Terrestrial waters represent less than 1% of the total water stored on Earth, play a major role in climate variability and are necessary for sustaining terrestrial life and human needs. Among the various reservoirs in which fresh water is stored (e.g., ice caps, glaciers, snowpack, soil moisture and groundwater), surface waters (rivers, lakes, reservoirs, wetlands and inundated areas) play a crucial role in the global biogeochemical and hydrological cycles.
As a part of the hydrologic cycle, investigating the spatio-temporal variation of the surface water storage (SWS) is fundamental to the study of the global water cycle while providing a critical parameter for water resources management.
Until recently, our knowledge of surface water dynamics relied on sparse in situ observations and hydrological models. Traditional in situ gauge measurements quantify the water discharge in river channels, but no information is provided by these measurements regarding the diffusive flow over floodplains associated to rivers or wetlands. Furthermore, the number of gauging networks is, in general, limited especially in remote areas with difficult access such as tropical regions. When available, discharge data and hydrological observations are often classified by governments due to transboundary issues and their access is restricted for scientific usage.
Recent efforts have been undertaken to quantify the surface freshwater storage and its variations at seasonal to inter-annual time scales using satellite observations. Estimates of SWS variations were obtained combining surface water extent observations with altimeter-derived height variations over rivers, lakes and wetlands. They allowed studying the impact of extreme climate events, such as the extreme drought that occurred in the Amazon Basin in 2005 or the ENSO-related droughts and floods in the Tonle Sap drainage area, on the surface hydrological reservoir.
In this study, we included topographical information from several digital elevation models (DEM) (i.e., SRTM, ASTER, MERIT, Tandem-X) to improve the interpolation of altimetry-based water levels over inundated areas, taking into account the slopes inside the drainage area. Altimetry-based water levels were derived from ERS-2, ENVISAT and SARAL data, and inundation maps from MODIS. Comparisons between the classical method and this new approach are performed in different drainage areas – the Inner Niger Delta, the Mekong Basin and the McKenzie Delta. A thorough discussion is made on the impact of the choice of the DEM and of the geoid model used (i.e., EGM96, EGM2008, EIGEN6) on the SWS variations.

Knowledge on dynamic of regional surface water storage is vital for local to regional water management and for the prediction of water-related hazards. In northeastern (NE) Brazil, majority of the reservoirs lack this information. In this study, the digital elevation models (DEM) of very high resolution generated from TanDEM-X data provided the bathymetry for over 2 000 reservoirs in four representative regions of a total area of 10 000 km2 in NE Brazil. Based on the reservoirs’ bathymetry and a time series of RapidEye satellite images acquired in the period of 2009-2017, the temporal and spatial variations of surface water storage in the four regions were obtained. For validation purposes, surface water storages derived from in-situ measurements were employed. This study showed that: 1) DEMs derived from TanDEM-X acquired in the deep dry season can reveal the full bathymetry for 90% of the reservoirs; 2) NE Brazil is densely dammed with 0.04-0.23 reservoirs km2, a surface water area of 0.78-4.13 ha/km2, and a storage capacity of 0.01-0.23 hm3/km2; 3) Among all the reservoirs, those < 10 ha account for the majority of the water body population but contribute least to the regional surface water storage; 4) From 2009 to 2017 the overall surface water storage decreased at an average rate of 4000 m3/(year·km2); 5) The decrease of surface water storage in the period of 2009-2017 showed high spatial heterogeneity. Our study has filled the data gap of the reservoir bathymetry and surface water storage capacity for the studied over 2 000 reservoirs in NE Brazil. The high consistence between the surface water storage yielded by remote sensing and that from in-situ measurements demonstrated the potential of monitoring the surface water storage with remote sensing approach. In addition, the surface water storages derived for these reservoirs for the past eight years can facilitate water management, e.g. by constraining and validating hydrological modelling in the regions or by estimating water losses caused by evaporation.

