Component Challenge Solution
Ecosystem survey (A) Optimizing survey effort across disparate environmental gradients, scales, and spatial contexts Implement stratified random survey design (e.g., Metzger et al 2013)
Fixed spatial grains may bias some survey data Explore utility of plotless survey methods (Cogbill et al 2018)
Subjective survey site selection may introduce spatial bias Coarse-grain (sensu Newman et al 2019) data to common raster grid cells (Miller et al 2004)
Training Data Compilation (B) and Finalization (C)
Measurement scales and units employed for biotic and abiotic field survey variables differ markedly
Convert survey measurements to presence/absence records (König et al 2019)
Disproportionate ratios of biotic to abiotic variables Pool species by common ecological traits (Kissling et al 2018)
Adequacy of survey data Utilize rarefaction methods (Chao and Jost 2012)
Effects of spatial survey grain and extent on training data structure Utilize rarefaction methods (Chao and Jost 2012); test for scale dependencies using independent data sets with differing spatial grains (Mod et al 2020)
Training data dimensionality impacts model performance Scale pools of test data across levels of ecological complexity; reduce or eliminate disjunctions (sensu Miller et al 2004) in test data
Predictor Data Compilation (B) and Finalization (C)
Determining relevant conceptual, theoretical, and statistical criteria for selecting predictor data
Draw on key theory from relevant disciples. Select predictors to resolve joint biotic-abiotic patterns and processes shaping ecosystems at various scales
Mismatched spatial resolution of predictors and response variables (Bryn et al 2021) Coarse-grain training data to match predictor grain (Newman et al 2019); assess effect of spatial scaling on model performance (König et al 2021)
Unequal numbers of biotic and abiotic predictors Employ pilot models to test the effect varying ratios, and combinations, of biotic and abiotic predictors has on ESPM performance (Brodie et al 2020)
Implications of pooling disparate predictors shaping ecosystem pattern across spatial grains Determine whether modelled responses are shaped more by local and or regional drivers (e.g., Soranno et al 2019), and whether responses vary among ecosystems and ecosystem constituents
Implications of grouping predictors of direct, indirect, and resource gradients (Austin 2013) Clarify expected responses of individual and aggregate ecosystem features (Table S1) to individual environmental gradients (Austin 2013)
Optimizing combinations of predictors Explore implications of co-variate trade-off (Brodie et al 2020)
Model Building and Assessment (D) Model algorithm selection Seek ESPM algorithms with adequate flexibility, functionality, and predictive capacity. Candidate algorithms may be adapted from community-level (e.g., Nieto-Lugilde et al 2018) or bioregion modelling (e.g., Hill et al 2020) techniques.
Determining model settings Settings (e.g., cross-validation, latent variables, random effects, residual associations) depend on selected algorithm and on results of trial model implementations; consult published methodological guidelines (e.g., Ovaskainen and Abrego 2020, Mokany et al 2022)
Evaluating model fit and performance Metrics for evaluating model fit and performance vary by algorithm and modelling intent; consult appropriate guidelines (Araújo et al 2019, Zurell et al 2020)
Translating model prediction to mapped spatial outputs Continuous (e.g., ordination) or discrete (e.g., region of common profile) post-hoc analytical outputs can be mapped in YUV colour space (e.g., Tikhonov et al 2020)
Intrinsic Predictions (E) Determine value of lower-order intrinsic (e.g., biotic and abiotic ecosystem constituents and properties) predictions for informing higher-order modelling objectives Improve understanding of commonalities, potential interactions, and properties of individual variables characterizing ecosystems or sites; disentangle the relative contributions of individual variables to assembly mechanisms; relate model results to existing ecological knowledge
Extrinsic Predictions (F) Select method for resolving the identities, features, and distribution of disparate ecosystems and ecosystem types Adapt classification methods and indices (e.g., similarity, clustering, ordination) developed for community-level (Ferrier and Guisan 2006) or bioregion modelling (Hill et al 2020)
Explore utility of less common tools: concordance analysis (e.g., Taranu et al 2020), ecological uniqueness indices (e.g., Dansereau et al 2022), or embedding techniques such as T-SNE (e.g., Sonnewald et al 2020).