Bayesian analysis to detect abrupt changes in extreme hydrological processes

- Define the regression model for change point analysis for extreme data.
- Construct a selection prior distribution.
- Obtain posterior samples of the regression model regularized by priors.
- Analyzing the posterior mean for detecting change points.

SummaryIn this study, we develop a new method for a Bayesian change point analysis. The proposed method is easy to implement and can be extended to a wide class of distributions. Using a generalized extreme-value distribution, we investigate the annual maximum of precipitations observed at stations in the South Korean Peninsula, and find significant changes in the considered sites. We evaluate the hydrological risk in predictions using the estimated return levels. In addition, we explain that the misspecification of the probability model can lead to a bias in the number of change points and using a simple example, show that this problem is difficult to avoid by technical data transformation.