Climate-informed clustering based nonstationary regional extreme flood events spatio-temporal evolution using hierarchical Bayesian modeling

Abstract

Study region: Xiangjiang River basin, is at the middle and lower Yangtze River. Study focus: Increasing evidence shows that large-scale climate indices are being linked to the spatio-temporal evolution of extreme flood events in many regions. It is widely recognized that identifying the response of flood risks to climate factors is investigated by nonstationary local hydrological frequency analysis (NLFFA). However, high uncertainty by utilizing NLFFA usually cannot achieve the accuracy for engineering design, creating need for nonstationary regional flood frequency analysis (NRFFA) for reducing uncertainty. This study presents a NRFFA for extreme flood risk under changing environment that is conditioned on the large-scale climate index. The nonstationary regional analysis is developed in a hierarchical Bayesian framework to support partial and full pooling, which can enhance climate-based regionalization skill as well as retain site-specific characteristics. Simultaneously, the modeling structure is designed to consider the spatial dependence across the basin. New hydrological insights for the region: Regional models for whole basin and climatically homogeneous sub-region respectively, and local models are compared: the regional models exhibit a great benefit with the estimates uncertainty reduction and allows a better quantification of large-scale climate effect on flood extremes. Moreover, the regional partial pooling model for homogeneous sub-region indicate that the regional model can be further improved by considering a prior pooling stations formed by similar significant climate response.

Keywords

• Nonstationary regional extreme flood events
• Spatio-temporal evolution
• Hierarchical Bayesian model
• Large-scale climate index
• Xiangjiang river basin