Sentinel-1 is a unique resource for global flood monitoring, providing systematic, weather-independent Synthetic Aperture Radar (SAR) imagery with unprecedented coverage. To overcome limitations of on-demand flood mapping services that depend on human operators to collect and interpret satellite images, a fundamentally new approach was adopted by the Global Flood Monitoring (GFM) service. This service, which was launched in 2021 as part of the Copernicus Emergency Management Service (CEMS), processes all Sentinel-1 land images acquired in VV polarisation fully automatically in near-real time. This article presents the first comprehensive analysis of GFM’s scientific achievements and challenges during its initial years of operation. To map floods reliably under diverse environmental conditions, GFM combines three complementary flood-mapping algorithms with reference water datasets to differentiate flooded areas from permanent and seasonal water bodies. The service also offers a novel flood-likelihood layer and contextual information to highlight areas where flood mapping is unreliable or not feasible. These data layers were derived from a global 20 m backscatter datacube containing approximately 379 billion land surface pixels. This datacube also made it possible to generate the first global Sentinel-1 flood archive (2015 to present). Our performance analysis shows that GFM typically delivers flood maps within five hours of image acquisition. However, a significant percentage of floods may go undetected due to coverage gaps. Initial evaluation results show that good accuracies are achieved for larger-scale floods and regions in the temperate and tropical zones, while accuracies are lower for smaller-scale floods and arid environments. The GFM service will continue to improve service quality by enhancing flood detection capabilities using improved algorithms and additional data, such as the VH channel from Sentinel-1 or L-band data from the upcoming ROSE-L mission.

Accurate flood mapping from synthetic aperture radar (SAR) imagery is essential for disaster response and risk management. Although deep learning (DL) models have advanced SAR-based flood mapping, they typically lack mechanisms to quantify and communicate predictive uncertainty, limiting their reliability in high-stakes applications. This study proposes a unified framework that captures and conveys two key sources of uncertainty in flood mapping: uncertainty arising from the model's limited knowledge, and uncertainty caused by noise and ambiguity inherent in the data. The former is estimated via feature density in the latent space of a density-aware neural network, while the latter is quantified using softmax entropy. These uncertainties are then communicated through conformal risk control under a user-defined risk level (alpha, delta): inputs associated with high model-knowledge uncertainty are flagged as out-of-distribution (OOD) and abstained from prediction, while data-related uncertainty guides the construction of set-valued predictions for in-distribution (ID) samples. Experiments on diverse real-world flood scenarios-including flooded built-up (FB) areas, flooded vegetation (FV), and flooded bare soil-demonstrate substantial gains in uncertainty quantification and predictive reliability. The proposed density-aware neural network achieves an average area under the receiver operating characteristic (AUROC) of 0.921 in OOD detection, outperforming Bayesian neural networks (BNNs) (0.684) and deep ensembles (DEs) (0.603). OOD abstention improves predictive safety, reducing the average water false negative rate from 36.9% to 28.5% under a risk level of (alpha = 0.05 and delta = 0.1), while prediction sets reduce average miscoverage from 4.4% to 1.6% compared to standard singleton predictions. By explicitly quantifying and communicating the uncertainty in DL predictions, this work strengthens trustworthy, risk-informed decision support for SAR-based flood mapping.

