A Novel Decision Support System for Early Flood Warning Using Cellular Automata and Support Vector Machine

Urban flood warning systems are designed to monitor flood propagation on floodplains. These early warning systems are based on accurate runoff and flood spreading prediction, and use (a) history rainfall-runoff information and (b) digital elevation data to predict water depth and flood extent in urban areas. However, these traditional decision support systems (DSSs) cannot be applied to real time processes due to the computational cost required for running two dimensional (2D) hydrodynamic models. In this paper, we propose a novel Support Vector Machine (SVM) ĝ"€ Cellular Automata (CA) based DSS, to deliver (near) real time flood warnings for flood defence managers and public authorities. The proposed DSS was tested and validated on Thamesmead, London. The results in this study indicated that the SVM model accurately predicted discharge compared to measured data, and the CA-based model using irregular big cells showed that the predicted water depth and flood extent were comparable to a commercial 2D hydrodynamic model's simulated results and the predictions only took less than 3 seconds to run for the whole floodplain. The proposed real time DSS could save thousands of lives for high risk flood zones.