Accurate inundated landscape mapping is essential for understanding hydrological connectivity, wetland dynamics and floodplain processes in tropical environments. However, optical and Synthetic Aperture Radar (SAR) sensors often produce inconsistent inundation estimates due to differences in sensor physics, wavelength sensitivity, vegetation and cloud penetration, and analytical methods. These challenges are amplified in wet–dry tropical systems, where cloud cover, heterogeneous vegetation and seasonal flooding complicate water detection. This study presents a unified, data-driven framework for comparing inundation products derived from Sentinel-2 optical imagery, Sentinel-1 C-band and PALSAR-2 L-band SAR in a tropical floodplain system in northern Australia. Inundated landscape maps were generated using supervised Random Forest classification and unsupervised multi-Otsu thresholding, followed by multi-sensor fusion analysis. Products were validated using wet- and dry-season field observations. Results showed strong seasonal and sensor-specific differences. Sentinel-2 provided the most balanced representation of dry-season floodplain vegetation and inundation, while Sentinel-1 achieved high wet-season accuracy (>0.95) under cloud-affected conditions. PALSAR-2 improved the detection of inundated vegetation and structurally complex environments, reaching an overall accuracy of 0.97. Multi-sensor fusion generally improved spatial coherence and class delineation, but performance varied across sensor combinations. These findings show that optical, C-band SAR and L-band SAR inundation products are not directly interchangeable in tropical wetlands. The proposed framework supports more robust interpretation, comparison and integration of multi-sensor inundation products for wetland monitoring, hydrological assessment and environmental management in data-limited tropical regions.

Floodplain inundation modelling is considered an essential part of river basin management for engineering, ecological and environmental perspectives. This study characterises inundation dynamics and quantifies river-wetland connectivity across the Daly River catchment using a two-dimensional hydrodynamic model (MIKE 21). The model was configured for the downstream reaches of the Daly River and its tributaries, focusing on floodplain regions subject to seasonal inundation. High resolution topographic data (1 m LiDAR) for the main river and 30 m FABDEM for surrounding areas) were used to construct the elevation model. A gridded land use and land cover (LULC) map was used to derive surface roughness in the model. We assessed inundation dynamics using five different high flood events from 2015 to 2024. To calibrate the MIKE 21 HD model, we analysed a representative flood event by benchmarking modelled water levels and extents against Sentinel-1–derived inundation maps and gauged water levels. Our results reveal spatial and temporal variability in floodplain inundation and highlight the influence of both seasonal and anthropogenic factors on inundation extent.

Water is essential to life and health of various ecological and social systems. Unfortunately, water is one of the natural resources most impacted by climate change, with increasingly intense hydro-meteorological extremes (floods, droughts, etc.) and growing societal demand. To help manage this vulnerable resource, it is vital to assess and monitor its availability on a regular basis, as well as to track its trajectory over time to better understand the impact of global change on it. Surface water (lakes, rivers, flood plains, etc.) represents an important component of total water resources, and it is of primary importance to monitor it to better understand and manage the consequences of climate change. Surface water resources provide populations around the world with essential ecosystem services such as power generation, irrigation, drinking water for humans and livestock, and space for farming and fishing.In this context, the SCO-CASCADES project implements end-to-end processing chains for satellite Earth observation data, including Sentinel-1 and 2 (S-1 and S-2), in order to provide surface water products (surface water body and inundation depth maps) that will be made available via an interactive platform co-constructed with identified users.In the first phase of the project a fully automated Sentinel-1 based processing chain has been implemented. This chain is based on automatic multiscale image histogram parameterization followed by thresholding, region growing and chain detection applied on individual, subsequent pairs, and time series of S1 images. This chain enables us to derive various products: i) an exclusion layer identifying areas where water cannot be detected on Sentinel 1 image (e.g. Urban and forested areas), ii) permanent seasonal water body maps, iii) a water body map for each S1 image, iv) an uncertainty map characterizing the water body classification uncertainty, v) an occurrence map providing the number of times (over the time series) each pixel was covered by open water.Here, we propose to present and evaluate the robustness of the processing chain and the resulting maps produced using multi-year S1 time series over two large scale sites: the Mekong flood plains between Kratie, the Tonle Sap lake and the Mekong Delta, and the Tsiribihina basin in Madagascar. The kappa score obtained from the comparison between S1 and S2-derived maps shows a good agreement yielding CSI and Kappa Cohen scores most of the time higher than 0.7 and sometimes reaching values higher than 0.9.

