OOS 22-10
Early warning indicators fail to forewarn of impending kelp forest regime shifts

Wednesday, August 7, 2013: 4:40 PM
101D, Minneapolis Convention Center
Mark Novak, Integrative Biology, Oregon State University, Corvallis, OR
Jane Watson, Dept. of Biology, Vancouver Island University, Nanaimo, BC
Mike Kenner, Ecology and Evolutionary Biology, University of California, Santa Cruz, CA
James A. Estes, Ecology and Evolutionary Biology, University of California, Santa Cruz, CA
Background/Question/Methods

Many ecological communities exhibit alternative states characterized by fundamentally different species abundances, with one state often being less desirable than the other.  The threat of tipping points between these states -- whereby even small, incrememental changes in a driving variable can lead to large-scale reorganizational changes in community structure -- has motivated the development of indicators to forewarn of approaching transitions.  In particular, a growing body of literature illustrates how the phenomena of critical slowing down (decreasing response rates to small stochastic perturbations) and flickering (switching back and forth between states in response to stochastic perturbations) can lead to increases in the temporal autocorrelation and the variance (across space and in time) of species abundances before the tipping point is reached.

We assessed the utility of a suite of such indicators to forewarn of tipping points in the southern California kelp forests of San Nicolas Island.  Like elsewhere around the globe, these kelp forests are characterized by two general states: a kelp-dominated state and an urchin-dominated state.  The time-series of kelp and urchin abundance dynamics we analyzed reflect 31 years of biannual sampling performed in fixed-location plots at six sites surrounding the island.

Results/Conclusions

Calibrated Bayesian change point analysis of kelp and urchin dynamics provided objective evidence for the occurence of multiple state-shifts of varying magnitude: posterior probabilities associated with the inference of a state-shift having occurred varied continuously from zero to one.  Two sites where population sizes varied cyclically showed no evidence for state-shifts having occured.

None of the spatial or temporal indicators we assessed showed evidence of being able to forewarn of impending tipping points:  Neither the normalized indicator values preceeding a given time-point, nor the run-length of increasing indicator values preceeding a given time-point, was positively associated with that time-point’s posterior probability of having exhibited a state-shift.  Furthermore, time-points exhibiting a near-zero state-shift probability were equally, if not more likely, to be preceeded by high indicator values and runs of increasing indicator values than were time-points exhibiting a high state-shift probability.  Rates of false-positive warnings were equally high for the two sites where no state shift occurred.  Our analyses therefore echo recent theoretical arguments warning against an overreliance on early warning indicators in management and conservation practices.