On the Interpretations of Joint Modeling in Community Ecology

In an era of global changes, developing reliable biodiversity models has become important research area. Species distribution are the common tools to understand and predict distributions species across space time. However, they fail explicitly account for interactions. To this aim, joint were introduced tease apart effect environment from that interactions, improve rare modeling, functional traits, predictive power models. Nevertheless, most announced advantages have remained unfulfilled, there is still a need better integrate interactions in response environmental change. Explaining modeling communities more than ever central goal ecology. Recently, (JSDMs), which extend (SDMs) by considering correlations among species, been proposed community analyses predictions while potentially inferring Here, we illustrate mathematical links between SDMs JSDMs their ecological implications demonstrate JSDMs, just like SDMs, cannot separate effects biotic We provide guide conditions under (or not) preferable modeling. More generally, call uptake clarification novel statistical developments field prediction value(s) one or variable(s) given other variable(s). Conditional could be derived through use residual correlation matrix. variables used (see later). article, covariates represent abiotic conditions. A missing covariate variable not included model but on variables. flexible generalization ordinary linear regression exponential family (Poisson, binomial, etc.) assuming some known transformation mean function predictor written multiple levels (hierarchical form). Hierarchical allows sharing information entities (mostly herein) facilitate parameter estimation, advantage commonly referred as ‘borrowing strength’. values two Joint directly observed usually irrespective Marginal typical output JSDMs. physiological dependence environment. relationships This outcome both parameters describe covariates. (joint) interpreted descriptions species’ niches. matrix after accounting interest modeled predicted. mostly presence–absences species.