Community- and ecosystem-level effects of multiple environmental change drivers: Beyond null model testing

Understanding the joint effect of multiple drivers of environmental change is a key scientific challenge. The dominant approach today is to compare observed joint effects with predictions from various types of null models. Drivers are said to combine synergistically (antagonistically) when their observed joint effect is larger (smaller) than that predicted by the null model. Here, I argue that this approach does not promote understanding of effects on important community- and ecosystem-level variables such as biodiversity and ecosystem function. I use ecological theory to show that different mechanisms can lead to the same deviation from a null model's prediction. Inversely, I show that the same mechanism can lead to different deviations from a null model's prediction. These examples illustrate that it is not possible to make strong mechanistic inferences from null models. Next, I present an alternative framework to study such effects. This framework makes a clear distinction between two different kinds of drivers (resource ratio shifts and multiple stressors) and integrates both by incorporating stressor effects into resource uptake theory. I show that this framework can advance understanding because of three reasons. First, it forces formalization of “multiple stressors,” using factors that describe the number and kind of stressors, their selectivity and dynamic behaviour, and the initial trait diversity and tolerance among species. Second, it produces testable predictions on how these factors affect biodiversity and ecosystem function, alone and in combination with resource ratio shifts. Third, it can fail in informative ways. That is, its assumptions are clear, so that different kinds of deviations between predictions and observed effects can guide new experiments and theory improvement. I conclude that this framework will more effectively progress understanding of global change effects on communities and ecosystems than does the current practice of null model testing.