Segmenting Synthetic Aperture Radar (SAR) images is crucial for many remote sensing applications, particularly water body detection. However, deep learning-based segmentation models often face challenges related to convergence speed and stability, mainly due to the complex statistical distribution of this type of data. In this study, we evaluate the impact of mode normalization on two widely used semantic segmentation models, U-Net and SegNet. Specifically, we integrate mode normalization, to reduce convergence time while maintaining the performance of the baseline models. Experimental results demonstrate that mode normalization significantly accelerates convergence. Furthermore, cross-validation results indicate that normalized models exhibit increased stability in different zones. These findings highlight the effectiveness of normalization in improving computational efficiency and generalization in SAR image segmentation.

As climate change exacerbates variability and non-stationarity in rainfall patterns, it is crucial to assess the predictive capabilities of forecasting models. Previous researches on rainfall-runoff modeling have focused on the impact of training dataset size on artificial neural networks (ANNs) results, with limited consideration of hydrometeorological diversity. This study first evaluates the influence of the training dataset length (1–15 years) on performance of a Long Short-Term Memory (LSTM) and a traditional conceptual model, Superflex, across 10 validation years. Next, training years are categorized based on hydrometeorological diversity (wetter, standard, drier). This clustering allows for experiments where models are trained on data from similar or different clusters, enhancing understanding of how data diversity, and therefore climate change, can affect model performance. Results indicate that the LSTM model is highly sensitive to training length, showing poor performance with short datasets (below three years), reaches similar performance to Superflex around six training years on average, and overperforms with 15 years of training. Conversely, Superflex maintains rather constant performance levels regardless of the dataset length. LSTM model benefits from diverse training data, achieving higher accuracy and reliability when trained on years with diverse hydrological typology. Despite their potential to outperform traditional models (with six or more training years on average), LSTM models are highly dependent on the quality and diversity of training data. In climate change scenarios, caution is needed when applying LSTM models to unfamiliar conditions, as their predictive accuracy may decline more rapidly than that of more traditional hydrological models.

Soil moisture content (SMC) is a critical parameter for agricultural productivity, particularly in semi-arid regions, where irrigation practices are extensively used to offset water deficits and ensure decent yields. Yet, the socio-economic and remote context of these regions prevents sufficiently dense SMC monitoring in space and time to support farmers in their work to avoid unsustainable irrigation practices and preserve water resource availability. In this context, our study addresses the challenge of high spatial resolution (i.e., 20 m) SMC estimation by integrating remote sensing datasets in machine learning models. For this purpose, a dataset made of 166 soil samples’ SMC along with corresponding SMC, precipitation, and radar signal derived from Soil Moisture Active Passive (SMAP), Integrated Multi-satellitE Retrievals for GPM (IMERG), and Sentinel-1 (S1), respectively, was used to assess four machine learning models’ (Decision Tree—DT, Random Forest—RF, Gradient Boosting—GB, Extreme Gradient Boosting—XGB) reliability for SMC mapping. First, each model was trained/validated using only the coarse spatial resolution (i.e., 10 km) SMAP SMC and IMERG precipitation estimates as independent features, and, second, S1 information (i.e., 20 m) derived from single scenes and/or composite images was added as independent features to highlight the benefit of information (i.e., S1 information) for SMC mapping at high spatial resolution (i.e., 20 m). Results show that integrating S1 information from both single scenes and composite images to SMAP SMC and IMERG precipitation data significantly improves model reliability, as R2 increased by 12% to 16%, while RMSE decreased by 10% to 18%, depending on the considered model (i.e., RF, XGB, DT, GB). Overall, all models provided reliable SMC estimates at 20 m spatial resolution, with the GB model performing the best (R2 = 0.86, RMSE = 2.55%).

