Background/Question/Methods   Energy availability as indexed by climate has emerged as a strong predictor of broad-scale patterns of species richness.  Yet many non-climate factors may prevent observed richness from reaching the potential richness limit set by available energy.  Most notably, habitat conditions significantly influence species occurrence and can vary greatly among regions with similar available energy.  Our objective was to assess the relative importance of climate and habitat conditions for predicting forest bird richness across the southeastern United States.  We estimated richness (accounting for heterogeneous detection probabilities) for 426 North American Breeding Bird Survey routes that were run annually during a 3-year period.  We derived temperature and precipitation predictors from long- (30-year) and short-term (3-year) climate data.  Habitat-related predictors included elevation, human population density, housing density, road density, and the amount and arrangement of forest cover.  Predictors were measured for a 20-km-radius area centered on each route.  Multivariate adaptive regression splines were used to build predictive models.  We estimated a variable’s predictive importance (PI) based on the increase in model mean square error (MSE) that would occur if that variable was removed from the model.  PI was a relative index that varied between 0 and 100; strong predictors had high PI values.

Results/Conclusions   The minimum-MSE model had an R² of 66.4% and involved four climate variables (long-term annual mean temperature, long-term summer mean precipitation, spatial variation in long-term annual mean precipitation, deviation between short- and long-term annual mean precipitation; average PI = 31.1) and four habitat variables (proportion of forest, mean forest patch size, forest patch size variation, variation in elevation; average PI = 48.3).  A simpler model with a comparable MSE and an R² of 66.3% involved one climate variable (long-term annual mean temperature; PI = 48.3) and three habitat variables (proportion of forest, mean forest patch size, forest patch size variation; average PI = 58.2).  Predictive strength was generally greater for habitat variables than it was for climate variables, indicating that predicting forest bird richness is more complex than implied by species-energy theory.  The two models defined a trade-off relevant to how conservationists may project effects of environmental change on richness.  Parsimony suggests use of the simpler model, with the disadvantage of not using the richer set of environmental predictors in the complex model.  An interest in addressing a greater array of environmental conditions suggests use of the complex model, with the potential disadvantage of lower prediction accuracy due to over-fitting.