Disentangling spatial and environmental effects: Flexible methods for community ecology and macroecology

Community ecologists and macroecologists have long sought to evaluate the importance of environmental conditions in determining species distributions, community composition, diversity across sites. Different methods been used estimate species–environment relationships, but their differences jointly fit disentangle spatial autocorrelation structure remain poorly studied. We compared how four broad families statistical models estimated contribution environment space variation occurrence abundance. These included redundancy analysis (RDA), generalized linear (GLMs), additive (GAMs), three types tree-based machine learning (ML) methods: boosted regression trees (BRT), random forests, trees. The component model consisted Moran's eigenvector maps (MEMs; RDA, GLM, ML), smooth splines (in GAM), or nonlinear modeling coordinates ML). simulated typical site-by-species data assess methods' performance (1) fitting models, (2) partitioning explained by predictors. observed marked among generally caused different performances structures. Generalized BRT with MEMs were most reliable for effects a wide range scenarios. remaining tended underestimate pure effects, because either underfitting structures overestimation when estimated. Performing on nine empirical datasets using these yielded contrasting results, especially estimation fraction variation. Our results suggest that previously overlooked performing partitioning, ML, offer flexible approaches analyze matrices. provide general guidelines usefulness under ecological sampling scenarios, distribution modeling, ecology, macroecology.