Purpose Crop growth models (CGM) are valuable tools for agricultural monitoring. However, the need for many input parameters, the uncertainties related to model parametrization and structure, and the lack of spatial information motivate the application of techniques such as data assimilation (DA). This paper proposes a DA framework to improve maize biomass estimation. Methods A particle filter (PF) was used to assimilate remotely sensed reflectance and soil moisture (SM) data, both independently and simultaneously, into the Agricultural Production Systems sIMulator (APSIM) model. Reflectance observations from Sentinel-2 were assimilated through coupling APSIM with the radiative transfer model (RTM) PROSAIL, while SMAP L-band SM products were directly assimilated into APSIM. Results The synthetic experiment, designed to evaluate the reliability of the proposed procedure, highlighted the strength of assimilating reflectance to constrain crop traits and of SM to reduce ensemble spread and improve robustness. Real-case results confirmed these findings. DA assimilation of SM especially contributed to improving overall biomass accuracy, particularly under data gaps and drought conditions. Although it did not consistently surpass single-source assimilation, the joint assimilation yielded consistent results. In 2022, it achieved a root-mean-square error (RMSE) of 2275.20 kg/ha, a normalized RMSE (nRMSE) of 44.99%, and a bias of 1081.90 kg/ha. In 2023, RMSE, nRMSE and bias were 1120.29 kg/ha, 14.79%, and 284.05 kg/ha, respectively. Furthermore, the joint assimilation led to a tighter ensemble spread than single source-assimilation. Conclusion The proposed framework demonstrates the potential of multi-source DA to enhance biomass estimation and support robust, spatially explicit crop monitoring.

Natural hazards cause hundreds of billions of dollars in losses annually, with impacts that disproportionately affect vulnerable populations. Earth observation (EO) provides comprehensive, repeatable measurements that are essential for assessing disaster impacts when ground-based systems fail. Over the past two decades, methods for post-disaster mapping impacts from EO data have evolved rapidly, from manual interpretation and spectral indices to deep learning and, most recently, foundation models. Yet this methodological diversity has outpaced systematic evaluation: most existing studies focus on individual hazards, single tasks, or specific method families, and lack reproducible cross-method comparison. In this paper, we provide a unified review and benchmark for EO-based disaster mapping. We define a taskdriven taxonomy distinguishing disaster extent mapping from infrastructure damage mapping, and systematically review methods across both tasks, spanning traditional, deep learning, and foundation model paradigms. We then benchmark over 30 methods under a standardized protocol across 13 datasets covering diverse hazards, sensors, and operational settings. Our studies reveal that architectural complexity yields diminishing returns for extent mapping, that change detection architectures benefit damage mapping but not extent mapping, and that foundation models do not yet consistently outperform task-specific baselines, particularly on non-RGB modalities. Based on these findings, we identify key priorities for advancing operational disaster mapping. Code and benchmark resources are publicly available at https://github.com/ChenHongruixuan/AnyDisasterMapping

Earthquakes are a destructive and oftentimes unanticipated force of nature. To facilitate timely disaster relief, very high resolution spaceborne observations can map urban destruction even over remote or inaccessible terrain. Fostering community-driven innovation on AI-based solutions for rapid mapping of building-level damage, ESA Φ-lab and the International Charter ’Space and Major Disasters’ jointly organized the AI for Earthquake Response competition. The activity was designed to emulate the needs and urge of real post-event activations. In its course, over 261 teams participated on the ESA Φ-lab Challenges platform.The main contribution of this work is to report the key setup and outcomes of the challenge as well as share with the community the winning strategies of the most competitive solutions. We will first provide an overview of recent and related work, then detail the core premises of the competition, including the two-phase structure of the challenge as well as its evaluation principles and data. We will also provide descriptions of the winning strategies of the best-performing teams, comprising details on data preparation, the data-driven modelling approach, and the respective team’s recap and discussion on their accomplishments. We will also review similarities or differences across models and distill key insights. Finally, we conclude by reviewing key findings and highlighting open challenges and opportunities for future contributions in rapid mapping for building damage assessment.We foresee this work to foster further innovation in the community, working towards data-driven rapid mapping that may in the future support real post-seismic activations and save human lives.

Earthquakes are a destructive and oftentimes unanticipated force of nature. To facilitate timely disaster relief, very high-resolution (VHR) spaceborne observations can map urban destruction even over remote or inaccessible terrain. Fostering community-driven innovation on artificial intelligence (AI)-based solutions for rapid mapping of building-level damage, the European Space Agency (ESA) and the International Charter "Space and Major Disasters" jointly organized the "AI for Earthquake Response" competition. The activity was designed to emulate the needs and urges of real postevent activations. In its course, more than 261 teams participated on the ESA Challenges platform and the best-performing AI model accom-plished an overall F1 score of 0.71. This work summarizes the competition's objective, data provided, and outcomes of the challenge. Descriptions for each of the three best-performing AI solutions and their workflows are provided, plus an overview summarizing their recipes for success. We foresee the event and this report as fostering further innovation in the community, working toward data-driven rapid mapping that may in the future support real postseismic activations and save human lives.

