Discussion
Overall, our findings suggest that catchment forest loss is associated with a reduction in the trophic position of apex predators, which we interpret as a general shortening of food chain lengths (Wolkovich et al., 2014). Decreases in food chain lengths are commonly driven by resource limitation because the number of viable successive trophic steps (i.e., predator-prey interactions) are limited by the total metabolic energy in a food web (Doi et al., 2009; Takimoto & Post, 2013). While terrestrial inputs into fresh waters have been shown to diminish with forest loss (Tanentzap et al., 2014), we do not believe that limitation of terrestrial resource subsidies is the primary driver of the changes observed. This is because average allochthonylevels (~7%) and its temporal invariance (Supplementary Fig. 1, 2), both suggest that terrestrial inputs are a relatively unimportant basal resource in the freshwater food webs that we studied, reinforcing shifting perceptions about the subject (Takimoto et al., 2008).
Despite this, there are some plausible mechanisms through which reductions in terrestrial subsidies can result in resource limitations. Firstly, the contribution of terrestrial inputs to overall resources may not necessarily be commensurate with their importance as a direct food source. For example, Hobbs et al. (2006) found growth rates to be higher in aquatic invertebrates feeding on a mix of leaf litter and algae when compared to individuals subsisting exclusively on either. The experimental data also show that the proportion of terrestrial carbon assimilated remained low in the mixed-diet treatment, at a value similar to those recorded in our study (~6%), despite the apparent positive effect on consumer growth (Hobbs et al., 2006). Secondly, terrestrial organic matter can also serve functions which are unrelated to nutrition. For example, terrestrial detritus adds mechanical structure and refugia to freshwater ecosystems—both beneficial to aquatic primary (Brothers et al., 2013) and secondary (Sass et al., 2006) production.
Although we found forest loss to be predictive of changes in food webs over time, the relative performance of the different forest loss metrics (Table 2; Fig. 4) suggests intricacies in our generalised conclusion. Here, we found that food web changes were better predicted by forest loss in the upstream catchment area than by overall forest loss across the entire catchment basin. This is unsurprising, as a majority of terrestrial subsidies are likely to enter stream systems with groundwater/runoff from the upstream catchment. We also found that the ratio between forest loss and forest gain (i.e., net change in forest cover) was a better predictor of food web changes than total forest lost (Table 2), suggesting that the impacts of deforestation may be attenuated by re-forestation. However, an increase in TP was recorded only in the Danum catchment, where upstream forest cover comprised part of a protected conservation area and was relatively unchanged over time (Fig. 2; Table 1). While sample size is limiting, it seems that as with most measures of environmental health (e.g., species diversity), freshwater food webs benefit more from forest preservation than from forest restoration (Hobbs et al., 2006), at least within the time interval (~20 years) of our study (Chazdon, 2008).
We note two possible confounding factors in our study, specimen body-size and substantial differences in catchment basin areas. The susceptibility of larger individuals to anthropogenic disturbances (Walters & Post, 2008) and a strong positive correlation between trophic position and body size (Brose et al., 2005) means that there may be a feasible alternative driving mechanism to the one we proposed. Specifically, forest loss could result in the loss of larger individuals (Walters & Post, 2008), hence lowering trophic positions of apex predators overall (Brose et al., 2005). While this may be the case where loss or similar anthropogenic impacts causes a reduction in trophic positions (Wilkinson et al., in revision), it is not likely to be relevant to our study. Here, our data show that fish individuals collected from both before and after were of comparable sizes (Supplementary Fig. 3).
Another factor likely to complicate the interpretation of our findings is the influence of ecosystem (or catchment) size on food web structure (Takimoto & Post, 2013). This is partially accounted for by our longitudinal study design because unlike cross-sectional studies, our primary response variable is less prone to conflation with baseline variations between survey locations. Thus, we were able to determine if observed trends in a response variable (e.g., food web structure) were driven by differences in the intensity of a disturbance regime (e.g., forest loss) independently of ecosystem size driven starting conditions (Takimoto & Post, 2013). We also found that when tested as predictors of food web change, ecosystem size performed more poorly than forest loss metrics measured at an equivalent spatial scale (i.e., whole catchment or immediate upstream area of catchment; Table 2).