A large array of species distribution model (SDM) approaches has been developed for explaining and predicting the occurrences of individual species or species assemblages. Given the wealth of existing models, it is unclear which models perform best for interpolation or extrapolation of existing data sets, particularly when one is concerned with species assemblages. We compared the predictive performance of 33 variants of 15 widely applied and recently emerged SDMs in the context of multispecies data, including both joint SDMs that model multiple species together, and stacked SDMs that model each species individually combining the predictions afterward. We offer a comprehensive evaluation of these SDM approaches by examining their performance in predicting withheld empirical validation data of different sizes representing five different taxonomic groups, and for prediction tasks related to both interpolation and extrapolation. We measure predictive performance by 12 measures of accuracy, discrimination power, calibration, and precision of predictions, for the biological levels of species occurrence, species richness, and community composition. Our results show large variation among the models in their predictive performance, especially for communities comprising many species that are rare. The results do not reveal any major trade‐offs among measures of model performance; the same models performed generally well in terms of accuracy, discrimination, and calibration, and for the biological levels of individual species, species richness, and community composition. In contrast, the models that gave the most precise predictions were not well calibrated, suggesting that poorly performing models can make overconfident predictions. However, none of the models performed well for all prediction tasks. As a general strategy, we therefore propose that researchers fit a small set of models showing complementary performance, and then apply a cross‐validation procedure involving separate data to establish which of these models performs best for the goal of the study.

Aims The choice of environmental predictor variables in correlative models of plant species distributions (hereafter SDMs) is crucial to ensure predictive accuracy and model realism, as highlighted in multiple earlier studies. Because variable selection is directly related to a model's capacity to capture important species' environmental requirements, one would expect an explicit prior consideration of all ecophysiologically meaningful variables. For plants, these include temperature, water, soil nutrients, light, and in some cases, disturbances and biotic interactions. However, the set of predictors used in published correlative plant SDM studies varies considerably. No comprehensive review exists of what environmental predictors are meaningful, available (or missing) and used in practice to predict plant distributions. Contributing to answer these questions is the aim of this review. Methods We carried out an extensive, systematic review of recently published plant SDM studies (years 2010–2015; n = 200) to determine the predictors used (and not used) in the models. We additionally conducted an in‐depth review of SDM studies in selected journals to identify temporal trends in the use of predictors (years 2000–2015; n = 40). Results A large majority of plant SDM studies neglected several ecophysiologically meaningful environmental variables, and the number of relevant predictors used in models has stagnated or even declined over the last 15 yr. Conclusions Neglecting ecophysiologically meaningful predictors can result in incomplete niche quantification and can thus limit the predictive power of plant SDMs. Some of these missing predictors are already available spatially or may soon become available (e.g. soil moisture). However, others are not yet easily obtainable across whole study extents (e.g. soil pH and nutrients), and their development should receive increased attention. We conclude that more effort should be made to build ecologically more sound plant SDMs. This requires a more thorough rationale for the choice of environmental predictors needed to meet the study goal, and the development of missing ones. The latter calls for increased collaborative effort between ecological and geo‐environmental sciences.

A key focus in ecology is to search for community assembly rules. Here we compare two community modelling frameworks that integrate a combination of environmental and spatial data to identify positive and negative species associations from presenceabsence matrices, and incorporate an additional comparison using joint species distribution models (JSDM).The frameworks use a dichotomous logic tree that distinguishes dispersal limitation, environmental requirements, and interspecific interactions as causes of segregated or aggregated species pairs. The first framework is based on a classical null model analysis complemented by tests of spatial arrangement and environmental characteristics of the sites occupied by the members of each species pair (Classic framework). The second framework, (SDM framework) implemented here for the first time, builds on the application of environmentally-constrained null models (or JSDMs) to partial out the influence of the environment, and includes an analysis of the geographical configuration of species ranges to account for dispersal effects.We applied these approaches to examine plot-level species co-occurrence in plant communities sampled along a wide elevation gradient in the Swiss Alps. According to the frameworks, the majority of species pairs were randomly associated, and most of the non-random positive and negative species associations could be attributed to environmental filtering and/or dispersal limitation. These patterns were partly detected also with JSDM. Biotic interactions were detected more frequently in the SDM framework, and by JSDM, than in the Classic framework. All approaches detected species aggregation more often than segregation, perhaps reflecting the important role of facilitation in stressful high-elevation environments.Differences between the frameworks may reflect the explicit incorporation of elevational segregation in the SDM framework and the sensitivity of JSDM to the environmental data. Nevertheless, all methods have the potential to reveal general patterns of species co-occurrence for different taxa, spatial scales, and environmental conditions.

Habitat filtering and limiting similarity are well‐documented ecological assembly processes that hierarchically filter species across spatial scales, from a regional pool to local assemblages. However, information on the effects of fine‐scale spatial partitioning of species, working as an additional mechanism of coexistence, on community patterns is much scarcer. In this study, we quantified the importance of fine‐scale spatial partitioning, relative to habitat filtering and limiting similarity in structuring grassland communities in the western Swiss Alps. To do so, 298 vegetation plots (2 m × 2 m) each with five nested subplots (20 cm × 20 cm) were used for trait‐based assembly tests (i.e., comparisons with several alternative null expectations), examining the observed plot and subplot level α‐diversity (indicating habitat filtering and limiting similarity) and the among‐subplot β‐diversity of traits (indicating fine‐scale spatial partitioning). We further assessed variations in the detected signatures of these assembly processes along a set of environmental gradients. We found habitat filtering was the dominating assembly process at the plot level with diminished effect at the subplot level, whereas limiting similarity prevailed at the subplot level with weaker average effect at the plot level. Plot‐level limiting similarity was positively correlated with fine‐scale partitioning, suggesting that the trait divergence resulted from a combination of competitive exclusion between functionally similar species and environmental micro‐heterogeneities. Overall, signatures of assembly processes only marginally changed along environmental gradients, but the observed trends were more prominent at the plot than at the subplot scale. Synthesis. Our study emphasises the importance of considering multiple assembly processes and traits simultaneously across spatial scales and environmental gradients to understand the complex drivers of plant community composition.