Ecological theory predicts ecosystem stressor interactions in freshwater ecosystems, but highlights the strengths and weaknesses of the additive null model

Abstract

Understanding and predicting how multiple co-occurring environmental stressors combine to affect biodiversity and ecosystem services is an on-going grand challenge for ecology. So far progress has been made through accumulating large numbers of smaller-scale individual studies that are then investigated by meta-analyses to look for general patterns. In particular there has been an interest in checking for so-called ecological surprises where stressors interact in a synergistic manner. Recent reviews suggest that such synergisms do not dominate, but few other generalities have emerged. This lack of general prediction and understanding may be due in part to a dearth of ecological theory that can generate clear hypotheses and predictions to tested against empirical data. Here we close this gap by analysing food web models based upon classical ecological theory and comparing their predictions to a large (546 interactions) dataset for the effects of pairs of stressors on freshwater communities, using trophic- and population-level metrics of abundance, density, and biomass as responses. We find excellent overall agreement between the stochastic version of our models and the experimental data, and both conclude additive stressor interactions are the most frequent, but that meta-analyses report antagonistic summary interaction classes. Additionally, we show that the statistical tests used to classify the interactions are very sensitive to sampling variation. It is therefore likely that current weak sampling and low sample sizes are masking many non-additive stressor interactions, which our theory predicts to dominate when sampling variation is removed. This leads us to suspect ecological surprises may be more common than currently reported. Our results highlight the value of developing theory in tandem with empirical tests, and the need to examine the robustness of statistical machinery, especially the widely-used null models, before we can draw strong conclusions about how environmental drivers combine.