At the Fort Bragg Military Installation, an army base in North Carolina, the habitat associations for the breeding bird species and the effects of the current prescribed fire program on the avifauna are virtually unknown. Fort Bragg encompasses one of the largest, fire-dependent longleaf pine systems existing today and is a mosaic of forested habitats. I used bird count data collected during 1994-1997 at 50-m fixed radius point count stations to examine bird species-habitat relationships in relation to fire treatment (i.e., fire intense longleaf pine woodlands versus fire suppressed mixed pine-hardwood and hardwood forests) and a riparian-upland habitat gradient, and at multiple spatial scales (i.e., the microhabitat and landscape). I used two-way factorial analyses to test for the effects of fire treatment and the riparian-upland habitat gradient on total bird abundance, species richness, and species relative abundance. To examine species-habitat associations at multiple spatial scales, I measured vegetation characteristics at a 50-m radius microhabitat scale, and I quantified landscape structural attributes at a 300-m to 1500-m radius landscape scale using a GIS database and the spatial analysis program FRAGSTATS. I then used logistic regression to determine which microhabitat and landscape variables were associated with the probability of occurrence for each species and which spatial scale was of greater relative importance to a species' occurrence. Finally, I tested logistic regression (LR) models and multiple linear regression (MLR) models, specific to the microhabitat scale, with independent data to evaluate their usefulness at predicting the occurrence and relative abundance for several breeding bird species.

Total bird abundance did not vary across fire treatment and species richness may be only slightly greater in fire suppressed habitats, even though this habitat offered greater structural complexity than the park-like longleaf pine, fire intense habitats. Both total bird abundance and species richness were highest within the riparian habitat of streamhead pocosins, which offered distinctive vegetative characteristics otherwise lacking in this landscape. The fire treatment and riparian-upland habitat gradient also were greatly associated with the relative abundance of many species. Four bird species assemblages were defined based on the relative abundance patterns across fire treatments and the riparian-upland gradient: longleaf pine, fire suppressed, drain (i.e., riparian habitat), and generalist assemblages. Continued longleaf pine restoration using growing season prescribed fire likely will cause a decline in species of the fire suppressed assemblage in mixed pine-hardwood and hardwood forests, including many Neotropical migrant songbirds, but will greatly benefit members of the longleaf pine assemblage, such as the Red-cockaded Woodpecker, Brown-headed Nuthatch, Prairie Warbler, and Bachman's Sparrow.

Breeding bird distributions in this fire-influenced, forest-dominated system were associated with attributes at both microhabitat and landscape spatial scales, though microhabitat attributes generally were of greater importance for the occurrence of most species. Microhabitat variation associated with the fire management gradient (intensely burned habitat versus fire suppressed habitat) and the riparian-upland gradient were the most frequent predictors in the species-habitat models. These results are similar to other studies documenting that microhabitat features were more influential than landscape features for birds in a naturally patchy or forest-dominated landscape.

The microhabitat LR (probability of occurrence) models performed best in presence/absence classification when tested with the same data used for model development (cross-validation tests), and the LR and MLR (relative abundance) models performed better for an independent two-year data set compared to an independent one-year data set (validation tests). Although most MLR models were not significantly biased when tested with an independent two-year data set, these models had relatively low precision, suggesting they can be used to predict species relative abundance across a large area but they may not be sensitive to changes in abundance at individual count stations. These model validation results suggest that modeling species occurrence, rather than both occurrence and relative abundance, would have been sufficient to describe general species-habitat associations and to produce reliable, predictive models sensitive to changes in microhabitat structure and composition.