The impact of a stage structure on the food webs
Out of 30,000 food webs, 5865 linked (19.6%) and 9099 unlinked webs (30.3%) persisted for 1000 growing seasons and passed the criteria, of which 2628 webs were shared. Assessed by a set of network structural metrics and characteristics quantified on the unweighted networks after removing extinct taxa (“ATN-filtered”; note that taxa that went extinct could be different between the linked and unliked webs), the persisting food webs in both scenarios largely overlapped in terms of the majority of the properties measured (Fig. A4). The number of persisting nodes ranged from 15 to 60, of which invertebrates comprised the majority and 3 to 30 nodes were life-history stages of fishes. The majority of the food webs did not have a top predator.
The distributions of measured food web properties between the linked and unlinked food webs overlapped substantially (Fig. 4a). Nonetheless, the bootstrap differences in means between the linked and unliked food webs (linked – unlinked webs) highlighted that some of the mean effect sizes were noticeable (Fig. 4b); the mean CVs of fish biomasses was about 3.3-orders of magnitude smaller, the number of nodes was greater by about 6, the slope of biomass spectra was 0.17 smaller, and the CV and skewness of normalized fish energy gain were 0.17 and 2.35 larger, respectively, in the linked webs than in the unlinked webs on average. These suggested that, when stages were linked relative to when unlinked, on average fish biomass dynamics were less variable, the biomass pyramids were more bottom-heavy, food webs sustained several more species or fish stages, and fish energy gain was more variable among links and dominated by small gains (indicated by the positive skews). Taken together, the results suggested that the life-history stage structure of fishes on average moderately promoted the stability of and diversity in these complex food webs, once they were able to persist. The sensitivity analysis showed that the patterns in the bootstrap differences in mean effect sizes remained in the same direction (negative or positive) as the baseline simulations across the sensitivity analysis simulations except several cases (Fig. A5). The bootstrap differences in mean effect sizes were qualitatively deviated from the baseline runs for a few measures when the maximum and minimum numbers of fish stages\(\left(\text{NStage}_{\max},\ \text{NStage}_{\min}\right)\) were smaller (#1), full interference competition was included (#13), or when \(q\) was reduced to 1.2 (#15). The number of persisting webs was markedly smaller (~25%) when \(q=1.2\) (data not shown).