Establishing nutrient thresholds in the face of uncertainty and multiple stressors: A comparison of approaches using simulated datasets

Various methods have been proposed to identify threshold concentrations of nutrients that would support good ecological status, but the performance of these methods and the influence of other stressors on the underlying models have not been fully evaluated. We used synthetic datasets to compare the performance of ordinary least squares, logistic and quantile regression, as well as, categorical methods based on the distribution of nutrient concentrations categorised by biological status. The synthetic datasets used differed in their levels of variation between explanatory and response variables, and were centered at different positions along the stressor (nutrient) gradient. In order to evaluate the performance of methods in “multiple stressor” situations, another set of datasets with two stressors was used. Ordinary least squares and logistic regression methods were the most reliable when predicting the threshold concentration when nutrients were the sole stressor; however, both had a tendency to underestimate the threshold when a second stressor was present. In contrast, threshold concentrations produced by categorical methods were strongly influenced by the level of the stressor (nutrient enrichment, in this case) relative to the threshold they were trying to predict (good/moderate in this instance). Although all the methods tested had limitations in the presence of a second stressor, upper quantiles seemed generally appropriate to establish non-precautionary thresholds. For example, upper quantiles may be appropriate when establishing targets for restoration, but not when seeking to minimise deterioration. Selection of an appropriate threshold concentration should also attend to the regulatory regime (i.e. policy requirements and environmental management context) within which it will be used, and the ease of communicating the principles to managers and stakeholders.