Up-to-date mapping of built-up areas is of paramount importance for urban planning, environmental monitoring, and disaster management. In recent years, there has been a growing interest in employing supervised machine learning and deep learning methods to map built-up areas using satellite SAR and optical data. However, the laborious and expensive task of gathering and maintaining a vast array of diverse training data poses a challenge to the widespread adoption of these methods for large-scale built-up area mapping. This paper presents a two-step framework enabling an automated extraction of built-up areas using Sentinel-1 and Sentinel-2 data. Initially, training data for built-up and non-built-up classes are automatically sampled and labeled from Sentinel-1 and Sentinel-2 data for a given area of interest. Subsequently, a cross-fusion neural network is trained using the samples from the first step to produce a built-up map for the entire study area. To enhance the network’s resilience to label noise, a contextual virtual adversarial training (CVAT) regularization is introduced within the mean-teacher architecture. Our proposed framework was tested on 48 different study areas across the world. Both quantitative and qualitative evaluations demonstrate its robustness and effectiveness for large-scale built-up area extraction. The versatility of our framework in generating accurate and up-to-date built-up information, which is essential for monitoring urban environments and assessing economic losses resulting from natural disasters, is highlighted through comparisons with four state-of-the-art global built-up products: Global Human Settlement Built-up map based on 2018 Sentinel-2 composites (GHS-BUILT-S2), World Settlement Footprint 2019 (WSF 2019), ESA World Cover, and Dynamic World.

We investigate the application of Federated Learning (FL) for ship detection across diverse satellite datasets, offering a privacy-preserving solution that eliminates the need for data sharing or centralized collection. This approach is particularly advantageous for handling commercial satellite imagery or sensitive ship annotations. Four FL models—FedAvg, FedProx, FedOpt, and FedMedian—are evaluated and compared to a local training baseline, where the YOLOv8 ship detection model is independently trained on each dataset without sharing learned parameters. The results reveal that FL models substantially improve detection accuracy over training on smaller local datasets and achieve performance levels close to global training that uses all datasets during the training. Furthermore, the study underscores the importance of selecting appropriate FL configurations, such as the number of communication rounds and local training epochs, to optimize detection precision while maintaining computational efficiency.

The development of the world economy in recent years has been accompanied by a significant increase in maritime traffic. Accordingly, numerous ship collision incidents, especially in dense maritime traffic zones, have been reported with damage, including oil spills, transportation interruption, etc. To improve maritime surveillance and minimize incidents over the seas, satellite imagery provided by synthetic aperture radar (SAR) and optical sensors has become one of the most effective and economical solutions in recent years. Indeed, both SAR and optical images can be used to detect vessels of different sizes and categories, thanks to their high spatial resolutions and wide swath. To process a mass of satellite data, Deep Learning (DL) has become an indispensable solution to detect ships with a high accuracy rate. However, the DL models require time and effort for implementation, especially for training, validating, and testing with big datasets. This issue is more significant if we use different satellite imagery datasets for ship detection because data preparation tasks will be multiplied. Therefore, this paper aims to investigate various approaches for applying the DL models trained and tested on different datasets with various spatial resolution and radiometric features. Concretely, we focus on two aspects of ship detection from multi-source satellite imagery that have not been attentively discussed in the literature. First, we compare the performance of DL models trained on one HR or MR dataset and those trained on the combined HR and MR datasets. Second, we compare the performance of DL models trained on an optical or SAR dataset and tested on another. Likewise, we evaluate the performance of DL models trained on the combined SAR and optical dataset. The objective of this work is to answer a practical question of ship detection in maritime surveillance, especially for emergency cases if we can directly apply the DL models trained on one dataset to others having differences in spatial resolution and radiometric features without the supplementary steps such as data preparation and DL models retraining. When dealing with a limited number of training images, the performance of DL models via the approaches proposed in this study was satisfactory. They could improve 5–20% of average precision, depending on the optical images tested. Likewise, DL models trained on the combined optical and radar dataset could be applied to both optical and radar images. Our experiments showed that the models trained on an optical dataset could be used for radar images, while those trained on a radar dataset offered very poor scores when applied to optical images.

