Methods for detecting regime shifts in large marine ecosystems: A review with approaches applied to North Pacific data


Mantua, N.J. 2004. Methods for detecting regime shifts in large marine ecosystems: A review with approaches applied to North Pacific data. Progress in Oceanography 60(2-4):165-182.


This paper provides a brief review of five analytical methods previously used or promoted for diagnosing regime shifts in marine ecosystems. The methods discussed are:

(i) Principal Components Analysis,

(ii) compositing Average Standard Deviates,

(iii) Autoregressive Moving Average and Intervention Analysis modeling,

(iv) VectorAutoregressive Process modeling, and

(v) Fisher Information.

Assessments of some the relative strengths and weaknesses for the different analytical approaches are also offered. Some of these methods are applied to a collection of fishery oceanographic time series for the North Pacific to illustrate aspects of their relative utility and limitations for diagnosing regime shift behavior.

One recommendation for future studies is to analyze biotic and abiotic time series separately in order to identify ecosystem state variables of interest and to better isolate ecosystem behaviors from other influences like environmental change. Methods that allow for quantitative assessments of the statistical significance of hypothesized regime shifts should be favored over those that do not.

Analyses of especially large collections of time series may benefit from first using a data compression technique and then applying one of the methods that are more appropriate for just one or a small number of time series. Because of the difficulties in observing and adequately documenting many aspects of marine ecosystem variability, it is crucial that future research attempts to combine empirical studies of large marine ecosystems with theoretical and modeling studies of other systems for which the dynamics and predictability are better understood. With such a comparative approach it should be possible to refine conceptual and simulation models while also identifying crucial gaps in existing observations.