Regional flood frequency modeling: a comparative study among several data-driven models

Correct prediction of flood discharge is important to design the hydraulic structures, to diminish the danger of its failure, and to minimize the environmental damages of downstream. The present study aims to investigate the application of machine learning methods for regional flood frequency analysis (RFFA). To achieve this aim, a number of 18 parameters of physiographic, climatic, lithology, and land use for the upstream watershed of stations were considered. Then, the best regional probability distribution function (PDF) was determined through the Kolmogorov-Smirnov (KS) test in each station for estimation of flood discharge with a return period of 50 years (Q50). The best input combination and best length of training datasets were determined by Gamma and M tests, respectively. Finally, RFFA was performed using adaptive neuro-fuzzy inference system (ANFIS), support vector machine (SVM), and genetic expression programming (GEP). Log-Pearson type III was identified as the best regional probability distribution based on the KS test for estimation of Q50. Gamma test results indicated that parameter of the perimeter, basin length, form factor, and mainstream length were selected as best input combination for RFFA. Also, according to the M-test results, the best length of training and testing datasets, 68% and 32% respectively, was determined. RFFA results indicated that the SVM, ANFIS, and GEP models had "good" performance (Nash-Sutcliff coefficient, NSE, equal to 0.75, 0.74, and 0.75, respectively).