This study proposes a dual approach to assess 20 Gridded Precipitation Products (GPPs) within a transboundary region with complex topography. GPPs were first compared with observed precipitation to assess their spatiotemporal accuracy. They were then integrated into a lumped (GR4J) and a semi-distributed (MGB-IPH) hydrological model to evaluate their impact on streamflow simulations for three basins. Even if most GPPs effectively captured the dominant north-south precipitation gradient shaped by the Andean topography, the results show significant variations in GPP effectiveness across the considered basins. The most reliable GPPs for streamflow simulation across Katari, Ilave, and Ramis basins are MSWEP, CHIRPS, and MSWEP when considering the lumped GR4J model and SM2Rain_CCI, IMERG_FR, and SM2Rain_CCI when considering the semi-distributed MGB model. This discrepancy among the models shows that GPPs' reliability assessment is sensitive to the model structure and that different conclusions could be made according to the selected model. Our findings show that the GR4j lumped model is barely influenced by precipitation bias due to its buffering capacity. In contrast, the semi-distributed MGB-IPH model is sensitive to precipitation bias in space and time and therefore is more suitable to reveal GPP inconsistencies. Overall, this study not only provides GPP reliability feedback but also new insights on the respective limits and advantages of different assessment methods (i.e., gauges, lumped and semidistributed models). These findings support the development of a practical framework for GPP selection according to the forecast use.

Accurate discharge prediction in hydrological forecasting relies on robust modeling. This study investigates the Long Short-Term Memory (LSTM) model’s performance, focusing on training dataset size and hydrometeorological patterns. Convolutional Neural Networks (CNNs) and Artificial Neural Networks (ANNs) are also considered for spatial and temporal dependencies. Data is drawn from the CAMELS-GB dataset (1975–2015, Saxons Lode, UK). Results show that LSTM performance varies, with years surrounding high water events (like 2004) performing poorly in training, and struggles in validation. Training with one year yields 23.03

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.

This study assesses the relative performance of Sentinel-1 and -2 and their combination with topographic information for plow agricultural land soil salinity mapping. A learning database made of 255 soil samples’ electrical conductivity (EC) along with corresponding radar (R), optical (O), and topographic (T) information derived from Sentinel-2 (S2), Sentinel-1 (S1), and the SRTM digital elevation model, respectively, was used to train four machine learning models (Decision tree—DT, Random Forest—RF, Gradient Boosting—GB, Extreme Gradient Boosting—XGB). Each model was separately trained/validated for four scenarios based on four combinations of R, O, and T (R, O, R+O, R+O+T), with and without feature selection. The Recursive Feature Elimination with k-fold cross validation (RFEcv 10-fold) and the Variance Inflation Factor (VIF) were used for the feature selection process to minimize multicollinearity by selecting the most relevant features. The most reliable salinity estimates are obtained for the R+O+T scenario, considering the feature selection process, with R2 of 0.73, 0.74, 0.75, and 0.76 for DT, GB, RF, and XGB, respectively. Conversely, models based on R information led to unreliable soil salinity estimates due to the saturation of the C-band signal in plowed lands.

•Data assimilation into a fully-distributed 2D-SWE model.•Joint assimilation of satellite-derived soil moisture and observed discharge.•Data assimilation using a Particle Filter-based approach.•The proposed method works better in events with isolated peaks of discharge.•In terms of discharge, an average NSE value of 0.74 was achieved.

