Carl Boettiger, Alan Hastings and I have a new review out on early warning signals and regime shifts. It’s about an idea that Carl and I have been kicking around for a couple of years and finally got around to writing up for a special issue of Theoretical Ecology.
There has been a lot of excitement in the past six years or so about early warning signs - statistical signals that indicate that an ecosystem is near a dramatic shift, like the collapse of a fishery or a switch from grassland to desert. Much of the interest in this work comes from the idea that these statistical signals could be general properties of ecosystems prone to such rapid changes. Thus, we might not need to know a great deal of detail about a specific ecosystem in order to use these kinds of signals to forecast change - a real benefit given the pace of global change and the paucity of data about a lot of ecosystems.
Our paper focuses on a flaw see with this approach: that these early warning signals are an emergent property of a particular model used to represent ecosystems that have the potential to collapse. This model, called the “saddle node bifurcation” is often incorrectly equated with the idea of regime shifts in general. However, it’s really just one of a number of mechanisms that can represent regime shifts. Yet because this model appears in several common mathematical representations of ecosystems, almost all work on early warning signals has assumed the saddle node mechanism.
Just because this model is commonly used to describe ecosystems doesn’t mean ecosystems actually behave this way. Alan published a paper in 2010 showing a number of counterexamples - alternative ways a model could represent an ecosystem with the potential for a rapid shift - in which these signals never appear. Conversely, there are several cases of models that exhibit early warning-like patterns without any imminent shift.
In our review, we try to illuminate what is known about warning signals observed in models that have a saddle node and contrast this to other models that have received much less attention. So far, for instance, empirical data have only been collected in systems for which the saddle-node is thought to be a very good representation (e.g., freshwater lakes, pitcher-plants). We need more examples of this type of model testing in different ecosystems with different potential mechanisms of regime shift.
Other models of ecosystems have the potential to provide different warning signals, but we need to know more about them, and we need better methods of deciding which model is appropriate to apply to any particular ecosystem. Early warning signals may be less generally applicable than initially hoped, but they can be quite powerful if properly applied.
Finally, this is my first paper in which pretty much all collaboration occurred over github, so you can go and see the entire revision history there. This worked out great and I’ll definitely be doing it again.