A Penalized Spline Model Based Nonstationary Regional Rainfall Frequency Analysis in the Bayesian Framework

Many countries are facing the challenge of managing water resources due to the recent increase in climate variability under climate change. Rainfall frequency analysis has been commonly applied to evaluate design rainfalls for the regions where the long-term observed rainfall data are readily available. Rainfall frequency analysis provides the hydrological quantities corresponding to the return period under the assumption of stationarity. However, most historical records showed marked nonstationarity in rainfall extremes, so the nonstationary frequency analysis approach has been recommended to better fit the extremes. This study proposes a novel approach to the existing nonstationary frequency analysis for especially estimating trends from a regional penalized spline model in a Hierarchical Bayesian modeling framework. Here, a regionally integrated trend across stations is obtained by simultaneously estimating a set of parameters associated with spline models and frequency analysis. The proposed models are compared with a simple linear model-based approach. The preliminary results confirmed that the partial pooling approach based on the hierarchical Bayesian model could better detect regional trends reliably and reduce the associated uncertainties, especially for the regions where long-term data were not available.

[Acknowledgement] This work was funded by the Korea Meteorological Administration Research and Development Program under Grant KMI 2018-07010.