Global mean sea level (GMSL) has been rising at a rate of 3.1 ± 0.5 mm/yr since 1993 as measured by satellite altimetry. The rate of rise is attributed to an increased mass loss from glaciers, Antarctic and Greenland ice sheets and to ocean warming. Land water storage that comprises snow, surface water, soil moisture, and groundwater storage, excluding glaciers also significantly contributes to sea level rise. In fact, the main source of interannual variability in global mean sea level comes from land water storage in response to natural climate variability. Yet, in terms of sea level trend, it is the most uncertain contribution. In this study, we analyzed land water storage variations using two global hydrological models, Water Gap Hydrological Model (WGHM) and Interactions between Soil, Biosphere, and Atmosphere–Total Runoff Integrating Pathways (ISBA-TRIP) model, as well as three mascon-based GRACE solutions (from CSR, JPL and GSFC) over 2003-2015. The land water storage contribution (expressed in Equivalent Sea Level –ESL) from 343 major hydrological basins (defined according to the Global Run-off Data Center (GRDC) shapefiles) were summed up to obtain the global mean land water storage contribution. The ensemble mean GRACE mascon-based data over the 346 basins contributes to 0.04 ESL mm/yr over the study period, whereas trends from the WGHM and ISBA-TRIP hydrological models account to 0.14 mm/yr and 0.08 mm/yr respectively. While several studies have shown opposite land water contributions to GMSL between GRACE and hydrological models (with GRACE producing a negative sea level contribution while the models show vice versa), our study shows positive contributions (even though almost zero in the case of GRACE mascons) in both cases. We also show comparisons at basin-scale, allowing us to identify the basins where GRACE and models agree well and those where significant discrepancies are observed.

Synthetic Aperture Radars (SARs) are well suited for water bodies mapping from space since SAR sensors can acquire data of equal quality day-and-night and almost regardless of weather condition. Moreover, SAR imagery has been often successfully used in mapping flooded areas based on the fact that the backscatter from calm and open water bodies is very low, allowing them to be separated from other land cover classes relatively easily. However, most of the studies in this field focus on mapping the flood extent with high accuracy for specific areas with a surface size inferior to one of the SAR image spatial coverage. Therefore, the flood study regions have typically been limited to a small subset of an entire SAR frame and the performance of the previous algorithms over a large-scale region with variable ground conditions has been rarely reported.
In this study, we propose an automatic change-detection based method that allows generating a flood record at a global scale using the entire ENVISAT archive. It consists of the following steps: (i) potential flood image selection; (ii) optimal reference image selection; (iii) flood mapping using change detection. Since the data processing required for generating a global product is computationally expensive, it is important to select the potential flood images at the very beginning.
The proposed method assumes that the backscatter distributions of water and non-water pixels in a SAR image are two different Gaussian distributions, which are mixed into a bimodal distribution. In addition, it hypothesizes that the distributions of change and non-change in the difference image derived from the reference-flood SAR image pair are also two different Gaussian distributions leading to a bimodal distribution. Following these working hypotheses, the proposed method makes use of hierarchical split-based approach in order to identify the sub-tiles, of a variable size, that verify the bimodal distribution assumption. This test procedure is applied to all possible combinations of SAR image pairs available in the archive. The resulting selected sub-tiles are hereafter named Bimodal Mask (BM). In a pair of non-flood images, the BM should be empty. For every flood image candidate, several BMs are generated using different combinations with several reference candidates. It is then defined that a SAR image can be considered as a potential flood image if 30% of the combinations can generate non-empty BMs. Based on a non-empty BM, Gaussian distributions of non-water backscatter can be parametrized for both the flood image and the reference image candidate, hereafter denoted as G₁ and G₂. Next, the Euclidean distance between these two Gaussian distributions is computed in order to measure the level of similarity between G₁ and G₂ and the optimal reference image is selected as the image with the lowest value. Then, flood mapping is carried out using the method proposed by Chini et al. (2017) based on the selected flood image and the reference image. In addition to this method, a filtering that considers the SAR incidence angle effect and the Height Above the Nearest Drainage (HAND) index has been applied to reduce false alarms.
In this test case, we focus on all SAR images acquired from four different ENVISAT tracks over the UK. Due to the lack of large-scale ground truth data, the evaluation of the results has been done in two steps. First, subsets of the flood maps over the Severn River areas are evaluated using a flood extent map that was manually digitized from high-resolution aerial imagery. Next, for the other regions and images where no ground truth data is available, the Environment Agency Flood Maps for Planning (EA FMP) of the UK are used to compare the remote sensing data with model results. The two images acquired during the flood event in July 2007 have been selected by the proposed method as flood images and corresponding flood maps have been generated. According to our results, the overall accuracies of both flood maps’ subsets are higher than 80% while the user’ accuracies of the flood class are higher than 90%. Moreover, a good agreement has been found between the entire frame of SAR-derived flood maps and EA FMP. The experimental results over the UK indicate that the proposed method has strong potential for the generation of a global flood record from the ENVISAT Archive.

