Stressor gradient coverage affects interaction identification

This study aims at understanding how observed inconsistencies in the response of biotic indicators to multiple stressors may result from different stressor gradient lengths being represented at different areas or temporal windows, either as the result of intrinsic natural causes or as the result of sampling bias. We simulated a pool of sites showing five types of interactive responses of indicators to two co-occurring virtual stressors, as well as several sampling constraints, resulting in different portions of each stressor's gradient being covered. The sampled gradient length showed a strong influence on the detection of single stressor effects, both in terms of statistical significance and goodness-of-fit. Increasing constraints on gradient coverage also led to an increasingly deficient identification of stressor interactions. The fail in detecting significant interactions largely dominated over switches between interaction types. The simulations indicated that datasets not fully capturing stressor gradients may hinder the ability to unveil underlying multiple stressor effects. As distinct portions of stressor gradients may be present at different contexts and may change over time, our simulations stress the importance of adaptive management strategies based on robust sampling designs to minimize potential statistical artefacts and uncertainties.

Automated summary

This study aims to understand why inconsistencies in the response of biotic indicators to multiple stressors may occur. The researchers found that the length of the stressor gradient represented in different areas or temporal windows can greatly impact the detection of single stressor effects and the identification of stressor interactions. Insufficient coverage of stressor gradients in datasets can hinder the ability to uncover underlying multiple stressor effects. The simulations highlight the importance of adaptive management strategies based on robust sampling designs to minimize potential statistical artefacts and uncertainties.