Understanding what makes food webs stable has long been a goal of ecologists. Topological structure and the distribution and magnitude of interaction strengths in food webs have been shown to confer important stabilizing properties. However, our understanding of how variable species interactions affect food web structure and stability is still in its infancy. Anthropogenic stress, such as acid mine drainage, is likely to place severe limitations on the food web structures possible due to changes in community composition and body mass distributions. Here, we used mechanistic models to infer food web structure and quantify stability in streams across a gradient of acid mine drainage. Multiple food webs were iterated for each community based on species pairwise interaction probabilities, in order to incorporate the variability of realistic food web structure. We found that food web structure was altered systematically with a 32-fold decrease in the number of links and a 2-fold increase in connectance across the gradient. Stability generally increased 6-fold with increasing acid mine drainage stress, regardless of how interaction strengths were estimated. However, the distribution of the stability measure, s, for some impacted communities separated into clusters of higher and lower magnitude depending on how interaction strengths were estimated. Management and restoration of impacted sites needs to consider their increased stability, as this may have important implications for the re-colonization of desirable species. Furthermore, active species introductions may be required to overcome the internal ecological inertia of affected communities.

community_data.csv: Benthic macroinvertebrates were collected with Surber samples (0.06 m²) from New Zealand streams across a gradient of acid mine drainage stress. Fish were also sampled using electrofishing methods. All animals collected had their body lengths measured and their biomass estimated using published length-weight regressions. Dry weights and density values are average estimates per taxa. 

site: site identifier. Corresponds to observations in "amd_gradient.csv"

taxa: Taxonomic name of sample. Fish are identified to species, most aquatic insects are identified to genus. 

dw.g: taxa average estimated dry weight, in grams 

density_m2: taxa average estimated densities, in m²

rel.ab: taxa relative abundance per site

amd_gradient.csv: The acid mine drainage stress gradient was determined with PCA. The site scores from PC axis 1 were used to infer level of stress (higher values represent greater AMD stress). 

site: site identifier, corresponds with "community_data.csv"

amd_stress: site specific scores from PC axis 1. negative values are unimpacted, and increasing positive values represent increasing AMD stress

niche_model_parameters.csv: The model parameters are coefficient estimates for a statistical model of interaction probabilities based on predator-prey body size relationships. The statistical model used is described in Bartomeus et al. (2016 Functional Ecology). Data from Broadstone stream and Tadnoll Brook (Woodward et al. 2010, Advances in Ecological Research) were used to parameterize the model. 

coefficient: the model coefficient

estimate: the estimated value of the coefficient from the parameterized model.