The Remote Imaging Support for Emergencies (RISE) web application is designed to democratize access to geospatial intelligence for emergency management. Developed by WASDI in collaboration with the Luxembourg Institute of Science and Technology (LIST) and the Shelter Research Unit (SRU) of the Luxembourg Red Cross (AICRL), and funded by the World Food Programme, RISE aims to address the significant barriers in the adoption of satellite-based Earth Observation (EO) data, such as processing complexity, data latency, limited expertise, and accessibility. By integrating state-of-the-art algorithms for processing EO data with a user-friendly web interface, RISE delivers actionable insights in near-real time. The application focuses on flood mapping, drought monitoring, and urban area analysis, providing critical geospatial intelligence to support informed decision-making during humanitarian crises. RISE has a global coverage, but a validation exercise is about to be conducted in collaboration with the Luxembourg Red Cross, aiming to assess the performance improvements and practical usability of RISE in real-world conditions. The expected outcomes include reduced costs, shortened response times, improved safety, enhanced accuracy, and strengthened preparedness for future crises.

Comparing images captured by disparate sensors is a common challenge in remote sensing. This requires image translation-converting imagery from one sensor domain to another while preserving the original content. Denoising diffusion implicit models (DDIM) are potential state-of-the-art solutions for such domain translation due to their proven superiority in multiple image-to-image translation tasks in computer vision. However, these models struggle with reproducing radiometric features of large-scale multipatch imagery, resulting in inconsistencies across the full image. This renders downstream tasks like heterogeneous change detection impractical. To overcome these limitations, we propose a method that leverages denoising diffusion for effective multisensor optical image translation over large areas. Our approach super-resolves large-scale low spatial resolution images into high-resolution equivalents from disparate optical sensors, ensuring uniformity across hundreds of patches. Our contributions lie in new forward and reverse diffusion processes that address the challenges of large-scale image translation. Extensive experiments using paired Sentinel-II (10 m) and Planet Dove (3 m) images demonstrate that our approach provides precise domain adaptation, preserving image content while improving radiometric accuracy and feature representation. A thorough image quality assessment and comparisons with the standard DDIM framework and five other leading methods are presented. We reach a mean learned perceptual image patch similarity of 0.1884 and a Fr & eacute;chet Inception Distance of 45.64, expressively outperforming all compared methods, including DDIM, ShuffleMixer, and SwinIR. The usefulness of our approach is further demonstrated in two Heterogeneous Change Detection tasks.

Monitoring oil drift by integrating multi-source satellite imagery has been a relatively underexplored practice due to the limited time-sampling of datasets. However, this limitation has been mitigated by the emergence of new satellite constellations equipped with both Synthetic Aperture Radar (SAR) and optical sensors. In this manuscript, we take advantage of multi-temporal and multi-source satellite imagery, incorporating SAR (Sentinel-1 and ICEYE-X) and optical data (Sentinel-2/3 and Landsat-8/9), to provide insights into the spatio-temporal variations of oil spills. We also analyze the impact of met–ocean conditions on oil drift, focusing on two specific scenarios: marine floating oil slicks off the coast of Qatar and oil spills resulting from a shipwreck off the coast of Mauritius. By overlaying oils detected from various sources, we observe their short-term and long-term evolution. Our analysis highlights the finding that changes in oil structure and size are influenced by strong surface winds, while surface currents predominantly affect the spread of oil spills. Moreover, to detect oil slicks across different datasets, we propose an innovative unsupervised algorithm that combines a Bayesian approach used to detect oil and look-alike objects with an oil contours approach distinguishing oil from look-alikes. This algorithm can be applied to both SAR and optical data, and the results demonstrate its ability to accurately identify oil slicks, even in the presence of oil look-alikes and under varying met–ocean conditions.

