VFI model, our framework avoids domain-specific retraining while delivering reliable intermediate frames. We validate the method using overlapping data from the KMA GK2A (10-minute full-disk, 2-minute Asia-Pacific) and KMA COMS (3-hour full-disk, 15-minute Asia-Pacific) satellites over the period July 25, 2019, to March 31, 2020.

Our results indicate significant improvements in both PSNR and SSIM metrics, confirming the model’s efficacy in three critical applications:

• Up-sampling Archived Geostationary Data

  • Enhancing the temporal resolution of older satellite imagery (e.g., COMS 30-minute or 3-hour intervals) to match or approximate modern satellite capabilities (e.g., GK2A 10-minute intervals). This harmonization facilitates unified climatology analyses spanning multiple generations of instruments.

• Sensor-Error Correction and Gap Filling

  • Recovering missing or corrupted frames resulting from attitude adjustments, sensor calibrations, or malfunctions on geostationary satellites. Ensuring a continuous record provides more robust inputs to operational forecasting and climate assessments.

• Delayed High-Frequency Observation Services

  • Enabling resource-constrained meteorological agencies to retrospectively produce and disseminate high-frequency satellite products (e.g., near 1-minute intervals) to improve nowcasting, risk assessment, and disaster preparedness.

Our preliminary findings show minimal computational overhead per inference step, making this cost-effective method feasible for near-real-time deployment and post-event analyses alike. By bridging temporal gaps in global satellite datasets, this technique supports advanced level-2 products such as cloud-tracking and convective-initiation alerts, thereby driving broader socioeconomic and scientific benefits in both developed and developing regions.

How to cite: Ryu, H. and Choi, Y.: Toward High-Frequency Satellite Observations in Data-Sparse Regions: A Zero-Shot Interpolation Framework for Missing and Historical Imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17456, https://doi.org/10.5194/egusphere-egu25-17456, 2025.

Tomographic methods, such as electrical resistivity tomography (ERT), induced polarization (IP) and seismic refraction tomography (SRT) are often effective for detecting geophysical targets in disparate real-world scenarios. However, a final reconstruction expressed only in terms of individual geophysical parameters (resistivity, chargeability, P-wave velocity) leaves room for ambiguity in complex sites exhibiting several transitions between layers or zones having different geophysical properties. In such cases, the sensitivity of the geophysical parameters for the various methods can differ significantly, so that a univocal interpretation based only on a visual comparison of the different models is often ineffective. To overcome these limits, in this work we present a machine learning-based quantitative approach for the detection of geophysical targets associated with both geological and anthropogenic scenarios. We integrate two-dimensional ERT, IP and SRT tomographic data with a soft clustering analysis by the Fuzzy C-Means (FCM) to obtain a final combined section, where each pixel is characterized by a cluster index and an associated membership value. The membership function of the Fuzzy C-Means is a good estimator of the accuracy of the subsurface reconstruction, as it ranges from 0 to 1, with 1 reflecting a high reliability of the clustering analysis. We apply this method to two case studies, related to the detection of leachate accumulation areas in a municipal solid waste landfill and to the bedrock characterization in a site prone to instability. In both cases, we detect the cluster associated with the geophysical targets of interest and our final sections are validated by a good agreement with the available direct information (boreholes and wells). The accuracy of the reconstruction is consistently high across most areas (membership values > 0.75), even though it is reduced in areas where the resolution of geophysical data is lower. Therefore, this approach may be a valuable automatic tool for optimizing the cost-effectiveness of projects where new constructions or remediation interventions have to be planned.

How to cite: De Donno, G., Cercato, M., Melegari, D., Paoletti, V., Penta de Peppo, G., and Piegari, E.: Fuzzy clustering of electrical and seismic data for the detection of geophysical targets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17957, https://doi.org/10.5194/egusphere-egu25-17957, 2025.

The Frequency Domain Electromagnetic (FDEM) method is a cost-effective geophysical technique that simultaneously studies the electrical and magnetic properties of a medium, providing data as in-phase and out-of-phase components of the electromagnetic field. Although FDEM yields valuable insights, its results can be complex to interpret, and the two EM field components are normally only visually inspected to support findings from other techniques. This study aims to enhance FDEM data interpretation using an unsupervised learning technique. The proposed approach seeks to automate and expedite the interpretative phase. By applying the K-Means clustering algorithm, we divided the FDEM data into several clusters based on specific intervals of the in-phase and quadrature components, resulting in integrated maps of EM components. Combining these maps with geological and archaeological insights helped identifying areas of potential archaeological interest. This method was applied to the Torre Galli archaeological site in Calabria, Italy, known for its significance in Iron Age studies.

Based on comparisons with the findings of earlier excavations and results from a magnetic survey, the proposed procedure shows promise in improving the efficiency and accuracy of the FDEM method in identifying areas of archaeological interest. This suggests that automating the interpretation process could lead to a better cost management and time optimization in geophysical and archaeological studies.

How to cite: Capozzoli, A., Paoletti, V., Cella, F., La Manna, M., and Piegari, E.: Integrating FDEM data with K-Means clustering for improved archaeological site identification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18082, https://doi.org/10.5194/egusphere-egu25-18082, 2025.

