Summary of model selection testing for interactions between multiple stressors on macroinvertebrate abundance and richness and the SPEAR and BMWP indices, based on the Akaike information criterion corrected for sample size (AICc).

Models are ordered from the best to the poorest fit according to Akaike weights (wi). K, number of estimated parameters for each model; Δi (delta AICc), difference in AICc value relative to the best model; wi, probability that a model is the best among the whole set of models. For each response variable, five models were constructed, which are ordered from the simplest model without interactions (model 1: null model, with no interactions) to the most complex one (model 5, containing the 3-way interaction). Models differ in the number of parameters according to the most parsimonious combination of structure and terms described in S9 Table. PT, pesticide toxicity (Tumax); SE, sedimentation index; NE, nutrient enrichment (SRP).

本研究依据赤池权重（Akaike weights, wi）将模型按拟合优度从优至劣排序。其中，K为每个模型的待估计参数数目；Δi（ΔAICc）为该模型与最优模型间的修正赤池信息准则差值；wi为某一模型在全部模型集合中为最优模型的概率。针对每个响应变量，共构建5个模型，按从不含交互作用的最简模型（模型1：零模型，无任何交互项）至包含三阶交互作用的最复杂模型（模型5）的顺序排列。各模型的参数数量由表S9中所描述的模型结构与变量项的最简约组合决定。缩写说明如下：PT为农药毒性（Tumax），SE为沉降指数，NE为营养富集（SRP）。