Differentiating between niche and neutral assembly in metacommunities: When null models fall short

Background/Question/Methods

In ecology, one attempt to link patterns in observational data with biological mechanisms is to compare diversity metrics with null distributions. Recently, this includes null deviation methods, which use presence/absence or abundance data to discern between niche and neutral community assembly forces while controlling for variation in alpha- and gamma-diversity. Despite increasing use in empirical studies, null deviation metrics have not been rigorously tested against communities with known assembly mechanisms. Using a simulated metacommunity of 25 species and 25 patches, we tested the ability of presence/absence and abundance-based null deviation metrics to distinguish between niche and neutral assembly under deterministic and stochastic processes. We examined null deviation values through time and across a gradient of pure niche ("species-sorting") to neutrally structured communities. We also compared the effect of starting conditions on the metrics’ performances. 

Results/Conclusions

In deterministic simulations, both presence/absence and abundance-based null deviation metrics properly distinguish between niche and neutral communities. The presence/absence metric exhibits a threshold between niche and neutral, while the abundance-based metric is able to capture intermediate communities. In stochastic simulations, the presence/absence metric fails to properly distinguish between assembly processes, and in fact detects neutrality in niche-structured communities and vice versa.  In comparison, the abundance-based metric correctly distinguishes between assembly processes even with demographic stochasticity. However, when legacy effects are incorporated into the model, the metric is no longer guaranteed to correctly predict community assembly mechanisms. In all simulations, null deviation values are constant across time in niche-structured communities, but are susceptible to population drift in neutral communities. We conclude that great caution is needed when applying these metrics to empirical data. We suggest only using the abundance-based metric for comparison studies with low stochasticity and equivalent starting conditions such that error from drift and legacy effects are minimized.