Flash flood susceptibility modelling using advanced machine learning algorithms. Case study of the Rheraya watershed,Morocco

Abstract

<p>A flash flood is a very rapid and surprising rise in water level affecting any part of a watershed. It is considered to be one of the most dynamic natural disasters for which action should be taken to minimize economic damage, negative effects and consequences. Flash flood susceptibility maps are important for a range of applications, including land use planning and the development of risks reducing strategies and early warning systems.</p><p>In the present study, the Rheraya watershed in Morocco was chosen to develop the modeling of flash flood susceptibility and subsequently to detect the areas most vulnerable to flooding.</p><p>For the development of this work, new advanced machine learning algorithms were adopted. Thus, four models were applied: Naïve Bayes (NB), K Nearest neighbor (KNN), Extreme Gradient Boosting (XGB) and Random Forest (RF). Twelve independent topographic, climatic and geo-environmental variables (elevation, slope, aspect, curvature…) and one dependent variable (flash-flood inventory) were used as inputs to our flash flood sensitivity models.</p><p>The results showed that RF is the most optimal model with an area under the curve (AUC) value of 0.86. The other AUC values of the models, i.e., XGB, NB, KNN, were 0.85, 0.76, 0.76 respectively. In the optimal model, areas with very high susceptibility to flash floods are well localized and specific, which will play an essential role in the planning and implementation of flood mitigation strategies in the region.</p><p><strong>Keywords: </strong>Flash floods, Rheraya watershed, Machine Learning, GIS, remote sensing.</p>