in Lake Ecosystems:
of Excellence in Ecology volume
Updated 15 July 2003
Much ecological thinking is organized around ideas about stability, and attempts to understand the variability of ecosystems as departures from a stable condition. In contrast, large-scale field observations, especially long-term ecological research, suggest that ongoing change and variability are the typical condition. Stability is an anomaly that begs explanation. These contrasting perspectives may be reconciled, in part, by notions of complex dynamics and multiple attractors. Shifts among two or more quasi-stable regimes are one of the simplest types of complex dynamics, and may explain some important changes in ecosystems. Even this minimal type of complexity has profound implications for ecosystem management, because it suggests that changes in ecosystems may be surprising, large, and difficult to reverse. This book is about anticipating regime shifts - capabilities, limitations, prospects, and some implications for design of management systems.
The ever-changing nature of ecosystems is richly illustrated by long-term ecological research. In many cases, long-term change is interpreted as a shift from one dynamic regime to another: oligotrophic to eutrophic, grassland to woodland, before versus after the top predator was lost. This book is about such ecological regime shifts. In retrospect, it is often possible to understand ecological regime shifts. Anticipating them is a different matter. Forecasting involves a model, and the past data are often consistent with several different models with contrasting implications for the future. This book addresses some challenges of modeling and forecasting regime shifts.
Lake and island ecosystems have provided many examples of regime shifts. The modularity of lakes and islands is a great advantage for understanding regime shifts, because similar regime shifts can be observed over and over again in many ecosystems. Here we will focus on lakes. Observed features of three types of regime shifts are summarized: eutrophication, depensatory loss of fish stocks, and trophic cascades.
Observations of regime shifts have evoked many kinds of models. While there are straightforward methods for sorting among rival models, there is considerable doubt about our ability to forecast future regime shifts based on any particular model. This chapter uses a simple simulation to illustrate the difficulty of identifying the best model for predicting regime shifts. Multiple types of evidence, including studies of underlying mechanisms, cross-system comparisons, deliberate ecosystem manipulations, as well as long-term observations, may be necessary. The use of multiple lines of evidence, and the persistence of uncertainty despite an extraordinary set of scientific information, are illustrated by current understanding of the eutrophication of Lake Mendota.
Comparison of contrasting models is a powerful tool in ecosystem studies. In this chapter, a long-term whole-lake experiment is analyzed using two different models - a linear empirical one, and a more mechanistic nonlinear one. Regime shifts involving multiple basins of attraction are possible in the mechanistic model, but not in the linear one. Both models fit data reasonably well. On the other hand, both models have systematic shortcomings that fail to capture important features of the data. For the scientist, the differences between the models, and their departures from data, are fodder for exciting hypotheses. For the lay public, the models evoke dramatically different mental pictures of how the lakes may change in the future. For the manager, the models raise troubling questions about how to evaluate risk.
If an ecosystem is subject to regime shifts, the best way to learn about the threshold between the regimes is to cross the threshold repeatedly, through experimental manipulation. Is it possible to improve our knowledge of a threshold without crossing it? A simple simulation model for management of a lake subject to eutrophication shows that it is possible to measure the threshold by crossing it. However, it is difficult to learn about the threshold while remaining in the oligotrophic regime, even with rather risky manipulations. The conclusions are pessimistic about the possibility of learning about thresholds from safe experiments in singular, unique ecosystems. However, in modular ecosystems such as lakes, islands, and small watersheds, it may be possible to transfer knowledge from one ecosystem to another. In these situations, experimentation to learn about thresholds may have significant payoff if analysts and managers are willing to assume that information is exchangeable across similar systems.
When data give comparable support to each of several different models, it is possible to devise decision rules that account for the information in each of the models. Simple simulations are used to demonstrate such rules for a fishery in which two dynamic regimes (one unproductive, the other highly productive) can exist. The risk of crossing the threshold to the unproductive regime is high unless the data used to calibrate the models are highly informative. Thus, even sophisticated decision rules that account for information in multiple models are not likely to avoid thresholds, unless the available data provide accurate estimates of the threshold. In the absence of such information, extremely cautious policy choices may avoid crossing the threshold. The key to managing systems subject to regime shifts lies in designing programs that anticipate the possibility of diverse events, not in planning around optimal forecasts.
The longer we study ecosystems, the more we see. Regime shifts are one of the remarkable phenomena that offer fascinating scientific opportunity for ecologists. This book suggests several promising research avenues, including (1) experimental study of regime shifts in modular ecosystems such as lakes, islands and watersheds, (2) combination of observational and mechanistic studies to improve understanding of regime shifts, and (3) analysis of multiple models, ranging from the empirical to the mechanistic, to understand regime shifts in long-term data.
The message to managers is that change and variability are common in long-term data from spatially extensive ecosystems, and we do not yet have sufficient understanding to forecast important changes like regime shifts. Models which imply that changes in ecosystems are easily reversible may gain considerable statistical support. However, it does not follow that the probability of irreversible change is zero, or that reversal of unwanted change will be easy. There is great risk in basing decisions on singular "optimal" models. Instead, managers should expect a wide range of possible ecological regimes in the future, and plan accordingly by embracing the expectations that emerge from diverse families of models. It's not a decision problem - it's a matter of design for the novel and the unexpected.
Many of the regressions presented in this book derive from Bayesian statistics. Bayesian methods are less familiar than some other statistical approaches in ecology, and for this reason it is appropriate to provide some methodological background. This appendix provides a brief summary of the Bayesian regression methods that are the basis of many analyses presented in the book.
Questions with this page? Contact firstname.lastname@example.org