Operational monitoring of total water stocks in reservoirs at a large scale suffers from a lack of knowledge on the bathymetry of such water bodies. When known, a measured water area can be directly linked to the total water volume. Digital Elevation Models can be very useful to this regard, but often reservoirs bathymetry is not visible since they are filled with water. In such a context, recent developments in Artificial Intelligence might be helpful. Using an inpainting approach, a U-Net algorithm was trained on several thousands of virtual lakes to estimate reservoirs bathymetry, using a true DEM as inputs. Its capacity to determine accurate bathymetry and derived surface to volume relationship has been tested and assessed over the Guadalquivir basin in Southern Spain. The R2score of 0.79 obtained over one reservoir on volume estimation along time is encouraging for the future operational monitoring of the whole basin.

In spite of astonishing advances and developments in remote sensing technologies, meeting the spatio-temporal requirements for flood hydrodynamic modeling remains a great challenge for Earth Observation. The assimilation of multi-source remote sensing data in 2D hydrodynamic models participates to overcome such a challenge. The recently launched Surface Water and Ocean Topography (SWOT) wide-swath altimetry satellite provides a global coverage of water surface elevation at a high resolution. SWOT provides complementary observation to radar and optical images, increasing the opportunity to observe and monitor flood events. This research work focuses on the assimilation of 2D flood extent maps derived from Sentinel-1 C-SAR imagery data, and water surface elevation from SWOT as well as in-situ water level measurements. An Ensemble Kalman Filter (EnKF) with a joint state-parameter analysis is implemented on top of a 2D hydrodynamic TELEMAC-2D model to account for errors in roughness, input forcing and water depth in floodplain subdomains. The proposed strategy is carried out in an Observing System Simulation Experiment based on the 2021 flood event over the Garonne Marmandaise catchment. This work makes the most of the large volume of heterogeneous data from space for flood prediction in hindcast mode paves the way for nowcasting.

Flood simulation and forecast capability have been greatly improved thanks to advances in data assimilation (DA) strategies incorporating various types of observations; many are derived from Earth Observations from space. This article focuses on the assimilation of 2-D flood observations derived from synthetic aperture radar (SAR) images acquired during a flood event with a dual state-parameter ensemble Kalman filter (EnKF). Resulting binary wet/dry maps are here expressed in terms of wet surface ratios (WSRs) over a number of subdomains of the floodplain. This ratio is assimilated jointly with in situ water-level observations to improve the flow dynamics within the floodplain. However, the non-Gaussianity of the observation errors associated with these SAR-derived measurements breaks a major hypothesis for the application of the EnKF, thus jeopardizing the optimality of the filter analysis. The novelty of this article lies in the treatment of the non-Gaussianity of the SAR-derived WSR observations with a Gaussian anamorphosis (GA) process. This DA strategy was validated and applied over the Garonne Marmandaise catchment (southwest of France) represented with a TELEMAC-2D hydrodynamic model, first in a twin experiment and then for a major flood event that occurred in January and February 2021. It was shown that assimilating SAR-derived WSR observations in complement to the in situ water-level observations significantly improves the representation of the flood dynamics. The GA process brings further improvement to the DA analysis while also demonstrating to be a nonessential element. This study heralds a reliable solution for flood forecasting over poorly gauged catchments thanks to available remote sensing datasets.

The challenges in operational flood forecasting lie in producing reliable forecasts given constrained computational resources and within processing times that are compatible with near-real-time forecasting. Flood hydrodynamic models exploit observed data from gauge networks, e.g. water surface elevation (WSE) and/or discharge that describe the forcing time-series at the upstream and lateral boundary conditions of the model. A chained hydrologic-hydraulic model is thus interesting to allow extended lead time forecasts and overcome the limits of forecast when using only observed gauge measurements. This research work focuses on comprehensively reducing the uncertainties in the model parameters, hydraulic state and especially the forcing data in order to improve the overall flood reanalysis and forecast performance. It aims at assimilating two main complementary EO data sources, namely in-situ WSE and SAR-derived flood extent observations.

Owing to advances in data assimilation, notably Ensemble Kalman Filter (EnKF), flood simulation and forecast capabilities have greatly improved in recent years. The motivation of the research work is to reduce comprehensively the uncertainties in the model parameters, forcing and hydraulic state, and consequently improve the overall flood reanalysis and forecast capability, especially in the floodplain. It aims at assimilating SAR-derived (typically from Sentinel-1 mission) flood extent observations, expressed in terms of wet surface ratio. The non-Gaussianity of the observation errors associated with the SAR flood observations violates a major hypothesis regarding the EnKF and jeopardizes the optimality of the filter analysis. Therefore, a special treatment of such non-Gaussianity with a Gaussian anamorphosis process is thus proposed. This strategy was validated and applied over the Garonne Marmandaise catchment (South-west of France) represented with the TELEMAC-2D hydrodynamic model, focusing on a major flood event that occurred in December 2019. The assimilation of the SAR-derived wet surface ratio observations, in complement to the in-situ water surface elevations, is illustrated to consequentially improve the flood representation.

Dam reservoirs account for only a small fraction of the global freshwater but they have major importance for human activities, providing water for agriculture, domestic, recreational, and industrial use, especially in a context of water scarcity induced by climate change [1] . They also have major impacts on the hydro- and ecosystems [2] . While they are subject to numerous studies across multiple disciplinary fields [3] , [4] , no exhaustive database of existing dam reservoirs is available, and the available data are limited to the largest reservoirs. Thus, there is a need for a method to automatically extract dam reservoirs from widely available data such as satellite images. This task is different from the broader water surface extraction task [5] , [6] as the goal is also to distinguish dam reservoirs from other waterbodies. Some approaches have been proposed to achieve it, for example by Van Soesbergen et al. [7] . However, this extraction task using only single images can be very difficult for very small waterbodies (below 1 hectare).

Wetlands and open waters are key components of the hydrological and carbon cycles but their spatio-temporal dynamics are still not well known at global scale. Current paper presents a new methodology to retrieve water fraction at coarse scale and high temporal resolution (one week) using L-Band multi-angular and dual polarisation remote sensing data from SMOS mission. The dataset labeled G-SWAF (or Global-SWAF) is an extention of the SWAF approach which did not consider the separate contributions of the Soil, Vegetation and water fractions. The comparison to existing datasets shows more water fraction detecting in Tropical areas but better consideration of high latitudes is still needed to be included in future studies. The use of such datasets with the future data from the SWOT (NASA/CNES) mission will provide global evaluation of inland open water volumes at 10 days scale.