Ecosystem, Abiotic, and Biotic Responses – Overview of Models
The final ecosystem, abiotic, and biotic response models explained, 41.4, 29.03, and 50.9 percent of deviance in observed dissimilarity. These values are consistent with those in published generalized dissimilarity models (Mokany et al. 2022) (see S1 for detailed model overviews). Matrix permutation tests, employed to cross-validate model performance, revealed all three formulations were statistically significant, with p-values ranging from 0.04 to < 0.01. The three models were each fit with relatively low numbers of predictors, ranging from 4 to 8 (Table 1).
The final ecosystem response model had six significant predictors (Table 1), including leaf area index, softwood basal area, canopy height, normalized vegetation difference index, hardwood basal area, and terrain ruggedness. The strongest predictors were biotic (jointly explaining 41.31% of deviance), and much of the variation in dissimilarity was accounted for by leaf area index (19.72%) and softwood basal area (11.55%). Terrain ruggedness was the only abiotic predictor making a significant, albeit minor (1.55%), contribution to model fit.
Terrain ruggedness, depth-to-water, canopy height, and geographic distance were significant predictors in the final abiotic model (Table 1). Two of these predictors were abiotic (jointly explaining 23.14% of deviance), with terrain ruggedness making the most important contribution (12.81%). Overall, the most influential biotic predictor was canopy height (5.46%). Among the significant predictors, geographic distance had the lowest strength (<1%).
Table 1. Summary of the relative contribution of predictors employed in the ecosystem, abiotic, and biotic response models. For each model, individual predictor coefficient sums are reported, while the proportion of total deviance in dissimilarity, explained by each predictor, is in parentheses. Predictors with higher values, have greater importance, for each given model. The three most important predictors per model are bolded; predictor types are distinguished by font colour: green (biotic), brown (abiotic), and black (geographic distance). Only those significant (p <0.05) predictors retained after matrix permutation with backward elimination (Fitzpatrick et al. 2024) are shown. Predictor data and acronyms are described in S1.