Multiresponse algorithms for community‐level modelling: Review of theory, applications, and comparison to species distribution models

Abstract

Community‐level models (CLMs) consider multiple, co‐occurring species in model fitting and are lesser known alternatives to species distribution models (SDMs) for analysing and predicting biodiversity patterns. Community‐level models simultaneously model multiple species, including rare species, while reducing overfitting and implicitly considering drivers of co‐occurrence. Many CLMs are direct extensions of well‐known SDMs and therefore should be familiar to ecologists. However, CLMs remain underutilized, and there have been few tests of their potential benefits and no systematic reviews of their assumptions and implementations. Here, we review this emerging field and provide examples in r to fit common CLMs. Our goal is to introduce CLMs to a broader audience, and discuss their attributes, benefits and limitations relative to SDMs. We review (1) statistical implementations and applications of CLMs, (2) their advantages and limitations, and (3) comparative analyses of CLMs and SDMs. We also suggest directions for future research. We identify seven CLM algorithms with similar data structures and predictive outputs as SDMs that should be most accessible to ecologists familiar with species‐level modelling, including five methods that predict assemblage composition and individual species distributions and two methods that model compositional turnover along environmental gradients. Community‐level models have been applied to numerous taxa, regions, and spatial scales, and a variety of topics (e.g. studying drivers of community structure or assessing relationships between community composition and functional traits). Studies suggest that the relative benefits of CLMs and SDMs may be case specific, especially in terms of predicting species distributions and community composition. However, CLMs may offer advantages in terms of computational efficiency, modelling rare species, and projecting to no‐analog climates. A major shortcoming of CLMs is their reliance on presence–absence community composition data. Studies are needed to assess the relative merits of SDMs and CLMs, and different CLM algorithms, with a focus on three key areas: (1) under which circumstances CLMs improve predictions for rare species, (2) how CLMs perform under different community compositions (e.g. relative abundance of rare vs. common species), including the extent to which co‐occurrence patterns are structured by biotic interactions, and (3) ability to project across time/space.

Biotic interactions in species distribution modelling: 10 questions to guide interpretation and avoid false conclusions

“…Before putting our faith in analytical results, we must verify whether the modelling approach is suitable for detection and representation of expected biotic interactions. For example, as Box 2 and There are five main approaches for modelling species associations in macroecological species distribution models (see Table 1 for more details; Nieto-Lugilde et al, 2018 andOvaskainen et al, 2017 for a recent review). All essentially look for excess or deficits in cooccurrence, relative to a baseline occurrence rate set by the environment or a random null model.…”

“…Recently, considerable interest has arisen in inferring biotic interactions from large-scale data through the statistical analysis of species (co)distributions. The methods for this task are in substantial flux (Gonz alez-Salazar, Stephens, & Marquet, 2013;Louthan, Doak, & Angert, 2006; Morales-Castilla, Matias, Gravel, & Ara ujo, 2015;Nieto-Lugilde, Maguire, Blois, Williams, & Fitzpatrick, 2018;Ovaskainen et al, 2017;Staniczenko, Sivasubramaniam, Suttle, & Pearson, 2017;Warton et al, 2015;Wisz et al, 2013;Zhang, Kissling, & He, 2012). The general idea of these methods is to look for an excess in co-occurrences (possibly indicating facilitation) and deficits in co-occurrence (possibly indicating competition).…”

Measuring competitive impact: Joint‐species modelling of invaded plant communities

“…Joint-species distribution models (JSDM) are extensions of standard species distribution models that have the potential to measure both competitive impact and species responses to environmental conditions using community composition data from sites along known environmental gradients (Kissling et al, 2012;Nieto-Lugilde, Maguire, Blois, Williams, & Fitzpatrick, 2018).…”

Inference of biogeographic history by formally integrating distinct lines of evidence: genetic, environmental niche and fossil

“…Continued improvements in ENMs are occurring, such as use of non-point occurrence data like range maps (Merow et al 2017), multi-species modeling (Nieto-Lugilde et al 2018, Zhang et al 2018, and accommodations for imperfect detection (Koshikana et al 2017). Another interesting avenue is to incorporate multiple molecular marker types and genomic-scale data.…”