Quantitatively Testing Predictions From Mechanistic Models: A Case Study for Island Biodiversity

A key test of an ecological model is whether it can quantitatively predict unseen aspects of data not used in model fitting. Previous studies have fitted models to data on island alpha diversity, but did not test how well these models predict other patterns of biodiversity. Here, we test the extent to which models fitted only to island alpha diversity can predict three other patterns of island biodiversity, including similarity of species composition among islands. We found that a neutral model produced decent predictions for 17 archipelagos, with the proportion of data points within the model's confidence intervals closely tracking the nominal coverages of the intervals, differing by 0.19 on average. Thus, as a first approximation, observed patterns of island biodiversity can be parsimoniously explained as the result of neutral competition and dispersal limitation. More broadly, our results demonstrate that neutral models can make accurate predictions of higher-order diversity statistics.