Modeling nonstationary extreme value distribution using large scale climate indices over South Korea

As the influence of anthropogenic climate change on hydrological extremes such as heavy rainfalls and floods has become a significant issue, nonstationary extreme value modeling for hydrological variables has been studied extensively. Large-scale climate indices are widely used as the covariates in extreme value modeling because they could physically account for the effect of climate variability. This study aims to perform nonstationary frequency analysis for extreme rainfalls over South Korea using appropriate large scale climate indices. Based on the correlation between the long term trend of daily annual maximum rainfall data and various climate indices, significant three climate indices were selected. The combinations of the selected climate indices were used as covariates of location and scale parameters in the generalized extreme value (GEV) model. A total of 59 candidate GEV models including stationary GEV and nonstationary GEV with time covariate were considered as candidate models. Considering both the goodness-of-fit and uncertainty measure using the rescaled Akaike information criterion (rAIC) and nonparametric bootstrap method, respectively, an optimum GEV model was finally selected. The performance of the selected model was assessed by model diagnostics.