The introduction of Sentinel-1 and -2 satellite data in the second half of the last decade provides the opportunity for monitoring of land cover with a high temporal and spatial resolution over large areas. The operative monitoring of temporary inundated areas on the Great Hungarian Plain is a challenge that has kept researchers and engineers in Hungary busy for over half a century. The inundations, called belvíz in Hungarian (Inland excess water, ponding, areal flood or surface water flood are expressions used in English) are the phenomenon where large areas are temporary covered with surplus surface water due to the lack of runoff, insufficient absorption capability of soil or the upwelling of groundwater.
We developed an operational system that can be used to monitor inland excess water on a weekly basis at national scale. The methodology is fully based on freely available high resolution optical and radar satellite data and is completely automated using python scripts.
The preprocessing consists of an automated procedure to download daily Sentinel-1 and -2 data from ESA Sentinel Data hub. The Sentinel-2 Level-2A data is mosaiced to the original swath and all bands are resampled to 10 meter. Subsets of the total swath can be created to reduce the area for calculation of the IEW maps.
In the first stage of the IEW detection, the multispectral Sentinel-2 data is used to create Modified Normalized Difference Water Index (MNDWI) maps. Based on training data derived from known permanent water bodies, a threshold is calculated to slice the MNDWI maps into binary water and no water maps. The ISODATA algorithm is used to cluster the Sentinel-2 bands into a large number of classes. The spectral distance of each of the class is compared with the reference water pixels that have been extracted from the same image based on the training data. The class that is closest to the reference is determined “water”; all other pixels are designated “no water”. In case of images with limited cloud cover both optical data based approaches provide good results, but since IEW usually occurs during bad weather conduction it is rarely enough to just use optical data for the detection of the inundations. In order to improve the detection, we added a second stage to the workflow based on active data. Sentinel-1 GRD images are collected from the same area as the Sentinel-2 data. Common preprocessing, including orbit correction, calibration, thermal noise removal, speckle filtering and terrain correction is performed on the images to prepare them for the IEW detection algorithm. Similarly to the MNDWI procedure, statistics of training water areas are extracted from the Sentinel-1 image and the upper and lower boundary of water is determined. These boundaries are then used to select all water pixels in the image.
Since Sentinel-1 and Sentinel-2 data from the same area is not collected on the same day, an algorithm was developed to combine all the individual IEW maps that are calculated within one week from the area under consideration. The algorithm tests for each pixel how many times within that week water was detected, and if the detection rate is above a predefined value, the pixel is considered “water”. If it is below the threshold, it is “no water”. The algorithm detects all water in the area and is not able to distinguish between permanent water and inland excess water, therefore a mask is applied to remove all permanent water from the final result.
We validated our results with high resolution satellite data and aerial photographs and found a high correlation with large IEW patches. Smaller patches are difficult to detect due to the resolution of the input data. The algorithm also shows reduced accuracy if only one type of source data is available. If only optical data is available, cloud and cloud shadows cause problems. If only radar data is available, the algorithm is overestimating the amount of water due to dark speckle. The current algorithm can be extended with other satellite data sources, and we also plan to evaluate the use of more advanced machine learning algorithms.

The registration of inland water covered areas and monitoring of their extent fluctuation in time is a crucial component for status analyses and scenario examination in ecology and environmental monitoring. Water utilities companies show high interest in the annual hydrological cycles of the open surface water reservoirs that they use for the production of drinking water and their sudden changes. Efficient and timely monitoring is required. Usually this task is being treated with in situ measurement stations, or field trips, generating data and observations that are then coupled with a locally installed decision support system. Resources depleting methods may be replaced or complemented by spaceborne EO data to provide a cost-effective solution for frequent and accurate monitoring of the water extent.
Numerous approaches have been proposed to perform inundation mapping by assimilating spaceborne data. These rely mostly on optical or radar data. Still spaceborne image analysis reaches its limits due to their temporal and spatial resolution. Moreover, possible non favorable atmospheric conditions may hinder inundation map derivation; consequently, hydroperiod estimations, especially in extended periods of cloud coverage. Facing the challenge WQeMS takes advantage of and adjusts a set of tools that were developed in H2020 ECOPΟTENTIAL project to provide water utilities companies with the adequate services and products that can be incorporated into their decision support systems.
The WaterMasks inundation mapping module [1][2], relies on the physics of light interaction with water and water with emerging vegetation to estimate the inundation extent from radiometrically corrected Sentinel-2 (S-2) data. It implements a novel automatic local thresholding approach for the classification of an area into the water and land classes. WaterMasks inundation maps achieve good results in term of accuracy and the approach is validated for its transferability to other areas. While results produced using S-2 data are very good in terms of accuracy, they exhibit one important limitation. Especially in areas with frequent and extended cloud coverage, accurate hydroperiod maps cannot be produced due to the very high temporal distance between suitable data. To combat this a novel machine learning approach for the fusion of S-2 and Sentinel-1 (S-1) data [3] was devised to be able to retrieve credible inundation maps even under cloudy conditions. This way the time step, upon which hydroperiod maps shall be generated may remain the same. This pixel centric methodology relies on inundation maps created from S-2 data to be used as reference data in the training process. A swarm of Sentinel-1 images, timely coinciding with the S-2 reference, are used to extract the features that the classification models will be trained upon. Results achieve good accuracy, with it improving substantially when the mean day distance is less than 30 (approx. 6 sequential S-2 image acquisitions). When cloud coverage exceeds the time windows of 30 days, results can be doubtful.
The scope of this study is to further expand on the conclusion derived from the application of the fusion approach regarding the mean day distance (mdd) of 30 days. Initially, results about the transferability of the fusion method to new areas with different geomorphological characteristics are presented. Then a discussion is carried out towards the balance that shall be achieved between the number of S-1 products to incorporate in the production of the hydroperiod vs. the accumulative loss in accuracy that these products bring into the hydroperiod estimation.
Polyphytos water reservoir is selected for testing, being the Greek pilot area of the WQeMS – “Copernicus assisted Water Quality emergency Monitoring Services” project. For the creation of the validation layers, Very High Resolution (VHR) Data were used in combination with maps and other relevant in-situ data provided by the user group (local stakeholders). Results will be presented that showcase the capacity that the approach introduces for water extent monitoring and its benefits vs. business as usual.