Artificial Intelligence (AI) is revolutionizing the field of geomorphology, offering a robust tool for objective and quantitative analyses. This pioneering study proposes an innovative framework based on Machine Learning clustering techniques, capable of classifying drainage patterns into multiple morphological classes. This work follows up on a related study in which an attempt at classifying 156 terrestrial and extraterrestrial (Mars and Titan) river networks was made. Rivers’ outlines are intrinsically noisy, difficult to isolate from the background, and can be ambiguous for the human eye. The previous works have been focused on accurately classifying patterns, using the expertise of morphologists, thus introducing a weak link, the human eye, in the chain. This time, a reliable, automatic, and scalable methodology has been obtained, leveraging computers’ precision, objectivity, and computational power. The HydroRIVERS dataset, a publicly available data bank containing vector data, was utilized in this study. All HydroRIVERS data layers are provided in a geographic projection (latitude/longitude), referenced to the WGS84 datum. Each data layer includes an attribute table with information on the morphometric characteristics of each river reach. The input parameters for the clustering models included morphometric features such as LENGTH_KM, DIS_AV_CMS, ORD_STRA, ORD_CLAS, and ORD_FLOW.
During a preliminary experiment, a local convexity test was conducted to determine the optimal number of clusters (k) to identify the best metric values. This test made sure that the number of clusters with the highest evaluation metric was selected, varying in a closed numeric interval. Each cluster corresponds to a specific river class. Significant results were obtained with k = 6, k = 8, k = 10, and k = 12. Subsequently, the K-Means algorithm was applied, grouping the dataset into distinct clusters based on the morphometric parameters. The results were remarkable, with 10 being the best value for k. The results indicate that the clustering algorithm is able to optimally separate the dataset, producing a high inter-cluster distance and a low intra-cluster distance. The dataset points along the features, as highlighted by the three principal components obtained by performing PCA on the final five-dimensional clustering resulting vector space, are well grouped in relatively small clusters, far away from each other. The next step involves using the centroids obtained from the analysis of the large dataset as a reference for classifying the 155 rivers. In general, the centroids obtained from this kind of Learning could be of great value to the scientific community, establishing a new and innovative way of discerning between different classes of rivers without having to manually analyze and inspect images. This approach promises efficient and accurate classification of both terrestrial and extraterrestrial drainage patterns.

How to cite: D'Aniello, M. and Donadio, C.: A novel approach using Machine Learning to objectively classify terrestrial and extraterrestrial river networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19838, https://doi.org/10.5194/egusphere-egu25-19838, 2025.

The big developments in Artificial Intelligence (AI) generated a mixture of fascination and apprehension, echoing historical responses to unknown phenomena. Like how medieval European societies interpreted certain natural events and instruments as magical, AI is currently perceived by many as a “black box,” blurring the lines between advanced technology and mystical force. This phenomenon fosters misconceptions and uncertainties about AI’s actual mechanisms, benefits, and risks. Consequently, it underscores the urgent need for science communicators and science museums to adopt an active role in enhancing the general public’s AI literacy and in debunking some of its enigmatic traits.

By drawing parallels with the Middle Ages—when objects such as mirrors and magnetite were often attributed supernatural capabilities—modern AI tools LLMs or image and video generators are frequently viewed as possessing a “magical” principle. The public’s limited grasp of how AI processes inputs and produces outputs further intensifies this impression. The lack of transparency (or “black box” effect) in deep learning algorithms, combined with the ambiguity of human language, has been shown to fuel both wonder and anxiety.

Science museums and communicators should have an active role by offering educational programs that demystify AI through different demographic and social activities as well as promoting the public debate. These initiatives could clarify AI’s underlying mathematical and computational principles, highlight practical examples of AI-driven applications, and examine ethical considerations surrounding its deployment. Public understanding of AI’s capabilities and limitations is crucial not only to temper undue fears but also to encourage informed engagement with emerging technologies.

How to cite: Azevedo Rodrigues, L.: Making Sense of AI: The Important Role of Education and Communication, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20509, https://doi.org/10.5194/egusphere-egu25-20509, 2025.

While precipitation is the primary driver of streamflow variability, temperature also plays a significant role. Temperature influences streamflow by modifying precipitation, evapotranspiration, and soil moisture. While this relationship is often studied using hydrological or black-box models, the causal effect of temperature dynamics on streamflow at the catchment scale is not fully understood. This study investigates the causal relationship between precipitation, temperature, and streamflow time series using the PCMCI+ causal discovery method. Having the causal structure, the total causal effect of temperature on stream flow is estimated. The analysis is conducted on CAMELS-GB (671 catchments) and LamaH (859 catchments) datasets to study the causal effects of temperature on streamflow across a wide range of catchments with different climate and physiographic characteristics. Preliminary results indicate that temperature significantly influences streamflow within a specific range, which changes over time for most catchments. The changes in the range within which the temperature has high causal effects on the temperature might be due to the shift in catchment storage and precipitation patterns, leading to a change in catchment response to temperature. These findings highlight the importance of identifying a relationship between temperature streamflow variability from a cause-and-effect perspective. This suggests that incorporating causal information can improve the modelling of the hydrological systems under changing climate. 

How to cite: Abbasizadeh, H., Maca, P., and Hanel, M.: Influence of Temperature on Streamflow Dynamics: A Multi-Catchment Analysis Using the PCMCI+ Causal Discovery Algorithm, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-292, https://doi.org/10.5194/egusphere-egu25-292, 2025.

