Figure 2. Ecosystem spatial pattern modelling workflow. The
sequenced workflow includes steps to: (a) incorporate field-based biotic
and abiotic ecosystem response data and remotely-sensed predictors; (b)
design and implement independent ecosystem, abiotic, and biotic response
models; (c) select model data and assemble a site-pair table for GDM;
(d) fit GDM models; (e) apply GDM model outputs to inform decision
making and generate follow-up research (e.g., empirical testing). Tree
and shrub graphics (Natural Resources Canada 2015); other graphics
(Microsoft 365 premium creative content; Creative Commons; or original
content).
classes that performed the best and top performing models were
subsequently run with matrix permutation (see S1). Matrix permutation
was used to cross-validate our models and to assess both model and
predictor significance. Here, model significance was determined by
comparing the deviance explained between permuted and unpermuted
formulations (Fitzpatrick et al. 2024). Matrix permutation was
implemented with backward elimination to exclude non-significant (p
that included significant predictors, identified through matrix
permutation. We report the outputs of the final ecosystem, biotic, and
abiotic response models (see S1).