179 – Bayesian Modeling in Physical Sciences and Engineering
Detecting Change-Points Using a Bayesian Approach for Climate Data
Andrew Bartlett
Southern Illinois University Edwardsville
Extreme weather and climate events such as hot spells, snow storms, and floods have recently had a major impact on the economy, environment, and human well-being. Thus, acting as a catalyst for concern about whether or not the climate is actually changing. One challenge when scientifically trying to determine whether or not the climate is actually changing is a change-point. A change-point is defined as any abrupt change or shift in the distribution and is the single most important contributing factor for inaccurate or accurate results. Traditional change-point methods focus exclusively on detecting an alteration or a shift in the arithmetic mean. In this paper we present a Bayesian change-point detection algorithm for detecting change-points in climate data. We first develop the theory for a Bayesian approach using a hierarchical model to estimate the location and number of change-points within a climatic time series. We then discuss the implementation of our Gibbs Sampler algorithm to obtain posterior probabilities of the location of multiple change-points. We finally investigate the performance of our Bayesian change-point approach through comparison with a standard frequentist method. Both methods are applied to simulated and real temperature data collected from Chula Vista, California.