Comparing theory to data: What, when, and how we explain the species-area relationship

Though the species-area relationship (SAR) is a mainstay of ecology, much confusion revolves around its behavior and generation.  Models used to explain the SAR usually fall into one of two categories: mechanistic models that include ecological processes such as dispersal, competition, and disturbance; and statistical sampling models that usually begin with the species abundance distribution and describe the increase in species as number of individuals sampled (area) increases.  In either case, these models are often tested in a limited fashion by comparing a single model to at most one SAR from each of several ecosystems.  Using data from 35 SARs collected on serpentine grassland, I illustrate the spatial and temporal variation in the SAR across sites and years that models attempting to explain the SAR should be able to describe.  These SARs vary across space and time between forms indicative of a power law, curvilinear relations indicative of random placement, and even sigmoidal patterns.  Moreover, models often implicitly assume a fully-nested sampling design, while data used to generate a SAR data are usually collected in a spatially nested manner or by aggregating disjunct plots to yield larger area.  Each of these sampling designs yields a different functional form of the SAR and so comparison to model predictions can be deceptive.