Supporting information: Size, foraging and food web structure

The DBM requires the body size of each species and four allometries; those of energetic content of individuals of each species, density of each species, the attack rate of each consumer on each resource species, and the handling time of each consumer on each resource species. We assumed that energetic content is linearly related with body mass and a common allometric scaling of density and body mass. We searched the literature for empirically motivated and parameterised allometries of handling time and attack rate and, due to its importance for food web structure, we focused on allometry of handling times. For a function to be directly applicable to the model, it would need to 1) be parameterised from data that cover the range of organism sizes observed in real food webs (about 17 orders of magnitude); 2) include information about effects of a large range of resource sizes on the foraging of a single predator species. Various functions that appear in the literature are given in table S1 and none appear applicable. 2 Fifteen real food webs Table S2 contains the original sources of predation matrices and common names used in this article. Only the scmown2 web in Cattin et al. (2004) was used. We did not model or analyse the Weddell Sea food web because the large number of species and links made the required computations exceed available processing resources. The predation matrices of each food web are presented in figure S1. Body sizes were compiled from a variety of sources (table S2). These real food webs contain producers, herbivores, carnivores, parasites, and parasitoids. The organisms display a range of feeding interactions includ-ing predation, herbivory, bacterivory, parasitism, pathogenic, and parasitoid. For nine of the 15 webs there was information available about which inter-action was represented by each individual feeding link. 3 Parameterising the Allometric DBMs As the literature does not contain suitably parameterised allometries for attack rates and handling times, we used optimisation to fit the parame-ters. Two methods of optimisation were used: complete enumeration (CE) 2 T ab l