Regime shifts are considered here to be low-frequency, high-amplitude changes in oceanic conditions that may be especially pronounced in biological variables and propagate through several trophic levels. Three different types of regime shift (smooth, abrupt and discontinuous) are identified on the basis of different patterns in the relationship between the response of an ecosystem variable (usually biotic) and some external forcing or condition (control variable). The smooth regime shift is represented by a quasi-linear relationship between the response and control variables. The abrupt regime shift exhibits a nonlinear relationship between the response and control variables, and the discontinuous regime shift is characterized by the trajectory of the response variable differing when the forcing variable increases compared to when it decreases (i.e., the occurrence of alternative "stable" states). Most often, oceanic regime shifts are identified from time series of biotic variables (often commercial fish), but this approach does not allow the identification of discontinuous regime shifts. Recognizing discontinuous regime shifts is, however, particularly important as evidence from terrestrial and freshwater ecosystems suggests that such regime shifts may not be immediately reversible. Based on a review of various generic classes of mathematical models, we conclude that regime shifts arise from the interaction between population processes and external forcing variables. The shift between ecosystem states can be caused by gradual, cumulative changes in the forcing variable(s) or it can be triggered by acute disturbances, either anthropogenic or natural. A protocol for diagnosing the type of regime shift encountered is described and applied to a data set on Georges Bank haddock, from which it is concluded that a discontinuous regime shift in the abundance of haddock may have occurred. It is acknowledged that few, if any, marine data are available to confirm the occurrence of discontinuous regime shifts in the ocean. Nevertheless, we argue that there is good theoretical evidence for their occurrence as well as some anecdotal evidence from data collection campaigns and that the possibility of their occurrence should be recognized in the development of natural resource management strategies. (C) 2004 Elsevier Ltd. All rights reserved.
