This session invites contributions that harness the power of artificial intelligence (AI), Earth observation (EO), and integrated geospatial infrastructures to address local and global environmental challenges. We focus on the practical application of time-series aerial and satellite remote sensing data, combined with advanced geospatial technologies, to monitor, model, and mitigate impacts related to climate change, natural hazards, and resource management.
We welcome submissions from applied and theoretical domains that emphasize:
• AI and machine learning for land cover change, biomass estimation, and hazard mapping
• Time-series analysis of multi-modal data (optical, SAR, hyperspectral, in-situ)
• Scalable, open-source, and cloud-native geospatial workflows
• 3D geological modeling and dynamic image management
• Integrated geospatial infrastructures for collaborative science and decision-making
Submissions should highlight innovative methodologies and real-world applications using analysis-ready data across the electromagnetic spectrum. The session aims to foster interdisciplinary collaboration and showcase how geospatial technologies can support science-based strategies for resilience and sustainability.
Arctic coastlines are highly sensitive to climate change, resulting in significant environmental and socioeconomic impacts. Monitoring changes in these regions is essential for understanding long-term dynamics, identifying the most vulnerable coastlines, and providing critical information to local and international stakeholders for effective climate adaptation strategies. Geoinformation technologies, particularly Earth Observation (EO) and Geographic Information Systems (GIS), have emerged as transformative solutions for tracking coastal changes across remote and data-scarce Arctic areas.
To this end, this study aims to exploit EO and GIS for developing a geospatial database that examines how coastlines in the Svalbard Archipelago and Iceland have evolved over the last four decades (1985–2025). Anniversary dates of Landsat satellite imagery were used, while all necessary preprocessing and imagery acquisition were performed in Google Earth Engine cloud platform. Coastline digitization was carried out in ArcGIS Pro using the direct photointerpretation method. In Svalbard, shoreline changes are mainly linked to ice-related processes, such as variations in ice cover. In Iceland, coastline variability reflects a combination of geomorphological processes and localized anthropogenic activity, particularly in areas affected by port development
Τhe database developed herein supports the continued monitoring of Arctic coastal environments and offer a foundation for further investigation of coastal change in both natural and human-influenced parameters. The produced geospatial datasets provide a baseline for future analyses contributes to relevant efforts for developing open access geospatial datasets that such as the EO-PERSIST platform. Such endeavors provide fertile ground to researchers and policymakers to better understand coastal dynamics and support evidence-based decision-making for Arctic coastal management.
Keywords: Arctic coastlines; Shoreline change; Earth Observation; Landsat; Geoinformatics; EO-PERSIST
Acknowledgement
The present research study is supported by the project “EO-PERSIST”, funded by the European Union’s Horizon Europe research and innovation program (HORIZON-MSCA-2021-SE-01-01, under grant agreement no. 101086386
How to cite: Kapsokefalou, S., Gartagani, P., Plessas, E., Nakos, C., Kafouris, I., Tselos, N. G., Detsikas, S. E., Litke, A., and Petropoulos, G. P.: A Geospatial Database for Monitoring Arctic Coastline Dynamics: Mapping Shoreline Change in Iceland and Svalbard, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1692, https://doi.org/10.5194/egusphere-egu26-1692, 2026.
Climate change causes significant glacial retreat in the Austrian Alps. Glacial retreat is linked to an increase in the number and size of glacial lakes. The emergence and growth of glacial lakes threatens alpine infrastructure and can cause glacial lake outburst floods (GLOF's). Therefore, it is important to monitor the spatio-temporal evolvement of glacial lakes. Remote sensing provides possibilities for cost-effective glacial lake monitoring. Besides, in recent years various deep learning-based models have been introduced as effective tools for computer vision tasks, including semantic segmentation.
In this study, a Unet-based semantic segmentation model architecture has been implemented, trained and assessed on a custom training dataset.
The dataset is based on an inventory of Austrian glacial lakes in 2015 and Sentinel-2 imagery and contains 386 image chips, 270 for training and 116 for testing, each measuring 512 by 512 pixels and including at least one glacial lake from the inventory.
The semantic segmentation model has been applied to a time series of Sentinel 2 imagery from 2015 to 2025 in order to create a time series of glacial lake maps. The final results will be used to examine whether the number of glacial lakes has increased in recent years and to examine spatio-temporal trends in glacial lake evolution. Potential impacts of the observed developments on (hiking) infrastructure will also be assessed.
The geospatial workflow is implemented using open-source tools and freely available datasets.
How to cite: van Ek, J.-J. and Hölbling, D.: Deep Learning Based Glacial Lake Mapping in the Austrian Alps, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6483, https://doi.org/10.5194/egusphere-egu26-6483, 2026.
