A new intelligence approach based on GIS-based Multivariate Adaptive Regression Splines and metaheuristic optimization for predicting flash flood susceptible areas at high-frequency tropical typhoon area

The main objective of this research work was to propose and verify a new soft computing approach based on Multivariate Adaptive Regression Splines (MARS) and Particle Swarm Optimization (PSO) for spatial prediction of flash flood susceptible areas. A high frequency tropical typhoon area located on Northwest of Vietnam was selected as a case study. For this purpose, a GIS database for the study areas was prepared, including 654 flash-flood inundations and 12 influencing variables (elevation, slope, curvature, toposhade, aspect, topographic wetness index, stream power index, stream density, Normalized Difference Vegetation Index, soil type, lithology, and rainfall), which were compiled from various sources. The database was used to build and verify the prediction model. We assessed the model's performance through various indices including Classification Accuracy Rate, Area under the Curve (AUC), Precision, and Recall. We also compared the model's usability with five state-of-the-art machine learning techniques including the Backpropagation Neural Network, Support Vector Machine, and Classification Tree. The results revealed that the hybrid PSO-MARS model outperformed other benchmark models in all the employed statistical measures. We conclude that the proposed model can be particularly suited for flash flood forecasting problems at high frequency tropical typhoon area.