Understanding the extent of urban flooding is crucial for assessing building damage, casualties and economic losses. Synthetic Aperture Radar (SAR) technology offers significant advantages for mapping flooded urban areas due to its ability to collect data regardless weather and solar illumination conditions. However, the wide range of existing methods makes it difficult to choose the best approach for a specific situation and to identify future research directions. Therefore, this study provides a comprehensive review of current research on urban flood mapping using SAR data, summarizing key characteristics of floodwater in SAR images and outlining various approaches from scientific articles. Additionally, we provide a brief overview of the advantages and disadvantages of each method category, along with guidance on selecting the most suitable approach for different scenarios. This study focuses on the challenges and advancements in SAR-based urban flood mapping. It specifically addresses the limitations of spatial and temporal resolution in SAR data and discusses the essential pre-processing steps. Moreover, the article explores the potential benefits of Polarimetric SAR (PolSAR) techniques and uncertainty analysis for future research. Furthermore, it highlights a lack of open-access SAR datasets for urban flood mapping, hindering development in advanced deep learning-based methods. Besides, we evaluated the Technology Readiness Levels (TRLs) of urban flood mapping techniques to identify challenges and future research areas. Finally, the study explores the practical applications of SAR-based urban flood mapping in both the private and public sectors and provides a comprehensive overview of the benefits and potential impact of these methods.

Deep neural networks (DNNs) have demonstrated remarkable success across various domains, including Earth Observation applications. Despite their achievements, DNNs do not quantify the uncertainty of their predictions, which is particularly crucial for high-stakes applications such as flood mapping. We applied density-aware deep neural networks for uncertainty quantification in SAR-based flood mapping through a single forward pass. The aleatoric uncertainty is captured through softmax entropy, while epistemic uncertainty is quantified using density in the latent feature space. Our image segmentation results illustrate that the employed density-aware deep neural networks exhibit good performance in uncertainty quantification, surpassing Deep Ensembles for out-of-distribution (OOD) data detection.

This study describes a method that combines synthetic aperture radar (SAR) data with shallow-water modeling to estimate flood hazards at a local level. The method uses particle filtering to integrate flood probability maps derived from SAR imagery with simulated flood maps for various flood return periods within specific river sub-catchments. We tested this method in a section of the Severn River basin in the UK. Our research involves 11 SAR flood observations from ENVISAT ASAR images, an ensemble of 15 particles representing various pre-computed flood scenarios, and 4 masks of spatial units corresponding to different river segmentations. Empirical results yield maps of maximum flood extent with associated return periods, reflecting the local characteristics of the river. The results are validated through a quantitative comparison approach, demonstrating that our method improves the accuracy of flood extent and scenario estimation. This provides spatially distributed return periods in sub-catchments, making flood hazard monitoring effective at a local scale.

Ship detection from satellite imagery using Deep Learning (DL) is an indispensable solution for maritime surveillance. However, applying DL models trained on one dataset to others having differences in spatial resolution and radiometric features requires many adjustments. To overcome this issue, this paper focused on the DL models trained on datasets that consist of different optical images and a combination of radar and optical data. When dealing with a limited number of training images, the performance of DL models via this approach was satisfactory. They could improve 5-20% of average precision, depending on the optical images tested. Likewise, DL models trained on the combined optical and radar dataset could be applied to both optical and radar images. Our experiments showed that the models trained on an optical dataset could be used for radar images, while those trained on a radar dataset offered very poor scores when applied to optical images.