[1] G. Kordelas, I. Manakos, D. Aragones, R. Diaz-Delgado, J. Bustamante, Fast and automatic data-driven thresholding for inundation mapping with Sentinel-2 data , 2018, Remote Sensing, 10, 910, DOI: 10.3390/rs10060910.
[2] G. Kordelas, I. Manakos, G. Lefebvre, B. Poulin, Automatic Inundation Mapping Using Sentinel-2 Data Applicable to Both Camargue and Doñana Biosphere Reserves, 2019, Remote Sensing Journal, 11(19), 2251, DOI: https://doi.org/10.3390/rs11192251
[3] I. Manakos, G. Kordelas, K. Marini, Fusion of Sentinel-1 data with Sentinel-2 products to overcome non-favourable atmospheric conditions for the delineation of inundation maps", 2019, European Journal of Remote Sensing, DOI: 10.1080/22797254.2019.1596757

The first radar altimetry missions were dedicated to the open ocean. However, continental water surfaces (enclosed seas, lakes, rivers, flooding areas...) can also be measured by satellite altimetry. For many years now, satellite altimetry is increasingly used to monitor inland waters all over the globe, and even more with the advent of delay doppler radar altimeter embedded on the Copernicus Sentinel-3 and Sentinel6-MF missions, and the future SWOT mission based on interferometric radar imagery. For these instruments, new algorithms are currently being developed to support improved data processing over hydrological surfaces in order to achieve significant accuracy improvements. There is therefore an increasing need for new in-situ systems to provide reference data for large-scale Calibration/Validation (Cal/Val) activities over inland water.

In this context, vorteX.io designed a lightweight remote sensing instrument, inherited from the specifications of radar altimeters on board altimetric satellites, capable of providing water height measurements with centimeter-level accuracy and at high frequency. Mounted on a flying drone, the system combines a LiDAR system and a camera in a single payload to provide centimetre-level water surface height measurements, orthophotos, water surface mask and water surface velocity throughout the drone flight. The vorteX.io system is the result of a review of existing in-situ systems used for Cal/Val of satellite altimetry in hydrology or operational monitoring of water heights (often to anticipate potential river floods or to monitor reservoir volumes). As the lightweight altimeter is inspired from satellite altimetry, water level measurements are directly comparable to satellite altimeter data. Thanks to the UAV capability, water measurements can be performed on long distances along rivers, and at the same location and time as the satellite pass. New hydrological variables are planned to be added in the next future (water surface temperature, river discharge, turbidity, …).

The drone-embedded lightweight altimeter has been successfully used during several measurement campaigns for the French space agency (CNES) as part of the Cal/Val of Sentinel-3A, Sentinel-3B, and Sentinel-6 missions. This innovative instrument is being considered as one of the means of the in-situ validation of the future SWOT mission for hydrology. We present here the results of the measurements performed by the vorteX.io VTX-1 altimeter in different hydrological contexts in France in 2020 and 2022.

Our environment and society are affected by climate change in many ways. Amongst others, the intensification of the hydrological cycle is resulting in more extreme drought and precipitation events, leading to an increase in the frequency and intensity of flood events. Although heavy monsoons, hurricanes and cyclones long seemed far-away phenomena for Western Europe, this impact also became painfully clear in that region during the summer of 2021.

