The impact of using different probability representations in application of equidistant quantile matching for bias adjustment of daily precipitation over the Daqing River Basin, North China

Abstract

This study focuses on how the representation of daily precipitation probability distributions may affect the application of equidistant quantile matching (EDCDFm). Five representations, that is, two parametric distributions and three nonparametric approaches, are selected. The parametric distributions include Gamma and Weibull while the nonparametric approaches include empirical cumulative distribution function (ECDF) and kernel density estimation method (KDE) as well as an improved KDE named diffusion-based kernel density estimation (DKDE). All five methods were applied to correct the daily precipitation of 11 stations over the Daqing River Basin, North China in 1981-2015. The results demonstrated that DKDE is closer to ECDF than the other three methods in the goodness-of-fit evaluation. Furthermore, nonparametric methods present advantages over parametric methods; especially, DKDE and ECDF are skilful equally and both of them display impressive comprehensive performance than other methods by multi-index. These findings suggest that representing the precipitation probability distributions accurately is beneficial for the bias correction and selecting one or more suitable indexes is of great significance to the verification of EDCDFm. This study can provide guidance for future water resources management of the Daqing River Basin.