Covariates for nonstationary modeling of extreme precipitation in the Pearl River Basin, China

Nonstationarity of extreme precipitation has been reported in previous research. As a consequence, the hydrological frequency analysis that is primarily used in water-related infrastructure design in a nonstationary context is a current interest in hydrology. Covariates are critical elements in the nonstationary modeling of hydrological extremes. However, a covariate can be any physical process that exerts influences on hydrological extremes and it is important to find the best covariates for modeling the nonstationarity. Furthermore, the significance of and the spatial differences among the best covariates have yet to be investigated. Thus, in this study, the Pearl River Basin (PRB) was used as the study area and the wet/dry season maximum daily precipitation (WMP, DMP) data were used to determine the best covariates for the extreme precipitation based on the generalized extreme value (GEV) theory. The significance of the best covariates and the spatial differences between them were also investigated. The results showed that the best covariates for extreme precipitation exhibited large differences in spatial distribution. Furthermore, the El Nino-Southern Oscillation (ENSO) was found to be the best covariate for both the WMP and DMP in most of the PRB. The results also showed that there were differences in the performances of different indices used to represent the same climatic factor and significance test should be performed for the best covariates because the best covariate may not be a significant covariate. In addition, significant uncertainties were found in the nonstationary modeling of extreme precipitation due to the introduction of covariates.