Comparison of piecewise linear change point detection with traditional analytical methods for ocean and climate data

The Earth’s atmosphere and oceans are largely determined by periodic patterns of solar radiation, from daily and seasonal, to orbital variations over thousands of years. Dynamical processes alter these cycles with feedbacks and delays, so that the observed climate response is a combination of cyclical features and sudden regime changes. A primary example is the shift from a glacial (ice age) state to interglacial, which is driven by a 100-thousand year orbital cycle, while the transition occurs over a period of hundreds of years. Traditional methods of statistical analysis such as Fourier and wavelet transforms are very good at describing cyclical behavior, but lack any characterization of singular events and regime changes. More recently, researchers have tested techniques in the statistical discipline of change point detection. This paper explores the unique advantages of a piecewise linear regression change point detection algorithm to identify events, regime shifts, and the direction of cyclical trends in geophysical data. It evaluates the reasons for choosing this particular change detection algorithm over other techniques by applying the technique to both observational and model data sets. A comparison of the proposed change detection algorithm to the more established statistical techniques shows the benefits and drawbacks of each method.