Non‐stationary modelling of extreme precipitation by climate indices during rainy season in Hanjiang River Basin, China

The extreme precipitation regimes have been changing as the climate system has warmed. Investigating the non‐stationarity and better estimating the changes of the extreme precipitation are valuable for informing policy decisions. In this study, two precipitation indices are employed to describe the extreme events, including maximum 5‐day precipitation amount (RX5day) and the number of very heavy precipitation days (R20). The generalized additive models for location, scale and shape (GAMLSS) is employed to characterize non‐stationarities in extreme precipitation events and related climate indices in 13 stations in the Hanjiang River basin (HJRB). Three models including stationary model without change (M0), non‐stationary models over time (M1) and non‐stationarity models with large‐scale climate indices (M2) as predictors, respectively, are considered to analyse occurrence rates of extreme precipitation. The optimal model and the significant predictors were selected by the Akaike information criterion (AIC). To investigate the main predictors at regional scale, the homogeneous subregions for precipitation extremes are identified by clustering analysis. Results indicate that: (a) the non‐stationarities of RX5day series and R20 series at all stations are identified in the HJRB; (b) extreme precipitation behaviour is significantly influenced by climate indices and non‐stationary model 2 to describe the changes of extreme precipitation is better than non‐stationary model 1, indicating the impact of large‐scale climate forcing on the changes of extreme precipitation regimes; (c) the HJRB can be categorized into three homogenous regions. The optimal distributions and the main predictors of extreme precipitation events in most stations of each subregion are basically the same; (d) the dominated climate indices influencing the extreme precipitation events are different in different regions and have regional patterns. The results highlight the modelling of extreme precipitation events under non‐stationarity conditions and provide information for developing strategies of mitigation and adaptation to climate change impacts on extreme precipitation. The optimal distribution and the main predictors of extreme precipitation events in most stations of the subregion is the same after regionalization by cluster analysis. The detailed information of the frequency of extreme precipitation can be better described by climate‐induced non‐stationary model using climate indices as