ALGORITHMS
Our main proposition is that instead of just one flood mapping algorithm, we will implement an ensemble approach integrating three robust, cutting edge algorithms developed independently by three leading research teams, and accept a pixel as flooded when at least two algorithms classify it as water. The motivation for choosing such an ensemble approach is to substantially improve the robustness and accuracy of the derived Sentinel-1 flood and water extent maps and to build a high degree of redundancy into the production service. We expect this to be true from the very beginning of the service, but certainly along the way of the operations, with its continuous maintenance and optimisation activities and major evolutions every few years. This is because the (internal) availability of three separate flood and water extent maps will help the development team to readily identify algorithmic shortcomings and/or bugs in the implementation of each of the three algorithms and mitigate any problems that may appear, thereby not only improving the quality of the service but also speeding up the initial set-up of the service and reducing its costs in the long term. In the frame of the quality assurance procedure, the ensemble approach will further allow differentiating between classification errors that can be attributed to shortcomings of individual algorithms and errors that are inherent to the SAR sensing instruments and their difficulty to capture the appearance or disappearance of surface water in particular situations. The user, on the other hand, will only receive “consensus flood maps” where two or even all three cutting-edge algorithms agree, giving her or him the extra confidence needed to use the fully-automatic Sentinel-1 flood data product instead of manually derived flood maps.
