Comparative study on the selection criteria for fitting flood frequency distribution models with emphasis on upper-tail behavior