Flood Susceptibility Mapping Using Optimized SVM Algorithm

This study describes optimization of Support Vector Machine (SVM) using meta-optimization algorithms including the Particle Swarm Optimization (PSO), Genetic Algorithm (GA) and Ant Colony optimization for flood susceptibility mapping at Zahedan Basin. To determine the best factors among the 19 factors used in previous studies and the relative importance of input variables, two experiments of Information Acquisition Ratio (IGR) and multilinearity of parameters were used soil moisture content factor which is identified as a highly effective factor (IGR=0.767),  is used directly in this study for the first time. The flood inventory map was prepared from Sentinel-1 satellite data processing and validated. The generated data set was used to map the flood prone areas using SVM model optimized with the mentioned algorithms. For the first time, in this research, ACO algorithm was used to optimize SVM model. The prediction accuracy of the models was evaluated using statistical criterion. All three algorithms used in this study improved SVM performance by 4% in average, while the PSO-SVM model had the best performance among the others with AUC=0.953, MSE=0.025, RMSE=0.158 and StD=0.405.