Testing metapopulation models with stream-fish assemblages

We assessed the utility of simple metapopulation models in an analysis of a decade (1976–86) of stream-fish censuses at 10 sites on the Cimarron River, Oklahoma. This assemblage is a good candidate for metapopulation analysis because there was substantial annual turnover of population, with frequent local extinctions and recolonizations. However, annual probabilities of colonization and extinction were usually not correlated with the fraction of sites occupied, a key prediction of many Levins-style metapopulation models. Instead, these probabilities were related to position in the stream gradient. For most species, colonization was less likely and extinction more likely in upstream than downstream sites. Consequently, the simple Levins-style metapopulation models failed to accurately predict the dynamics of most species. Metapopulation models that incorporate spatial variability in colonization and extinction probabilities may be more successful for populations that are distributed across environmental gradients.