Statistical analyses
Descriptive statistics were expressed as arithmetic means and standard errors (SE). When feasible, errors were propagated to maintain an estimate of variance. We tested correlations between our predictor variables and herbivory levels as well as the annual proportion of elemental fluxes using linear mixed effect modelling (‘lmer’ function in ‘lme4’ package, Bates et al. 2015). Confidence intervals (95%) were constrained with 999 bootstrap simulations. The relative position above the valley bottom was included in the models as Z-scores based on the elevation grouped by transect, i.e. yielding values close to -1 for the lowest site, close to 0 for the middle site, and close to +1 for the highest site. For herbivory levels, we tested volumetric soil water content, mean growing season soil temperature, annual solar radiation, relative position above the valley bottom and absolute site elevation as abiotic predictors, while leaf C:N and CT:C ratios were included as biotic predictors. When testing these as fixed effects, the other variables were classified (4 approximately normally distributed groups per variable) and included as random effects. Further, the transect number was included as a random effect in all models (9 groups). For the annual proportion of elemental fluxes, we tested herbivory level, leaf production, leaf nutrient (N or P) content, and nutrient (N or P) resorption as fixed effects. Nutrient resorption was included, as strong nutrient resorption would decrease the nutrient flux through senesced litter, and result in a larger proportion of the annual nutrient fluxes through insect deposits for a given leaf area loss. Variables were transformed when needed to comply with model assumptions and assure similar variance for all variables. All statistical analyses were conducted in R version 3.6.2 (R Core Team, Vienna, Austria).
To assess the consistency in the relationship between local and regional elevation and herbivory, we linearly regressed BIH against the local elevation (elevation centred per transect) and the regional elevation (absolute elevation centred). We included green leaf N-content (% of mass) plotted against elevation to check the design, as we expected a more consistent positive relationship between leaf N and elevation (Körner 1989, Read et al. 2014), while the expectation to foliar loss to herbivores was less straight forward (see introduction).