Species interactions are dynamic processes that vary across environmental and ecological contexts, and operate across scale boundaries, making them difficult to quantify. Nevertheless, ecologists are increasingly interested in inferring species interactions from observational data using statistical analyses of their spatial co-occurrence patterns. To test the accuracy of various co-occurrence methods in inferring trophic interactions, we utilized two long-term datasets of pond-breeding amphibian species co-occurrences from Mt. Rainier National Park (Washington) and Mt. Hood (Oregon). We compared four statistical methods for inferring species interactions from their spatial associations to known interactions for these amphibian species. We used the best performing method, the Markov network, to further investigate the sensitivity of interaction inferences to spatial and temporal scale-dependencies and trophic structuring.
Results/Conclusions
We found significant turnover, or dissimilarity, among the resulting spatial interaction networks; none correctly identified the known interaction network among amphibian species. We also show that inferred interactions are dependent upon the spatial and/or temporal scale of observation and that predator presence can permeate indirect effects throughout the network and dilute the detectability of competitive interactions among prey species. Overall, our study highlights a limiting frontier in co-occurrence theory and the disconnect between widely implemented methodologies and their ability to accurately infer interactions in trophically-structured communities.