River discharge prediction is critical for water resource management, yet equifinality—where multiple model configurations achieve similar accuracy—complicates process understanding. We explored this phenomenon using Long Short-Term Memory (LSTM) models trained on UK river basins, incorporating geomorphic descriptors derived from Digital Terrain Models and other environmental features from the CAMELS-GB dataset, including land cover, soil, and climate variables. Explainable AI techniques revealed that the models rely on different, yet equally effective, combinations of correlated features to achieve comparable performance. This variability underscores the complexity of hydrological systems and highlights the importance of integrating explainability and domain knowledge in machine learning to enhance model interpretability and robustness.

How to cite: Chen, Q., Mudd, S., and Moulds, S.: Equifinality in River Discharge Prediction Revealed Through Explainable AI , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-482, https://doi.org/10.5194/egusphere-egu25-482, 2025.

Accurate modeling of soil water movement in the unsaturated zone is essential for effective soil and water resources management. Physics-informed neural networks (PINNs) offer promising potential for this purpose, but necessitate retraining upon changes in initial or boundary conditions, posing a challenge when adapting to variable natural conditions. To address this issue, inspired by the operator learning with more universal applicability than function learning, we develop a physics-informed deep operator network (PI-DeepONet), integrating physical principles and observed data, to simulate soil water movement under variable boundary conditions. In the numerical case, PI-DeepONet achieves the best performance among three modeling strategies when predicting soil moisture dynamics across different testing areas, especially for the extrapolation one. Guided by both data and physical mechanisms, PI-DeepONet demonstrates greater accuracy than HYDRUS in capturing spatio-temporal moisture variations in real-world scenario. Furthermore, PI-DeepONet successfully infers constitutive relationships and reconstructs missing boundary flux condition from limited data by incorporating known prior physical information, providing a unified solution for both forward and inverse problems. This study is the first to develop a PI-DeepONet specifically for modeling real-world soil water movement, highlighting its potential to improve predictive accuracy and reliability in vadose zone modeling by combining data-driven approaches with physical principles.

How to cite: Ye, Q., Huang, Z., Zheng, Q., and Zeng, L.: Predicting Water Movement in Unsaturated Soil Using Physics-Informed Deep Operator Networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1863, https://doi.org/10.5194/egusphere-egu25-1863, 2025.

Recent applications have demonstrated the strength of deep learning (DL) in information extraction and prediction. However, its limitations in interpretability have delayed its popularity for use in facilitating advancement of hydrologic understanding. Here we present a framework using explainable artificial intelligence (XAI) as a diagnostic tool to investigate distributed soil moisture dynamics within a watershed. Soil moisture and its movement generated by physically based hydrologic model were used to train a long short-term memory (LSTM) network, whose feature attribution was then evaluated by XAI methods. The aggregated feature importance presents abrupt rise in the model’s nodes located in riparian area, indicating threshold behavior in runoff generation and development of hydrologic connectivity at the watershed scale, which helps explain the rapid increase in streamflow. This work represents a demonstration of the potential of XAI to uncover underlying physical mechanisms and to help develop new theories from observed data.

How to cite: Ye, S., Li, J., Chai, Y., Liu, L., Sivapalan, M., and Ran, Q.: Using explainable artificial intelligence as a diagnostic tool , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2661, https://doi.org/10.5194/egusphere-egu25-2661, 2025.

How to cite: Huynh, N. N. T., Garambois, P.-A., Renard, B., and Monnier, J.: A General Framework for Integrating Neural Networks into Numerical Resolution Methods for Spatially Distributed Hydrological Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2740, https://doi.org/10.5194/egusphere-egu25-2740, 2025.

Flood susceptibility mapping (FSM) plays a crucial role in proactive flood risk management, particularly in light of increasing fluvial flooding events. Traditional FSM methods, such as physics-based and qualitative approaches, are hindered by either high computational demands or inherent uncertainty. To address this, machine learning (ML) models have become an increasingly popular FSM approach, though commonly cited as black-box approaches due to the difficulty associated with understanding their underlying mechanisms. In order to better understand the ML approaches used for FSM, this study uses the gradient-weighted class activation mapping (Grad-CAM) to interpret flood susceptibility predictions of a convolutional neural network (CNN) for the Don River watershed in Ontario, Canada. Grad-CAM is an explainable algorithm highlighting input regions that are influential to the output, aiding the user in understanding and visualizing model selected important features used to arrive at the prediction. Grad-CAM results are compared to the commonly used shapley additive explanation (SHAP) algorithm. SHAP is used to calculate the relative contribution of each input onto the output, and provides a benchmark for comparisons due to its popularity.

A two dimensional CNN with an architecture of two convolutional layers, two pooling layers and a fully connected layer is used to predict flood susceptibility. The inputs to the CNN include topographical and climactic variables across the entire watershed, with a 60-40% training and testing split respectively. The results of the CNN were compared against the floodplain map of the Don River. Using the area under curve- receiver operating characteristics (AUC-ROC) as a performance metric, the CNN exhibits high performance with an AUC-ROC of 0.96.

