We investigated the ability of ecological neutral models to predict patterns of species abundance. We found differences in both the number of species (richness) and species dominance between observed and modeled communities of zooplankton and aquatic invertebrates. Shown are data from a metacommunity of 49 small rock pools. The three pools shown differ substantially in volume and total community size, J. Each bar represents the multi-year average relative proportion of detritivore species represented at each abundance rank, in decreasing order from left to right.

The data were collected over a 13-year period, recording changes in the densities of generalist predators and their detritivore prey. We divided the species abundance data into two groups representing periods of high predator density (HPD; predators > 21.5 percent of community; red bars) and low predator density (LPD; predators < 21.5 percent; blue bars). Each pool experienced approximately equal periods of HPD and LPD. High predator density was associated with an increase in the diversity of detritivores. On average, the number and evenness of prey species increased during HPD periods. Black vertical lines represent standard error.

The grey, yellow, and green bars show the species proportions predicted from neutral models that were parameterized using data from the observed communities. The predictions were based upon a different migration probability (m): grey bars, m = 1.00, yellow bars, m = 0.001, green bars (bottom panel), m = 0.01. The agreement between observed proportions and modeled proportions for different values of m changes with predator density. This indicates that a key parameter of neutral models, the migration probability, is influenced by cross-trophic species interactions (i.e. predation).

In Hubbell's neutral community model, the number of species in a community, and the proportion of the community represented by each species, depend on four parameters: 1) J, the community size (number of individuals), 2) S, the number of species in the metacommunity, 3) P_i, the proportions of each species in the metacommunity, and 4) m, the probability of migration between community and metacommunity. We set the value of J to the size of each pool, estimated from samples, and established values for S and P_i using Hubbell's metacommunity algorithm (Hubble 2001). This approach leaves only one parameter, m, left to estimate.

The value of m determines the species richness and distribution of species relative proportions in a model community. To find the putative value of m that best fits the observed species patterns of each pool we compared the species richness and relative proportions of the observed communities to those predicted by neutral models parameterized with different migration probabilities. For year of data, each pool contains a specific number of individuals, J and species, S. We plotted the J-S distribution for each pool and compared it to that of the corresponding model community using a Wilcoxson matched pairs test. Each pool sample and simulated community also generates a distribution of ranked species proportions. For each pool, we compared the average communitiy proportion represented by each rank to that of the model communities, using a Chi-square test. We used migration probabilities that span four orders of magnitude: 1.0, 0.10, 0.01, and 0.001.

We combined the results of the Wilcoxson and Chi-square tests to determine the value of the migration parameter, m, that best characterizes the three pools. The best estimate of m is the value that yields no significant difference (P > 0.05) between observed and simulated communities for both species richness and species relative proportions. Thus, when both test statistics are not significant for a given value of m, we consider the tests to have been successful in estimating the value of m for that community. We consider the assignment of m for to have failed for a particular community if the P-value of both tests is not > 0.05 for any of the four values of m tested, or if more than one value of m yields a non-significant result for both tests.