Bayesian change point analysis for extreme daily precipitation

Abstract

Change point (CP) analysis of extreme precipitation plays a key role to incorporate non-stationarity in flood predictions under climate change. This article provides a Bayesian method to detect the CP frequently appearing in extreme precipitation data. Unlike most published work based on a normal distribution, we allow for the model to follow a generalized Pareto distribution to fit extreme precipitation over a high threshold with a CP, which can effectively utilize tail behaviour of the distribution. The Bayesian CP detection is investigated on four models: a no change model, a shape change model, a scale change model, and both a shape and scale change model. Model selection is performed using the Bayes factor and model posterior probability; the posterior means of the unknown CP and the model parameters before and after the CP can be obtained based on the selected CP model. Simulation studies and a real data example are provided to demonstrate the proposed methodologies. Finally, model uncertainty issues in the frequency analysis are extensively discussed. It is found that considering the abrupt and sustained CP in extreme precipitation is important when performing hydraulic or hydrologic design.