The study highlights the potential of CNNs for flood susceptibility mapping, as well as compares two explainable machine learning algorithms, helping to further their application within FSM. Explainable algorithms are essential to decision makers in flood risk management for proactive planning and resource allocation. Future work should explore expanding the scope to predict flood susceptibility at a nationwide level.

How to cite: Khalid, R. and Khan, U. T.: Explainable convolutional neural network for flood susceptibility mapping in Southern Ontario  , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3859, https://doi.org/10.5194/egusphere-egu25-3859, 2025.

Floods are the costliest hazard in Canada in terms of direct infrastructure damage. Flood susceptibility modelling (FSM) identifies flood hazard areas; input features are dependent on the study area and modelling methods, which affect the reliability and accuracy of FS maps. Typical features in FSM are static topographical inputs (digital elevation model, land use, wetness index, height above nearest drainage, etc.). Though meteorological variables have been included in FSM, they are often low temporal resolution (e.g. annual); seasonal meteorological variables are often not included. The 2023 Canadian National FS map was developed using machine learning (ML) ensembles, with features that include historical flood events and 30 years of climate data. This research initiates the update to the existing Canadian FS map by expanding the suite of input features used and comparing the impact of three feature selection methods (partial correlation, partial mutual information, combined neural pathway strength) on three types of ML algorithms: random forest, artificial neural network (ANN) and convoluted neural network (CNN). The expanded set of features includes geospatial indices and flood-specific meteorological data such as spring temperature, precipitation, and vapour pressure. Data from preceding seasons to specific flood events is also included. Preliminary findings from the feature selection methods show that including seasonal flood-specific meteorological data provides important information leading to better model performance. Model performances of the three algorithms were comparable. Random forest with extreme gradient boosting led to the highest model performance (AUC = 0.98, F1 = 0.94), followed by CNN (AUC = 0.0.96, F1 = 0.90). ANN ensemble with leave-one-out-cross-validation resulted in the lowest model performance (AUC = 0.91, F1 = 0.85). Results contribute to the development of an improved national FS map for Canada.

How to cite: Dunbar, K. E., McGrath, H., and Khan, U. T.: Enhancing flood susceptibility modelling in Canada: Integrating seasonal meteorological data, feature selection and machine learning approaches, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3871, https://doi.org/10.5194/egusphere-egu25-3871, 2025.

Effective flood forecasting is critical for informed decision-making and timely emergency response. Traditional physical models, which rely on fixed-resolution spatial grids and input parameters, often incur substantial computational costs, limiting their capacity to accurately predict flood peaks and provide prompt hazard warnings.  This paper introduces methods to ensure physical consistency in machine learning models, aiming to develop a fast, stable, accurate, cross-regional, and downscaled neural flood forecasting foundation model. Specifically, we present a Physics-embedded Neural Network, which integrates the momentum and mass conservations of flood dynamics into a neural network. Additionally, we combine this Physics-embedded Neural Network with a diffusion-based generative model, enhancing physical process consistency for long-term, large-scale flood forecasting. We also briefly introduce other models that integrate physics and machine learning, such as the FloodCast model by incorporating hydrodynamic equations into its loss function to maintain physical consistency, and the UrbanFloodCast model by learning physical consistency from urban flood dynamic data. The performance of these models will be analyzed using our proposed FloodCastBench dataset, a comprehensive collection of low-fidelity and high-fidelity flood forecasting dataset and benchmark. Results from the dataset demonstrate that incorporating physical consistency significantly enhances flood forecasting accuracy, demystifies the black-box nature of machine learning frameworks, and increases confidence in addressing dynamical systems. Finally, we propose a Spatiotemporal Foundation Model capable of forecasting floods across a variety of scales and regions.

How to cite: Xu, Q. and Zhu, X. X.: Towards Physics-consistent Foundation Models for Flood Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4428, https://doi.org/10.5194/egusphere-egu25-4428, 2025.

Catchments are the fundamental units of hydrological analysis and integrate a vast number of physical, biological, and anthropogenic processes. Traditional hydrological modelling approaches, however, adopt a bottom-up perspective, aggregating small-scale physical principles to predict large-scale catchment behaviour. While effective for prediction, this approach can fall short in advancing our understanding of emergent processes and their interactions given the strong dependence on a priori assumptions. To address this gap, causal discovery algorithms offer a promising alternative by moving beyond simple correlation to directly identifying the dynamic causal structures emerging at the catchment scale. In this study, we applied the PCMCI+ algorithm to the CAMELS-US dataset in combination with a subsequent causal effect estimation. We explored how and to what extent dynamic causal structures can be learned from hydro-meteorological data alone, and which catchment properties and conditions influence their expression. We find that causal discovery in hydrology faces challenges due to non-stationarity, unsuitable conditional independence tests, and unmet methodological assumptions. Despite these limitations, our approach reconstructed physically plausible relationships controlled by meaningful catchment properties. These results highlight the potential of causal discovery in hydrology, where it could serve as a complementary framework for model evaluation studies or as an integral part of the model development process.

How to cite: Strahl, D., Gnann, S., Wiesner, K., and Wagener, T.: Using Causal Discovery to Identify Drivers and Controls of Streamflow in Large Sample Hydrology, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4845, https://doi.org/10.5194/egusphere-egu25-4845, 2025.

