Developing an Ensemble Machine Learning Approach for Enhancing Flood Damage Assessment

Abstract

Climate change has caused fundamental changes in the pattern of rainfall worldwide. Climate change can alter precipitation patterns, consequently intensifying the frequency and severity of flash floods in specific regions, including Iran. It is important for communities to be prepared for these events and to take steps to mitigate their impact. Full control or damage management of the resulted floods through structural measures is not always feasible due to economic, technological, environmental and social limitations. Therefore, often non-structural measures play an important role in reducing probable damages and casualties. The significance of advanced systems for both short- and long-term flood forecasting cannot be overstated. In this article, short-term flood prediction model is discussed using Ensemble Prediction Systems (EPSs) Machine Learning algorithms (ML) and HEC-HMS hydrological model. Also, in order to achieve high accuracy in the assessment of flood-damaged areas, remote sensing techniques have been used. The results show that the use of EPS improves the speed and accuracy of the daily prediction model (R² = 0.8). Also, with the use of Sentinel-1 radar satellite images and the simultaneous use of supervised learning algorithms, a suitable estimate of the evaded area has been made for seven selected floods in the Kan basin, which is a mountainous region in the north of Tehran, in 2015–2022 period.