Embeddings provide a compact representation of data in a lower-dimensional vector space, enabling faster and more efficient analysis compared to direct processing of high-dimensional data. Satellite imagery is an example of such data, as it is characterized by large volume and high dimensionality. With the rapid development of AI foundation models, embedding-based approaches can increasingly replace classical remote sensing techniques in tasks such as classification and regression, while maintaining or even improving the quality of results.
This work leverages the Global Embeddings Dataset from the Copernicus Data Space Ecosystem, which contains embeddings generated by multiple models, including SSL4EO DINOv2, SigLIP, DeCUR, and MMEarth. These models differ in sensing modality, input resolution, and embedding dimensionality, enabling diverse analyses based on heterogeneous data sources. Data standardization using the MajorTOM format facilitates automated processing and seamless integration of embeddings derived from different models.
Following the MajorTOM standard, more than 8 million images, comprising 9.368 trillion pixels of raw data, were processed to generate over 170 million embeddings from approximately 62 terabytes of satellite data. This scale demonstrates the feasibility of embedding-based approaches for efficient management and analysis of large-scale Earth observation datasets.
Embedding-based representations enable effective detection of environmental changes, which can be categorized as either abrupt events, such as wildfires, deforestation, or floods, or long-term processes, including river desiccation and gradual ecosystem degradation. Such change detection capabilities are applicable across multiple domains, including urban development, defense, and environmental monitoring. By operating on compressed representations, embeddings allow for efficient similarity and change analysis over temporal sequences, significantly accelerating the processing of large satellite data archives.
How to cite: Ostrowski, B., Bojanowski, J. S., Kluczek, M., and Czerkawski, M.: AI Foundation Models for Near Real-Time Environmental Monitoring from Satellite Data , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9658, https://doi.org/10.5194/egusphere-egu26-9658, 2026.
River organic pollution exhibits pronounced spatiotemporal dynamics in response to environmental changes. However, the traditional method of tracking chemical oxygen demand (COD) and/or other organic pollution indicators at fixed locations over expansive regions is labor-intensive, time-consuming, and inadequate for achieving full spatial coverage. To address this limitation, here we developed a Random Forest algorithm using Landsat satellite data in conjunction with sub-daily (every 4 hours) COD data at 1,997 sites across China. The proposed model achieved high accuracy, with a root mean square error of 0.52 mg/L and a mean absolute percent difference of 13.01%. Additionally, the model was robust across clear, algae-laden, turbid, and black-smelling waters. Then, the algorithm was applied to investigate the spatiotemporal variations of COD concentration in Chinese rivers during 1984-2023. Across China, high river COD concentrations were observed in the eastern Songliao (3.56 ± 1.11 mg/L), Haihe (3.00 ± 0.89 mg/L), and Huaihe (3.57 ± 0.67 mg/L) basins. Anthropogenic activities could explain 79.39% of the spatial variability in COD concentrations, and the cropland distribution had a significant impact. During 1984-2023, 73.58% of China's rivers exhibited significant changes in COD concentrations (p < 0.05). With respect to the 800 mm isoprecipitation line, 56.62% of the southeastern rivers showed decreasing trends; in contrast, 84.25% of the northwestern rivers displayed increasing trends in COD concentrations. The temporal variations in COD concentrations were driven by the combined effects of factors including rainfall, vegetation coverage, and human activities; their relative contributions were 0.02 – 42.45%, 0.07 – 68.76%, and 0.06 – 90.31% for COD changes in different provinces. This study underscores the advantages of using satellite data to efficiently and dynamically monitor organic pollution in river systems, providing crucial technical and data support for such monitoring efforts on a large scale.
How to cite: Liu, D., Yan, N., Qiu, Z., Zhang, C., and Yan, Y.: Landsat monitoring reveals the history of river organic pollution across China during 1984-2023, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4239, https://doi.org/10.5194/egusphere-egu26-4239, 2026.
Semi-enclosed seas are particularly sensitive to regional climate forcing because their exchange with the open ocean is limited. Despite this sensitivity, the magnitude of their warming relative to global climate model projections remains insufficiently constrained. This study examines both historical and future sea surface temperature (SST) changes in the Mediterranean, Aegean, Marmara, and Black Seas surrounding Türkiye using a hybrid framework that integrates in-situ observations, satellite reanalysis, and machine learning techniques.
More than three decades (1993–2023) of monthly coastal SST records from 21 stations are analysed together with Copernicus Marine Environment Monitoring Service (CMEMS) reanalysis data. Linear trend analysis reveals statistically significant warming across all basins, with observed SST increases reaching approximately 2.0 °C since 1993. When placed in a global context, these regional warming rates are approximately 1.5–2 times higher than the CMIP6 ensemble mean global ocean warming over a comparable period.
