Source food webs as estimators of community web structure

Taxonomically restricted ''source webs'' are commonly used to represent the community food webs of which they are part. This raises a methodological problem if source webs provide biased estimates of food web structure. We use four high quality, extensive food webs containing multiple source species to measure the sensitivity of food web metrics to the number of source species used to generate a web. The total number of species (S), linkage density (US), directed connectance (L/S-2), and the fractions of basal (B), intermediate (I), and top (T) species are all sensitive to the number of source species. Further, the pattern of variation for the latter fractions is inconsistent and web dependent, indicating that source webs are inappropriate for characterizing these properties. Linkage densities increase with the numbers of source species in all four cases, with webs based on single or few sources severely underestimating values obtained for the full webs. Connectance shows more constrained decreases with increasing numbers of sources, suggesting that multiple-source webs may provide reasonable estimates of connectance for community webs.