Detection of regime shifts in the environment: testing “STARS” using synthetic and observed time series

Abstract

Abstract While marine populations change all the time, sometimes regime shifts involve an entire ecosystem, resulting in crucial and sometimes permanent alterations in the ecosystem trophic web and services. A commonly used method to detect shifts in marine systems is the Sequential t-test Analysis of Regime Shifts (STARS). In this work, we chose to analyse the limits and performance of STARS because of its free open-source software and wide use. For the first time, we tested the STARS algorithm using synthetic time series and autoregressive integrated moving average time series, designed to resemble natural observations. We then applied the information obtained from these tests to investigate the STARS detections on an observed time series, that of Calanus finmarchicus in the North Sea. Our tests indicated that in no experiments did STARS detect 100% of the artificial change points at the exact time of the shift. In most cases, STARS tended to anticipate the shift by a few time units. Overall, we determined STARS to be a good method to detect shifts in observed natural time series, so long as the exact time of the shift is not necessary and the possibility of false positives is taken into account.