Abstract
This study provides the results of flood frequency analysis adopting a regional approach using annual maxima?s of peak flows (APF) of eight catchments located on various small rivers of KhyberPakhtunkhwa, Pakistan. Initial screening reveals that the recorded data of APF for all catchments are independent, random, free from significant trend and identically distributed. L-moments based heterogeneity measure indicates that the study region is homogeneous. The results of |Z-Dist| statistic and L-moment ratio diagram being goodness of fit measures are in favor of Generalized Pareto (GPA) distribution among five candidates of regional distribution. For the ungauged sites, flood quantiles have been estimated through OLS regression and artificial neural networks (ANN). The estimated quantiles using ANN method are relatively accurate compared to OLS regression. The historical assessment indicates that quantile estimates obtained through ANN and index flood method are close to the highest recorded APF values for shorter as well as longer return periods for each site.