Nonstationary extreme flood/rainfall frequency analysis informed by large-scale oceanic fields for Xidayang Reservoir in North China

Climate is a primary driver for extreme rainfall and flood events. In this paper, the temporally changing flood risk associated with the annual maximum 30-day rainfall (30-day AMR) and the annual maximum daily inflow (AMDI) related to the Xidayang Reservoir catchment is analysed from a climatic context. This is the largest catchment in Daqing River Basin in North China and is highly prone to floods related to the East Asian summer monsoon. Two climate factors, the average May-June-July sea surface temperature anomalies in areas of the northern Indian Ocean and western Pacific Ocean, are identified. They show a high negative correlation with the AMDI and 30-day AMR. Bayesian nonstationary models are then developed for the AMDI and 30-day AMR using the climate predictors identified as covariates. We compared three types of models of AMDI and 30-day AMR: (1) time-invariant, (2) linear temporal trend and (3) climate informed, and found that the climate-informed models exhibit the best performance according to Deviance Information Criterion (DIC) and 90th percentile Bayesian coverage rate for both AMDI and 30-day AMR. A significant decreasing trend is identified in the AMDI and the 30-day AMR, which is found to be associated with the climate predictors. Leave one out cross validation (LOOCV) is used to demonstrate that these models have decent skill in predicting year-to-year variability in flood risk. This can help to provide flood/rainfall dynamic management measures for reservoirs in Daqing River Basin using information from before the beginning of monsoon season, thus facilitating adaptation to a changing climate.