Evaluation and modeling of flood risk in Kal-e Shur Sabzevarbasin using Machine learning algorithms

The present study aims to compare the efficiency of Machine learning algorithms including BaggingbasedRough Set(BRS), Recurrent Neural Network (NNETR), Boosted Regression Tree (BRT), to evaluateflood susceptibility and identify vulnerable areas in Kal-e Shur basin, so that the most critical factorsaffecting flood occurrences in the catchment area can be identified and investigated. This study proposed ahybrid Flood Susceptibility Mapping (FSM) framework based on ۳ learning models. First, the flooddistribution map with ۲۵۵ points was prepared, and then, the points were classified in a ratio of ۷۰ to ۳۰ fortraining and validation. Among the ۱۹ parameters effective in the occurrence of floods in the basin, nineparameters (slope, land use/cover, lithology, distance to river, elevation, drainage density, and Slope LengthFactor (SL-Factor), precipitation, and soil are recognized as essential factors. Among the natural parameters,loose and permeable formations, areas without vegetation, the concavity of the ground surface and upstreamrunoff are the most effective of them. The integration of learning algorithms indicated the high efficiency ofthese algorithms in determining the flood-susceptible zones with high flood risk. Identifying flood-proneareas and providing effective flood control and management strategies are essential measures to reduce flooddamage.

کلیدواژه ها:

Machine learning ، BRT ، Risk susceptibility ، Kal-e Shur basin ، Flood.