Changepoint Detection for Ecosystem Processes

In practice, there is a need to account for time-varying changes in ecosystem processes that affect fisheries productivity. While there is ample ecosystem modeling theory and more is welcome, we submit that empirical changepoint detection are rapidly developing, tractable to interpret, and can be employed now to refine hypotheses of ecosystem dynamics. Changepoint detection is an algorithmic approach to testing whether a set of ecosystem parameters have similar patterns through time. Do the mean, variance, or both mean and variance of the process change? For example, recent meta analyses of stock-recruitment dynamics have shown that, for many stocks, regime-shift patterns in recruitment production exist. This contradicts the stationarity assumption often made for analyses of recruitment and more importantly for setting biological reference points. We briefly some review statistical models for changepoint detection. We apply the R-Package changepoint to assess concordance in changepoints for the Pacific Decadal Oscillation, the Oceanic Nino Index, and the three primary tuna stocks in the Western Pacific Ocean, skipjack, yellowfin, and bigeye tuna. Results suggest common tipping points in time exist for these series. We note that changepoint detection methods can help improve understanding and inform predictive models for fisheries management.