This article reviews the state of the art in the use of space-borne observations for analyzing extreme rainfall and flood events in Africa. Floods occur across many space and timescales, from very localized flash flood events to slow propagation of discharge peaks in large rivers. We discuss here how satellite data can help us understand the genesis and impacts of these flood events, monitor their evolution, and better constrain prediction models, thereby improving early warning and population protection. To illustrate these topics, we reanalyze major flood events that occurred in Niger, Mozambique, Central African Republic and Ivory Coast, using satellite information.

The Global Flood Monitoring (GFM) system of the Copernicus Emergency Management Service (CEMS) addresses the challenges and impacts that are caused by flooding. The GFM system provides global, near-real time flood extent masks for each newly acquired Sentinel-1 Interferometric Wide Swath Synthetic Aperture Radar (SAR) image, as well as flood information from the whole Sentinel-1 archive from 2015 on. The GFM flood extent is an ensemble product based on a combination of three independently developed flood mapping algorithms that individually derive the flood information from Sentinel-1 data. Each flood algorithm also provides classification uncertainty information that is aggregated into the GFM ensemble likelihood product as the mean of the individual classification likelihoods. As the flood detection algorithms derive uncertainty information with different methods, the value range of the three input likelihoods must be harmonized to a range from low [0] to high [100] flood likelihood. The ensemble likelihood is evaluated on two test sites in Myanmar and Somalia, showcasing the performance during an actual flood event and an area with challenging conditions for SAR-based flood detection. The Myanmar use case demonstrates the robustness if flood detections in the ensemble step disagree and how that information is communicated to the end-user. The Somalia use case demonstrates a setting where misclassifications are likely, how the ensemble process mitigates false detections and how the flood likelihoods can be interpreted to use such results with adequate caution.

Hydrological modelling is one of our main tools for flood forecasting. It is essential to help stakeholders planning disaster emergency response. Although a consensus has been reached in the scientific community accepting as beneficial the integration of satellite data with hydrological models, the related assimilation techniques are still evolving. Particle filter-based assimilation techniques have proven their usefulness in various hydrological studies, but degeneracy and sample impoverishment remain their main limitations, as the ensemble of particles lose variety and only a few of them retain a significant weight in the posterior probability distribution after one or a few assimilation steps. Although this new technique has already started to be implemented in large-scale hydrological models, it has still been briefly applied to fully distributed high resolution hydrological models. In this study, a particle filter-based method (Tempered Particle Filter) is utilized to jointly assimilate SMAP soil moisture data and discharge timeseries into Iber+, a fully distributed hydrodynamic model that combines rainfall runoff and shallow water modelling. We use as a case study a 200 km2 catchment located in the northwest of Spain. The results indicate that: (1) soil moisture prediction can be strongly biased if only the streamflow at the catchment outlet is assimilated; (2) streamflow prediction, and especially surface soil moisture prediction, can be improved through the joint assimilation of streamflow and soil moisture observations; (3) it is possible to directly relate the antecedent soil moisture condition of the catchment to some of the hydrological model parameters.

With floods becoming more severe and frequent due to climate change and growing urbanisation, there is a crucial need to improve flood management systems. Flood hazard assessment using hydrodynamic models is more challenging at a large scale, because discretizing an area using a fine mesh is essential for obtaining accurate results, but is found to be highly expensive computationally. The emergence of sub grid models in the past few decades has enabled faster simulations by using coarser cells while preserving small-scale topography variations within one cell. In this context, we propose a modelling framework based on the shallow water 2D model with depth-dependant porosity (SW2D-DDP). We evaluate this approach by setting up a standard 2D shallow water model (SW2D) considered as a benchmark. The 2007 flood event of the River Severn is used as a test case. Our preliminary results demonstrate a high accuracy (~90% during flood peak) and a low computational cost with a ~350 runtime reduction factor of the proposed model compared to a standard model. This opens up new perspectives for large scale applications over areas where bathymetric data is not available.

