Bayesian Changepoint Analysis of the Annual Maximum of Daily and Subdaily Precipitation over South Korea

Abstract

Abstract Bayesian changepoint analysis is applied to detect a change point in the 30-year (1976–2005) time series of the area-averaged annual maximum precipitation (A3MP) for the six accumulated time periods (1, 3, 6, 12, 24, and 48 h) over South Korea. Using noninformative priors, Bayesian model selection is performed by posterior probability through the Bayes factor, and the exact Bayes estimators of the parameters and unknown change point for the selected change model are obtained. To investigate the significance of the mean differences in the six A3MP between before and after the change point, posterior probability and 90% highest posterior density credible intervals are examined. The results show that a single change occurred around 1997 in the A3MP without regard to the accumulated time periods over South Korea. This is strongly consistent with the abrupt increases in the intensity and frequency of heavy precipitation after 1997. The A3MP after the change point (1997) significantly increased more than 15% compared with the A3MP before the change point. The intensification of A3MP resulted in a great increase of the annual total precipitation (about +18%), especially a greater increase of the heavy precipitation amount (+51%) and frequency (+48%) over South Korea.