Water resources management depends heavily on hydrological modeling for reservoir operation and risk mitigation, especially in data-scarce regions. Hybrid approaches that combine artificial intelligence and conceptual models offer great potential for accurate streamflow prediction. However, their implementation can be time-consuming and applied in different configurations. This study comprehensively compares two promising hybrid frameworks: the Conceptual-Data-Driven Approach (CDDA) and the Ensemble Approach. The analysis was conducted in the Upper Blue Nile Basin in Ethiopia over the period from 2002 to 2019. Six baseline models were developed, including CNN-LSTM (data-driven), NAM and HBV-Light (lumped), and SWAT+, WEAP, and HEC-HMS (semi-distributed). All models achieved NSE ≥ 0.85 during the validation period, with CNN-LSTM performing best (NSE = 0.94). Each model was integrated into the two hybrid frameworks using Random Forest (RF) or Artificial Neural Networks (ANN). Results showed that the Ensemble Approach outperformed CDDA by combining two conceptual models. ANN performed better than RF across both frameworks. Hybrid modeling significantly improved semi-distributed models, while lumped and data-driven models showed minimal benefits. In the Ensemble Approach, normal and extreme flows simulated using semi-distributed models performed best when supported by CNN-LSTM or lumped models. Our analysis also demonstrated the robustness of the Ensemble Approach for selecting the supporting model. These findings emphasize the value and feasibility of the Ensemble Approach for improving streamflow prediction and better supporting decision-making in data-scarce regions. Nevertheless, a thorough understanding of the opportunities in hybrid modeling requires further research with a specific focus on operational forecasting.

How to cite: Mohamed, A., M. Ali, A., Ali, A., Hassan, O., E. Elbasheer, M., and Abdelaziz, M.: Can CNN-LSTM and lumped models improve (extreme) streamflow prediction of semi-distributed models? A comparative analysis of two hybrid frameworks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5052, https://doi.org/10.5194/egusphere-egu25-5052, 2025.

Model Intercomparison Projects in the Earth Sciences have shown, that the outputs of
Earth System Models often show large variations and can therefore give quite different results,
with no single model consistently outperforming others. Examples include Global Water
Models (GWMs), as well as Global Climate Models (GCMs). The high computational costs
of running such models make comprehensive statistical analyses challenging, a common issue
with many complex models today. Machine learning models have become popular surrogates
of slow process-based models, due to their computational speed, at least once trained. This
speed makes it possible to use techniques from Explainable AI (XAI) to analyze the behavior
of the surrogate model.
Here, we analyze long-term averages of the GWM ’Community Water Model’ (CWatM)
for different parts of the global domain for actual evapotranspiration E_a, total runoff Q and
groundwater recharge R. We train an artificial neural network on the model’s input and output
data and use three different strategies to assess the importance of input data: LassoNet for sub-
set selection and feature ranking, along with Sobol’ indices and DeepSHAP for interpretability.
Our results show that subset selection can effectively reduce model complexity before XAI
analysis. For some hydrological domains the number of relevant input
variables for a chosen output reduces to less then 15 variables out of 98 model inputs, while
others remain more complex, requiring many variables for performances with R² > 0, 8.

How to cite: Faber, L., Wiesner, K., Tang, T., Wada, Y., and Wagener, T.: Using Explainable Artifical Intelligence (XAI) to Analyze the Behavior of Global Water Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5727, https://doi.org/10.5194/egusphere-egu25-5727, 2025.

Physics-based hydrological modelling provides great opportunities for risk assessment and water resources management. However, diagnostic model evaluation and quantitative uncertainty assessment remain a challenge: (1) Model choices, boundary conditions, and prior assumptions about input, parameter or data uncertainty might be hard to formulate or justify; (2) rigorous propagation of uncertainties struggles when the analysed model structure is not “true”, and (3) a full propagation of uncertainties is often computationally prohibitive for complex models.

Alternative approaches promote the extraction of information directly from data, thereby avoiding overly strict physics-based constraints and the pitfalls of uncertainty quantification. Challenges of these data-driven approaches include the lack (or difficulty of) explainability, transparency, and transferability to unseen scenarios.

To explore the frontier of where those two perspectives (should) converge, we investigate the potential of surrogate models (computationally cheaper, data-driven representations of complex models) as a binding link with several potential benefits: (1) they alleviate the computational burden and thereby allow for a fully Bayesian uncertainty analysis; (2) they are flexible enough to overcome structural deficits of the original complex model, thereby enabling a better predictive performance, and (3) being data-driven, we can elegantly fuse the information from available data into their training process.

Methodologically, we propose a weighted data-integrated training of surrogates via two competing approaches that differ technically, but also philosophically, and reveal complementing insights about the strengths and weaknesses of the physics-based model and about the additional information in the available data, thereby facilitating deeper system understanding and improved (hybrid) modelling. We demonstrate the proposed workflow on didactic examples and a real-world case study. We expect this approach to be generally useful for modelling dynamic systems, as it contributes to more realistic uncertainty assessment and opens up ways for model development.  

How to cite: Guthke, A., Reiser, P. L., and Bürkner, P.: Training Surrogates with Knowledge and Data: A Bayesian Hybrid Modelling Strategy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6278, https://doi.org/10.5194/egusphere-egu25-6278, 2025.