Future SST evolution is explored using a Long Short-Term Memory (LSTM) model trained on bias-corrected SST time series and supplemented with large-scale climate indices, namely ENSO and NAO. Comparison with observations shows that the model reproduces SST variability with a high level of agreement (R² > 0.9; RMSE ≈ 0.4–0.6 °C), while the projected trajectories remain physically plausible under both SSP2-4.5 and SSP5-8.5 scenarios. The projections point to a persistent warming signal throughout the 21st century, with the strongest increases concentrated in the Mediterranean and Aegean Seas.
Taken together, these results suggest that SST warming in Türkiye’s semi-enclosed seas is higher than coarse-resolution CMIP6 ensemble mean global ocean warming estimates. This finding emphasises the importance of regionally resolved observations and data-driven analyses for coastal climate assessment. The hybrid framework applied here offers a scalable approach for monitoring semi-enclosed marine systems and for informing climate-related decision-making at regional scales.
How to cite: Erkoç, M. H.: Are Türkiye’s Semi-Enclosed Seas Warming 1.5–2 Times Faster Than CMIP6 Global Projections?, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14132, https://doi.org/10.5194/egusphere-egu26-14132, 2026.
Long-term tree-species information is fundamental for quantifying ecosystem services and forest climate resilience, yet multi-decadal mapping is often constrained by sparse field samples and heterogeneous satellite archives. Here we present a sensor-agnostic sample-transfer pipeline that reconstructs annual dominant tree-species distributions in the Qilian Mountains (north-eastern Tibetan Plateau) from a single-year field training dataset.
We harmonize optical missions (Landsat-5/7/8 and Sentinel-2) with multi-frequency SAR archives (ERS and Envisat C-band, Sentinel-1 C-band, and ALOS PALSAR L-band) and build a unified annual feature space combining spectral variables, SAR backscatter metrics, terrain predictors, and phenological descriptors. Phenology is derived from NDVI time series using Harmonic Analysis of Time Series (HANTS), yielding noise-robust seasonal metrics that remain comparable across sensors and years. To overcome the absence of historical labels, we transfer class labels from 6,268 field samples to each target year through a dual-constraint similarity screening: (i) feature-vector magnitude (Euclidean distance) and (ii) feature-vector direction (cosine similarity / spectral-angle-based measure). A thresholding rule discards ambiguous points and retains only reliable migrated samples. Annual maps are then generated using a Random Forest classifier (bagging and majority vote), while class imbalance is mitigated via downsampling and SMOTE.
Across nine sensor-integration periods spanning 1986–2024, the sample-transfer component remains stable despite changing sensors (transfer accuracy: 86.0–94.4%), and the resulting tree-species classification maintains consistently high accuracy (95.7–98.8%). Year-by-year assessments indicate overall accuracy and Kappa typically above 0.95 for most years and classes; performance reductions are mainly confined to rare taxa with limited observations. The final products provide consistent annual maps for six dominant tree genera (Betula, Juniperus, Picea, Populus, Pinus, and Larix) together with shrub–grass vegetation, cropland, water, and bare land, enabling robust quantification of multi-decadal changes in area fractions, spatial patterns, and centroid migration.
By coupling multi-sensor feature harmonization, HANTS-based phenology, and a dual-constraint sample-transfer strategy, this workflow offers a practical and generalizable route to recover multi-decadal tree-species dynamics from limited field data in mountain ecological barrier regions.
How to cite: Yu, J., Li, Y., Zhang, F., Wang, C., Wang, Z., and Gou, X.: Reconstructing Annual Dominant Tree-Species Distributions (1986–2024) in the Qilian Mountains via Multi-sensor Sample Transfer and Random Forest, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7381, https://doi.org/10.5194/egusphere-egu26-7381, 2026.
Bees for Development (BfD) and its local partners have supported beekeeping initiatives in the Kwahu Afram Plains, Ghana, since 2019, with the objective of promoting forest conservation and empowering beekeepers to reduce drivers of forest loss while enhancing forest recovery. Evaluating the conservation results of such community-based interventions requires independent, spatially explicit evidence of environmental outcomes. Key indicators of success include the absence of tree loss and burned areas within a 250 m radius of established apiary sites.
While satellite-based change detection of tree cover and burned areas is well established at regional and global scales, fine-scale monitoring at localized, spatially distributed sites remains relatively understudied and methodologically challenging. Such analyses require careful calibration and validation to detect subtle changes at small spatial extents. In this study, we assess the performance of multiple change detection algorithms for monitoring tree cover loss and fire disturbance within a 500 m diameter surrounding apiary locations, with a focus on fine-scale detection.