Synthetic Aperture Radar (SAR) is particularly suited to monitor floods from space, thanks to its ability to penetrate clouds and its independence of an external illumination source. The Copernicus program has boosted the field of SAR remote sensing with the launch of the Sentinel-1 constellation, the first SAR sensors to provide free imagery and global coverage. Throughout the past years, many SAR-based algorithms for flood mapping have been developed, ranging from single scene to time series based and from manual to highly automated approaches. Typically, a high degree of automation and global applicability are pursued, to enable fast mapping independent of the flooded region. However, by doing so, locally available information might not be fully exploited.

The TerraFlood algorithm (Landuyt et al., 2021) was originally developed with and for the Flanders Environment Agency, responsible for operational water management in the Flanders region. The algorithm combines hierarchical thresholding and region growing, both on the pixel and object level. Aiming to fully exploit locally available data, it requires a SAR image pair (containing a flood and pre-flood image) and several ancillary data layers, including elevation, land cover, and flood risk, as input. The output map discriminates permanent water, open flooding, long-term flooding, possible flooding, flooded vegetation, and possibly flooded forests from dry land. Invisible forested areas, forested areas for which the flood state is unknown, are indicated too.

The algorithm’s accuracy and robustness, both for emergency mapping and automated monitoring, were assessed based on maps of 36 flood events that occurred between 2016 and 2020. Besides near-real time flood maps, also reduced products like flood recurrence maps can be provided. End users are provided free and easy access to all products through the Terrascope (1) platform. The algorithm immediately proved its value during and after the summer 2021 floods. As these floods hit hard in the Walloon region, maps were also provided to and used by several Walloon institutions. Feedback from both end users will be used to further improve the algorithm and service. Additionally, potential improvements for vegetated and dense urban areas, the two main pitfalls of the TerraFlood algorithm and Sentinel-1 imagery in general, remain under investigation.


(1) Terrascope (www.terrascope.be) provides analysis-ready satellite data and derived products. Registered users can access Sentinel-1, -2 and -5P and PROBA-V data as well as land cover, vegetation indices and elevation products.

L. Landuyt, F. M. B. Van Coillie, B. Vogels, J. Dewelde and N. E. C. Verhoest, "Towards Operational Flood Monitoring in Flanders Using Sentinel-1," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 14, pp. 11004-11018, 2021.

An unprecedented number of people are currently living with the risk of catastrophic flooding. As the population density in the floodplains increases, along with a corresponding increase in the frequency and magnitude of floods resulting from climate change impacts, the flood-vulnerable population estimates will only increase in the future. Accurate flood inundation information is thus, not only vital to efficiently manage rescue and response operations, but also bolster preparedness for future flooding to reduce economic losses. Satellite remote sensing provides a cost-effective solution to provide a synoptic view of flood-affected areas at both large and local scales. Specifically, Synthetic Aperture Radar (SAR) sensors with their weather and solar illumination independent imaging capabilities are uniquely suited to observing flooded areas, which are often covered by thick clouds making the use of optical sensors rather challenging.
SAR data are, however, affected by a myriad of uncertainties, which makes the estimation of uncertainties in the flood maps as well as assessing the quality of the classification strategy, absolutely critical to ensure reliability of the mapping. For example, the accuracy of SAR-based flood maps are strongly influenced by the inundated land cover type, given the high level of surface roughness sensitivity for microwaves. Another important factor which makes flood map evaluation particularly challenging, is the fact that they typically result in binary classifications, rendering most metrics sensitive to class prevalence almost unusable. Despite several calls to action from members of the scientific community and literature showing the impact of poorly chosen evaluation metrics on downstream decision-making, the validation strategies for flood mapping have not significantly evolved in the last decades. One of the key reasons could be the lack of convincing alternative metrics and an incomplete understanding of the possible consequences of failures in diligent accuracy assessments.
The present study seeks to identify the impact of using the currently prevalent evaluation metrics in flood mapping literature for map comparison (for instance, when choosing the best between several different classifiers) and recommend best practices for binary flood map evaluation. The performance of several machine learning classifiers (Random Forest, Classification and Regression Trees, and Support Vector Machines) for Sentinel-1 based flood detection was evaluated using diverse test cases. Confusion matrix based standard metrics (e.g. Overall Accuracy, Critical Success Index) were used and compared with alternative strategies to demonstrate the challenges of using currently popular objective functions for binary classifications. Alternative strategies, included prevalence-aware validation data sampling and land-use based error characterization. The flood map accuracy was evaluated against concurrent cloud-free Sentinel-2 based water masks, and an expert classified flood map based on multiple data sources and manual cleaning. The expert map was used as a benchmark to examine the quality of the Sentinel-2 data, specifically for binary flood map validation purposes.
Results indicate that the error characteristics of flood maps are determined by a variety of underlying factors which are not sufficiently captured by the objective functions currently in use. There is an urgent need to reassess how mapping accuracy is determined for satellite-based flood extents, particularly due to the rapidly increasing dependence on Earth Observation for flood damage estimation and subsequent insurance payout triggers. The outcomes from this study pave the way for more rigorous and statistically robust accuracy assessments, ultimately benefiting all stakeholders through increased reliability of satellite-derived flood extents.