Neural networks are efficient and effective in predicting system states in hydrology. However, most current approaches lack hydrological flow partitioning, do not allow for training on measurements of multiple variables, or lack capability to tightly integrate physically-based components. To address these shortcomings I propose and evaluate an approach referred to below as Dynamical System Neural Network (DSNN). DSNN is a feedforward neural network with an architecture that resembles the organisation in components of the real-world system it represents. In hydrology, the DSNN represents each water flow (e.g. seepage, snow melt) by a collection of input, hidden, and output neural layers, where each input is the state of a hydrological storage (e.g. groundwater storage influencing seepage) or other variable (e.g. air temperature influencing snow melt). These components are interconnected to form a single neural network of the complete dynamical system considered, where all storages and flows are explicitly quantified. If physical understanding of a flow and its parameterization is available, a known formulation can be used as a replacement of a neural network component. The DSNN is applied forward in time, backpropagating gradients over all timesteps. It can be run in spatially lumped or semi-distributed mode. To demonstrate the approach, a DSNN is presented of the Austrian Dorfertal (Kals) Alpine catchment containing snow and subsurface water storages and associated flows including streamflow. The DSNN is trained, validated, and tested on daily streamflow over ~40 years. To explore the capability of the DSNN in estimating the magnitude and dynamics of internal system storages (snow water equivalent, subsurface water storage) and flows (evapotranspiration, sublimation, snowmelt, seepage), the DSNN is first trained and tested with streamflow data generated by a conceptual model. The DSNN turns out to be capable of reproducing - with a satisfactory level of precision - the system states and fluxes calculated by the conceptual model, with decreasing performance when measurement error is added to the artificially generated streamflow data before training. To explore its predictive performance, the DSNN is applied on measured streamflow data for the Dorfertal, comparing multiple DSNN setups that represent all flows as neural network components or only a subset of flows where remaining flows are represented with a standard conceptual model (e.g. linear reservoir). Preliminary results indicate that in predictive performance, in most setups, the DSNN outperforms a standard conceptual model trained on the same streamflow data, with NSE values for testing of 0.74 and 0.71, respectively. This preliminary result indicates DSNN to be a promising approach for blending process-based and neural network based modelling as well as for training (i.e. calibration) of neural network models on measurements of multiple hydrological variables as these are all explicitly represented by the DSNN and can thus be incorporated in the loss function (e.g. streamflow, snow depth, groundwater, evapotranspiration).

How to cite: Karssenberg, D.: Dynamical system neural network for hydrological modelling, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7382, https://doi.org/10.5194/egusphere-egu25-7382, 2025.

How to cite: Zheng, Y., Wang, C., and Jiang, S.: Advancing distributed hydrological modeling with hybrid machinelearning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7770, https://doi.org/10.5194/egusphere-egu25-7770, 2025.

The integration of remote sensing observations into hydrological modeling frameworks presents a significant opportunity for improving spatial and temporal predictive capabilities across continental domains. This research introduces a novel hybrid distributed hydrological model that addresses key challenges in computational efficiency, by using a GPU-enabled computational infrastructure, and in predictive accuracy by assimilating multi-source remote sensing datasets, specifically satellite-based soil moisture and evapotranspiration, at a high spatial resolution (1km×1km) and temporal scale (hourly). The model addresses critical challenges in regional hydrological forecasting by leveraging advanced data assimilation techniques and machine learning methodologies.

The proposed hybrid modeling framework synthesizes physically-based distributed hydrologic modeling principles with data-driven machine learning approaches, facilitating a more comprehensive representation of land surface hydrological processes. A key innovation is the GPU-enabled cell-to-cell routing algorithm, which enables fast and efficient computational processing of complex hydrological connectivity and water movement across large spatial domains. By integrating remote sensing observations, the methodology enables enhanced initial condition specification and improved parameter estimation, particularly in regions characterized by sparse ground-based measurement networks.

Preliminary analytical results demonstrate significant improvements in model performance, particularly in capturing spatial and temporal variability of hydrological states and fluxes. The approach substantively advances current methodological capabilities in hydrological forecasting, offering a promising framework for developping enriched tensorial numerical solvers, addressing complex hydroclimatic prediction challenges in data-limited environments.

How to cite: Ettalbi, M., Garambois, P.-A., Huynh, N.-N.-T., Ferreira, E., and Baghdadi, N.: GPU-Enabled Cell-to-Cell Routing in a High Resolution Hybrid Distributed Hydrological Model with Multi-Source Remote Sensing Data Assimilation: A Continental-Scale Computational Approach , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9176, https://doi.org/10.5194/egusphere-egu25-9176, 2025.

The formation of floods, as a complex physical process, exhibits dynamic changes in its driving factors over time and space under climate change. Due to the black-box nature of deep learning, its use alone does not enhance understanding of hydrological processes. The challenge lies in employing deep learning to uncover new knowledge on flood formation mechanism. This study proposes an interpretable framework for deep learning flood modeling that employs interpretability techniques to elucidate the inner workings of a peak-sensitive Informer, revealing the dynamic response of floods to driving factors in 482 watersheds across the United States. Accurate simulation is a prerequisite for interpretability techniques to provide reliable information. The study reveals that comparing the Informer with Transformer and LSTM, the former showed superior performance in peak flood simulation (NSE over 0.6 in 70% of watersheds). By interpreting Informer’s decision-making process, three primary flood-inducing patterns were identified: precipitation, excess soil water, and snowmelt. The controlling effect of dominant factors is regional, and their impact on floods in time steps shows significant differences, challenging the traditional understanding that variables closer to the timing of flood event occurrence have a greater impact. Over 40% of watersheds exhibited shifts in dominant driving factors between 1981-2020, with precipitation-dominated watersheds undergoing more significant changes, corroborating climate change responses. Additionally, the study unveils the interplay and dynamic shifts among variables. These findings suggest that interpretable deep learning, through reverse deduction, transforms data-driven models from merely fitting nonlinear relationships to effective tools for enhancing understanding of hydrological characteristics.

