An explanation for the unequal distribution of life forms across the Earth's surface has been a persistent and problematic question in modern ecology ever since these patterns were first noted, over 100 years ago. Most empirical research supports one of three environmental hypotheses to explain these patterns: environmental energy (ambient environmental energy or ecosystem productivity); climatic variability; or habitat heterogeneity. This research examines these hypotheses using better datasets than those commonly considered, and using a consistent methodology that addresses often neglected statistical and analytic details.

The environmental datasets used in this study are derived from time series of satellite and ground station data, including the Daymet climate data, and net primary productivity data from the GLOPEM model. Species richness is derived from the individually modeled vertebrate distributions provided by the individual state Gap Analysis Projects for the western US states of California, Oregon, Washington, Idaho, Montana, Wyoming, Utah, and Colorado, which define the spatial extent of this study. The study methodology relies upon the summary of results from many model variants for each hypothesis. These variants are constructed by creating regression models at each of four different spatial scales (8, 16, 32, and 64 km grid cells), for each class of vertebrates (amphibians, birds, mammals, reptiles, and all), and over each of the eight states considered. Preliminary studies found that ordinary least squares would be a sufficient model form, although conditional autoregressive models were extensively considered. Other preliminary work examined issues of spatial autocorrelation and variable selection.

The results indicate that the energy/productivity hypothesis consistently outperforms all other hypotheses in explaining species richness, across almost all spatial scales, geographic regions, and vertebrate classes. The performance of the climatic variability and habitat heterogeneity hypotheses varies for particular states or vertebrate classes. Vertebrate data quality was important; results for Colorado and Washington were frequently unusual, suggesting an incompatibility between their modeled vertebrate distributions and those of other states. Models of reptile richness also often showed substantially different characteristics than those for other vertebrates. Overall the results provide additional support to the energy/productivity hypothesis, from a more comprehensive methodological basis.