A novel methodology for estimating flood depth and volume by combining Sentinel-1 derived flood extents and high-resolution LiDAR DTMs will be presented. Utilising the improved temporal resolution from Sentinel-1, the changes in flood volume during the 2015-16 winter floods across northern England are analysed, showcasing new insights into the water movement and storage. The data garnered from this technique can provide valuable information to emergency planners, and be used for data assimilation and validation of hydrodynamic models.

Inundation extents have been determined using a pixel-based classification methodology, which combines change detection and variable histogram thresholding (Clement et al. 2018). A river-at-bankfull dataset, created from the LiDAR DTM, is combined with the flood extents to create a surface water map. Shoreline heights are extracted from the LiDAR data, and a grid with overlapping cells is created over the flood. For each cell, a water surface is estimated using least squares regression, and from this flood slope, depth and volume can be derived. Water surface heights are validated against in-situ stage measurements, and flood depths are compared to hydrodynamic model outputs. The method proposed has been developed using UK specific datasets, such as the LiDAR DTM, however, the presentation will also look at the global applicability of the proposed methods by substituting the country-specific datasets with globally available, lower resolution ones, such as SRTM.

A discussion will be provided on some of the challenges when combining flood maps and terrain data to estimate volume. This includes accounting for the unknown bathymetry of the river network, with most DTMs showing the river height at the time of acquisition. “Volume above Bankfull”, and its change through the course of a flood, has been used in this analysis due to the unknown quantity of water in the rivers. Furthermore, the mismatch in spatial resolution between the inundation maps and the terrain data can result in flood boundaries being inaccurately located when overlaid onto the topography. This can be further exacerbated by SAR imagery potentially underestimating the flood boundary due to emergent vegetation, along with being unable to identify flooding accurately in woodland and urban regions. A solution is found using the local DTM statistics for the region around the shoreline to identify the probable flood height.

The creation of algorithms for reliably determining flood depths and volumes allows for more detailed monitoring and understanding of the dynamics of a flood. Future improvements and opportunities utilising the described method will be identified and discussed, such as the required data for applying globally and near real-time monitoring.

The event of Hurricane Harvey generated a wealth of high-resolution satellite EO data over several weeks. Rapidly processing the acquired satellite imagery provided critical information on the large-scale flooding that was unfolding.

This study presents an application of a recently introduced near real-time flood forecasting system that makes use of regularly acquired satellite EO data to keep model predictions on track. We take advantage of a recently developed satellite EO data processing algorithm enabling an efficient and fully automated generation of probabilistic flood maps using sequences of Sentinel-1 Synthetic Aperture Radar (SAR) images acquired during the flooding event. This algorithm is based on Bayes theorem and assigns to any pixel of a SAR image its probability of being water (i.e., flooded) using its backscatter values. This algorithm first estimates automatically the probability distribution functions of backscatter values associated with flooded (open water) and non-flooded pixels respectively, by decomposing the histogram of backscatter values. These distributions are next used to estimate for each pixel the probability of being flooded.

We further demonstrate that the sequential assimilation of these data sets into the 2D LISFLOOD FP hydraulic model of the Colorado River (Texas) can substantially reduce the uncertainty of the flood extent and water elevation predictions over subsequent time steps. The probabilistic flood maps derived from four Sentinel-1 images are sequentially assimilated into the model using a variant of the Particle Filter. The particle ensemble of flood extent maps coinciding with the satellite overpasses are generated by propagating inflow and rainfall uncertainty through the flood prediction system. The results of the assimilation experiment are evaluated using independent optical satellite imagery as well as water level recordings from a gauge station along the stream.

We show that the proposed approach holds promise for improving flood forecasts at locations where observed data availability is limited, but satellite coverage exists.