A topological similarity-based bootstrapping method for inferring food web parameters

A food web is a representation of trophic interactions in an ecosystem. Food webs are often generated by a single dataset aggregated from one or several surveys. Point estimates of food web parameters can be calculated from the data, but it remains an open question as to how their corresponding interval estimates can be quantified. Although conventional methods such as bootstrapping represent potential solutions, they tend to underestimate several network parameters. Here, we propose a simple bootstrap-based resampling procedure for inferring food web parameters. First, for a particular food web parameter, we obtain its point estimate by calculating the corresponding statistics from the original food web. Second, we generate a resampled food web by sampling with replacement the same number of species from the original food web, and for each resampled species we record how many prey items it consumes in the original food web. Third, a resampled species is allowed to consume its original prey species if such a species is also present; if not present, it instead consumes the resampled species that is most topologically similar to its original prey species. Many resampled food webs can be generated in this manner, and we calculate particular food web statistics for each of them. These form a sampling distribution from which interval estimates of the true food web parameter can be determined. We demonstrate our methodology on two different food web datasets and discuss its application in comparing food webs of various sizes and connectance.