How to cite: Xu, Y. and Lin, K.: Uncovering the Dynamic Drivers of Floods through Interpretable Deep Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10256, https://doi.org/10.5194/egusphere-egu25-10256, 2025.

Earth system data, measured by satellites and terrestrial stations and simulated by increasingly complex models, provide valuable information for identifying functional relationships within the Earth system. These relationships are essential for understanding complex interactions and predicting changes, for example, in climatic or ecological processes, but often only occur in certain spatiotemporal sections or within certain threshold values. With the increasing spatiotemporal resolution of remote sensing products and models, a manual analysis is impractical, and hypothesis-driven approaches can lead to undiscovered hidden relationships. Previous work proposed the SONAR (automated diScovery Of fuNctionAl Relationships) decision-tree algorithm to automatically search for functional relationships in earth system data without a-priori assumptions. We analyzed the proposed algorithm using artificially generated data to evaluate SONAR's functionality.  We tested if the choice of statistical indicator (Pearson’s r, Spearman’s ρ, Kendall’s τ, and Mutual Information) influences the functionality of the SONAR algorithm and which factors are important for the identification of functional relationships. Using 1512 synthetic data sets and the developed SAMPI (Similarity of A Manifested and Prototypical decision tree Indicator) coefficient, we demonstrate how the performance of the algorithm changes under different variations of the data sets - including the number of designated splits, the presence of interfering variables and the strength and nature of the underlying functional relationships. In particular, we show which statistical indicator provides the best results under these conditions. The results demonstrate that the SONAR algorithm is very versatile, especially when employing the most reliable statistical indicator. The SONAR algorithm could, therefore, have far-reaching applications, for example, in analyzing climatic patterns or investigating dependencies between environmental factors.

How to cite: Thöne, C., Bäthge, A., and Reinecke, R.: The effects of different statistical indicators in the new decision-tree-based SONAR algorithm for automated detection of functional relationships in Big Earth Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10444, https://doi.org/10.5194/egusphere-egu25-10444, 2025.

Terrestrial evaporation (E) is an essential climate variable, linking water, energy and carbon cycles. As E is influenced by the state of the atmospheric boundary layer, vegetation and soil, its modelling is a complex task, resulting in a myriad of simulation approaches. To combine the strong predictive skills of data-driven models with the interpretability and physical consistency of process-based models (PBMs), a new research field of differentiable modelling has emerged¹. 

Here, we present a differentiable framework for E estimation, facilitating online training of NNs as intermediate PBM components. It is inspired by the GLEAM framework for estimating E, which applies offline training (i.e., outside the PBM) of neural networks (NNs) predicting evaporative stress^2,3. Building upon the Julia SciML ecosystem’s implementation of universal differential equations⁴, a wide array of numerical methods are available for solving the PBM’s ordinary differential equations (ODEs) and calculating the parameter sensitivities⁵. In this way, the effect of the numerical methods on the obtained hybrid model can be investigated, moving beyond the direct automatic differentiation through explicit Euler solutions of ODEs as often applied in other hydrological hybrid modelling approaches. 

References

¹Shen, C., Appling, A.P., Gentine, P. et al., Differentiable modelling to unify machine learning and physical models for geosciences, Nat. Rev. Earth. Environ., 4, 552–567, 2023, https://doi.org/10.1038/s43017-023-00450-9

²Koppa, A., Rains, D., Hulsman, P. et al., A deep learning-based hybrid model of global terrestrial evaporation, Nat. Commun., 13, 1912, 2022, https://doi.org/10.1038/s41467-022-29543-7

³Miralles, D. G., Bonte, O., Koppa, A. et al., GLEAM4: global land evaporation dataset at 0.1° resolution from 1980 to near present, preprint, 2024, https://doi.org/10.21203/rs.3.rs-5488631/v1 

⁴Rackauckas, C., Ma, Y.,  Martensen, J. et al., Universal differential equations for scientific machine learning, ArXiv, 2020, https://doi.org/10.48550/arXiv.2001.04385 

⁵Sapienza, F., Bolibar, J., Schäfer, F. et al., Differentiable Programming for Differential Equations: A Review, ArXiv, 2024, https://doi.org/10.48550/arXiv.2406.09699 

How to cite: Bonte, O., Miralles, D. G., Koppa, A., and Verhoest, N. E. C.: Universal differential equations for estimating terrestrial evaporation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11693, https://doi.org/10.5194/egusphere-egu25-11693, 2025.

How to cite: Chang, A. Y.-Y., Leonarduzzi, E., Grams, C. M., and Humphrey, V. W.: A Physics-Constrained Emulator for High-Resolution Soil Moisture , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12068, https://doi.org/10.5194/egusphere-egu25-12068, 2025.

