Predicting richness and composition in mountain insect communities at high resolution: a new test of the SESAM framework

Aim The aim of this study was to test different modelling approaches, including a new framework, for predicting the spatial distribution of richness and composition of two insect groups. Location The western Swiss Alps. Methods We compared two community modelling approaches: the classical method of stacking binary prediction obtained fromindividual species distribution models (binary stacked species distribution models, bS-SDMs), and various implementations of a recent framework (spatially explicit species assemblage modelling, SESAM) based on four steps that integrate the different drivers of the assembly process in a unique modelling procedure. We used: (1) five methods to create bS-SDM predictions; (2) two approaches for predicting species richness, by summing individual SDM probabilities or by modelling the number of species (i.e. richness) directly; and (3) five different biotic rules based either on ranking probabilities from SDMs or on community co-occurrence patterns. Combining these various options resulted in 47 implementations for each taxon. Results Species richness of the two taxonomic groups was predicted with good accuracy overall, and in most cases bS-SDM did not produce a biased prediction exceeding the actual number of species in each unit. In the prediction of community composition bS-SDM often also yielded the best evaluation score. In the case of poor performance of bS-SDM (i.e. when bS-SDM overestimated the prediction of richness) the SESAM framework improved predictions of species composition. Main conclusions Our results differed from previous findings using community-level models. First, we show that overprediction of richness by bS-SDM is not a general rule, thus highlighting the relevance of producing good individual SDMs to capture the ecological filters that are important for the assembly process. Second, we confirm the potential of SESAM when richness is overpredicted by bS-SDM; limiting the number of species for each unit and applying biotic rules (here using the ranking of SDM probabilities) can improve predictions of species compositio