We integrate multi-sensor satellite data from Landsat, Sentinel-1, and Sentinel-2, and apply statistical time-series approaches including AVOCADO and BEAST. Algorithm performance is evaluated using high-resolution reference data from PNEO, WorldView, and Google Earth imagery, complemented by field-based ground observations. The study investigates different change detection methods, especially for localized, fine-scale conservation impact assessment and provides insight into practical and independent monitoring frameworks for community-led conservation initiatives.
How to cite: Hwang, B., Keywood, C., and Lowore, J.: Change detection of tree cover and burned areas at apiary sites in the Kwahu Afram Plains, Ghana, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18379, https://doi.org/10.5194/egusphere-egu26-18379, 2026.
The upper Yellow River is a nationally important water-conservation region where forest three-dimensional structure supports carbon storage and hydrologic buffering. In this topographically complex forest–shrub–grass mosaic, watershed-scale constraints of forest structure by climate remain poorly quantified, and uncertainty in forest extent can propagate into subsequent structural mapping and attribution. We produced a GEDI (Global Ecosystem Dynamics Investigation) height-constrained hierarchical forest mask and generated 10–30 m forest cover with probability and quality-assurance layers for the upper Yellow River.
Using the Taohe River Basin as a representative catchment, UAV LiDAR benchmarks were upscaled with Sentinel-1/2 time-series metrics, topographic predictors and GEDI information to derive 30 m wall-to-wall canopy height, aboveground biomass (AGB) and canopy entropy, the latter representing vertical structural complexity. Independent evaluation indicates reliable performance (R² ≈ 0.71–0.84; canopy entropy R² = 0.836).
Forest structure–climate relationships were examined with a Potential–Realized framework. Climate-constrained structural potential was estimated using conditional upper-quantile models, and climatic limitation was quantified as the departure of realised structure from its potential. Hydrothermal thresholds are well defined: canopy height potential peaks at 2–4 °C mean annual temperature (≈ 25 m) and approaches saturation beyond 540–560 mm annual precipitation; the joint optimum for canopy height and AGB occurs under ~450–550 mm precipitation and 5–8 °C mean temperature; canopy entropy maximises near 2–4 °C, 450–550 mm, aridity index ≈ 1.2, and potential evapotranspiration ≈ 600 mm. Using downscaled CMIP6 projections and derived vapour pressure deficit (VPD), we quantify potential contraction and overshoot risk (current structure/future potential) and provide analysis-ready layers for climate-risk screening in water-source regions.
How to cite: Wang, C., Yu, J., Li, Y., Zhang, F., Wang, Z., and Gou, X.: GEDI-constrained forest mapping and Potential–Realized climate limits of canopy height, biomass and structural complexity in the upper Yellow River, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7579, https://doi.org/10.5194/egusphere-egu26-7579, 2026.
Reliable feature identification, long time series of identified features, and tools to explore them provide substantial benefits for weather forecasting, process understanding, climate information provision, and the evaluation of climate model outputs. Moreover, expert use of these features within an established feedback loop enables the creation of high-quality training datasets for further application development and for training machine learning (ML) models.
EUMETSAT and its Member States are building a collaborative environment for joint manual or ML-assisted annotation, model development, and the database of identified features within the European Weather Cloud (EWC) to support these developments. The EWC is a cloud-based collaboration platform for meteorological application development and operations in Europe, and to enable the digital transformation of the European Meteorological Infrastructure.
EUMETSAT is compiling a database of long time series of meteorological features identified from various satellite datasets. This database will support the analysis of the development and interrelationships of these features, enabling new insights for both nowcasting and climate science. For example, the Deutscher Wetterdienst plans to use this environment to characterise convective storms using FCI-derived cloud-top features—such as overshooting tops—for nowcasting, with the primary aim of training an ML algorithm to automatically identify storm tops.
Initial work has begun on identifying tropical storms from long Himawari time series, alongside a feasibility study on additional feature types. Early results from the feasibility study, demonstrating the potential for performing feature identification on long time series of Earth-observation data, will be presented.
The joint working environment is available in the EWC, which is open to authorised users from ECMWF and EUMETSAT Member and Co-operating States for official duties and R&D projects. It consists of data-proximate cloud infrastructure, alongside the EWC Community Hub, which enables collaborative development, code and ML model sharing, and the exploitation of meteorological applications.
How to cite: Erdmann, A.-L., Tervo, R., Holl, G., Karatosun, A., Huckle, R., Schulz, J., Halbig, A., Hogervorst, F., and Brugaletta, L.: Identifying Earth System Features from Satellite Data , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19609, https://doi.org/10.5194/egusphere-egu26-19609, 2026.