Terrestrial evaporation (E) is a critical component of the water cycle, returning nearly 60% of continental precipitation to the atmosphere and dissipating approximately 50% of surface net radiation. A prevalent approach for estimating E involves computing a theoretical maximum, known as potential evaporation (E_p), and scaling it based on a multiplicative stress factor, often referred to as “evaporative stress” (S) or “transpiration stress” (S_t) when specifically applied to plant transpiration. Like stomatal or surface conductance, S_t is governed by a complex nonlinear interplay of environmental drivers such as soil moisture, air temperature, radiation, and atmospheric vapor pressure deficit. This complexity is not yet fully understood, which further hampers its accurate physical modelling and limits our ability to comprehend transpiration’s sensitivity to the changing environment.

The fourth generation of the Global Land Evaporation Amsterdam Model (GLEAM4) has yielded a global dataset of transpiration by integrating multi-source remote sensing data following a hybrid approach, in which E_p is computed based on a process-based model and S_t is calculated by employing deep neural networks. These neural networks are trained on global eddy covariance and sap flow measurements for both tall and short vegetation, and are informed by a set of environmental controls or biotic factors. These factors include soil moisture, vapor pressure deficit, atmospheric CO₂ concentration, wind velocity, air temperature, downwelling shortwave radiation, LAI, and vegetation optical depth. Beyond the predictive capabilities of these deep neural networks, the relationships between environmental controls and S_t within these neural networks remain under exploration, leaving uncertainty as to whether GLEAM4 accurately represents real-world processes. To explore the relationships, we employ the SHapley Additive exPlanation (SHAP) method, which quantifies the marginal contributions of predictors to model predictions, offering insights into the relative importance of environmental drivers in determining S_t.

Our findings highlight dominant S_t drivers across various climatic regimes and ecosystems, revealing their contributions' temporal evolution. Additionally, we investigate how S_t responds to shifts in environmental conditions, including climate and vegetation changes, water stress, atmospheric aridity, and rising CO₂ levels. Our study enhances global understanding of transpiration dynamics and provides critical insights into the impacts of diverse hydroclimatic drivers, thereby supporting broader applications within the hydrology and climate communities.

How to cite: Ruan, F., M. Baez-Villanueva, O., Bonte, O., Koppa, A., Li, W., Camps Valls, G., Yang, Y., and G. Miralles, D.: Global Vegetation Stress Drivers based on Hybrid Modelling and Explainable AI, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12928, https://doi.org/10.5194/egusphere-egu25-12928, 2025.

Vegetation plays an important but complicated role in modulating land-atmosphere interactions and the water cycle. Under global change, increasing vegetation greenness trends have been observed, which further complicate the control of vegetation in the earth system. Despite growing interest in the role of vegetation in the hydrological processes, large uncertainties still exist, particularly when it comes to the underexplored response of streamflow to vegetation greening. In this study, we explore the watershed-relevant biophysical controls of vegetation greening on streamflow. In order to do so, we develop a hybrid ecohydrological model. This model adheres to the water balance principles, while it simultaneously has a flexible structure that enables integrating physical insights from observational data. The multi-task learning optimization ensures physical consistency across a range of processes and temporal frequencies, which allows us to investigate the cascading impacts of vegetation changes across the water cycle, leading up to the streamflow as an end-process. Ecohydrological insights are directly derived from observational data, while physically meaningful model parameters reflect how ecosystem functions and hydrological processes respond to vegetation changes. We find that the marked change in streamflow can be attributed to vegetation change controls on diverse biophysical processes. Our research highlights the potential of hybrid models to capture complex earth system processes by exploiting multiple observational data streams, machine learning and physical constraints.

How to cite: Blougouras, G., Brenning, A., Migliavacca, M., and Reichstein, M.: Hybrid hydrological modelling of the biophysical impacts of earth’s greening on streamflow, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13311, https://doi.org/10.5194/egusphere-egu25-13311, 2025.

Cyanobacterial blooms have become more frequent and intense in Lake Superior since 2012, primarily due to increased nutrient loads, with phosphorus being the main limiting factor. To protect water quality, extensive monitoring of lakes and streams is crucial, but it is not cost-effective or practical to measure nutrients frequently across all ecosystems. This study presents a cost-effective, transferable solution using machine learning (ML) models to predict phosphorus concentrations and loads based on conventional water quality parameters like streamflow, dissolved oxygen, conductivity, turbidity, transparency, and total suspended solids. The research introduces an explainable hybrid ML framework combining probabilistic principal component analysis (P2CA) with several ML models, including Bagging Ensemble Learning, Boosting Ensemble Learning, Gaussian Process Regression, and Support Vector Regression, to enhance prediction accuracy. Results demonstrate that the P2CA-Boosting Ensemble Learning model consistently outperforms other approaches. To confirm its effectiveness, the developed model was tested with the same input data from a different river catchment, proving it works well in different environments. This study highlights the potential of combining P2CA with Boosting Ensemble Learning as a powerful tool for water quality management in streams and rivers.

How to cite: Kumar, A. and Zhang, K.: Development of a hybrid machine learning model to predict total phosphorus in streams over the north shore of Lake Superior, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13799, https://doi.org/10.5194/egusphere-egu25-13799, 2025.