Previous studies have shown that the decrease of temporal interferometric synthetic aperture radar (InSAR) coherence could be exploited to detect the appearance of floodwater in urban areas. However, as of today, approaches based on this principle only make use of single co-polarization images for identifying the presence of floodwater in the double-bounce feature. In this study, we take advantage of both co- and cross-polarization images to detect significant decreases of the multitemporal InSAR coherence in order to enhance the mapping of floodwater in urban areas. We consider that not only double-bounce scattering, but also multiple-bounce may occur in urban areas depending on how the building facades are oriented with respect to the synthetic aperture radar (SAR) sensor's line of sight. The Sentinel-1 (S-1) mission is particularly well suited for applying and testing this kind of approach due to the systematic availability of dual-polarization data. Using as a test case, the widespread flooding in the city of Houston, USA, caused by Hurricane Harvey in 2017, we demonstrate that the proposed methodology leads to an increase of the accuracy of the urban flood maps from 75.2% when only using the VV polarization, to 82.9% when using the dual polarization information.

Data Assimilation can improve forecast accuracy of flood inundation models by an optimal combination of uncertain model simulations and observations. Particle Filter (PF) has gained interest in the research community for its ability of dealing with non-linear systems and with any kind of observation and model error distributions. Using PF, one of the most difficult issues to deal with are degeneracy and sample impoverishment. In this study, we have investigated a novel approach, based on a Tempered Particle Filter (TPF), aiming to circumvent these issues, and increase the persistence in time of the assimilation benefits. Flood probabilistic maps derived from Synthetic Aperture Radar (SAR) data are assimilated into a flood forecasting model. The process is iterative and includes a particle mutation in order to keep diversity within the ensemble. Results show an improvement of the model forecast accuracy as a result of the assimilation with respect to the ensemble without any assimilation (Open Loop, OL): on average the Root Mean Square Error decreases by 80% at the assimilation time and by 60% 2 days after the assimilation. A comparison with a standard PF, the Sequential Importance Sampling (SIS), where degeneracy occurred, is carried out. Results are similar at the assimilation time but the increase of performance using the TPF are lasting longer. For instance, TPF-based RMSE are still lower than the OL RMSE 3 days after the assimilation, while the SIS-based RMSE becomes larger than the RMSE-OL 2 days after the assimilation.

With growing urbanisation and climate change, flooding is likely to become even more frequent and severe. Therefore, it is essential to constantly monitor water level changes at a large scale. Synthetic Aperture Radar (SAR) images are often used for flood mapping as they allow a rather straightforward detection of water bodies, at almost all weather and illumination conditions. As the number of satellite observations and global digital elevation models (DEMs) are becoming more available, this mapping approach is gaining in popularity. In this study, we propose three different approaches to retrieve water level maps based on the combination of satellite and topographic data, hereby referred to as the global Height Above Nearest Drainage (HAND), the local HAND and the local DEM methods. The three approaches are based on the optimization of a threshold applied on the topography (HAND or DEM) data enabling a best fit with the SAR-extracted flood map. The optimized threshold values provide at the same time the normalized water levels (with respect to the drainage network). The water depth map is thus computed from the difference between the water level and the DEM. The global HAND method applies a single optimized threshold to the HAND map, over the entire area of interest. The local HAND method is based on the same concept but optimizes and applies the HAND threshold value locally, using a sliding window. The local DEM thresholding method employs the same principle as the second method but directly on the DEM. We evaluate these methods using hydraulic model simulation results and ground truth data, and we carry out several experiments using various SAR images (Envisat, TerraSAR-X and Sentinel-1) and topographic datasets (SRTM, CopDEM and LiDAR). The best results are obtained while combining a high-resolution image (e.g:TerraSAR-X) with: 1) a high-resolution dataset (e.g: LiDAR DEM), using the local DEM approach, or 2) a coarse-resolution dataset (e.g: Srtm DEM), using the global HAND approach. RMSDs on the derived water depth maps reach respectively 0.52m and 0.93m.