<p>Over the last ten years, satellite imagery has become one of the most effective means to observe offshore oil spills, especially over a large area, thanks to their high spatial resolution and wide swath coverage. In particular, the use of images acquired by multi-sensors, including Synthetic Aperture Radar (SAR), optical, and visible/near-infrared, allows not only to early and quickly detect oil spills but also to monitor oil drift in short-terms (several hours) and long-terms (one or two days). Such approach has not been assessed in-depth in previous studies probably due to the lack of satellite data. A case study was presented in [1] to observe the movement of oil slicks through the combination of data acquired by SAR satellites, SAR airborne system, optical instrument, and in-situ observations. However, it is very complicated to implement such method for the other cases of oil spills since it requires many different platforms and sensors that are not systematically available. Therefore, this paper focuses uniquely on the collocation of sequential images acquired by various satellite sensors (SAR and optical) for oil drift monitoring in short-terms (several hours) and long-terms (up to 24–36 hours).</p> <p>For instance, to observe oil spills caused by a wrecked ship offshore Corsica (France) in Oct. 2018, we combine the images acquired by Sentinel-1 SAR (Oct. 8, 2018, 05:27:57 UTC, 17:21:45 UTC; Oct. 9, 2018, 17:14:27 UTC), Sentinel-2 optical sensor (Oct. 9, 2018, 10:20:21 UTC), and Radarsat-2 SAR (Oct. 9, 2018, 04:04:15 UTC). The techniques of oil spill detection from SAR and optical images are different. They are based on an advanced image processing procedure that will be presented at the conference. Due to the time lags between these images, we can estimate the movement of the detected oil slicks in terms of distance, velocity, and direction for 6, 12, 24, 36 hours of observation. Finally, we compare the oil drift results with the hourly data of met-ocean variables (surface wind and current) to assess the impact of the latter on oil drift. Surface wind fields (0.25° × 0.25° grid) and current vectors (0.083° × 0.083° grid) are extracted from the ERA-5 [2] and CMEMS [3] reanalysis data, respectively.</p> <p>[1] C. Brekke, M. M. Espeseth, K.-F. Dagestad, J. Röhrs, L. R. Hole, and A. Reigber, “Integrated analysis of multisensor datasets and oil drift simulations—a free-floating oil experiment in the open ocean,” J. of Geophysical Research: Oceans, vol. 126, e2020JC016499, 2021, doi: 10.1029/2020JC016499.</p> <p>[2] H. Hersbach et al., “ERA5 hourly data on single levels from 1959 to present,” Copernicus Climate Change Service (C3S) Climate Data Store (CDS). (Accessed on <05-12-2022>), 10.24381/cds.adbb2d47, 2018.</p> <p>[3] Global Ocean 1/12° Physics Analysis and Forecast updated Daily (Accessed on <05-12-2022>), doi: 10.48670/moi-00016.</p>

With the emergence of accurate high resolution remotely sensed datasets of hydrological variables, opportunities arise to study hydrological processes at an unprecedented scale and resolution. We took this opportunity to study spatiotemporal drought patterns over the country of Luxembourg. A daily 100x100 m2 soil moisture dataset based on the VanderSat technologya will be analysed in conjunction with a 6-day 60x60 m2 soil moisture dataset retrieved from Sentinel-1 data (Pulvirenti et al., 2018; van Hateren et al., 2021), and the Copernicus 1x1 km2 daily soil moisture product based on the TU Wien algorithmb. First of all, the consistency between the different products will be tested, as well as their ability to resolve small-scale variability in soil moisture. Then, the soil moisture products are compared to in situ soil moisture data, meteorological data and vegetation indices during major droughts in the last decade (2018, 2022). We will compare small scale spatial and temporal patterns of drought indices with land use, geology and elevation to see how the indices developed during these droughts and how they depend on the local landscape.ahttps://data.public.lu/en/datasets/soil-humidity-in-luxembourg-2002-2022/bhttps://land.copernicus.eu/global/products/ssmPulvirenti, L. et al., ‘A Surface Soil Moisture Mapping Service at National (Italian) Scale Based on Sentinel-1 Data’, Environmental Modelling & Software 102 (April 2018): 13–28, https://doi-org.ezproxy.library.wur.nl/10.1016/j.envsoft.2017.12.022.van Hateren, T.C. et al., ‘Optimal Spatial Resolution of Sentinel-1 Surface Soil Moisture Evaluated Using Intensive in Situ Observations’, in 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, 2021, 6311–14, https://doi-org.ezproxy.library.wur.nl/10.1109/IGARSS47720.2021.9553041.