Integrating multiple data sources with species distribution models to estimate the distribution and abundance of northern bobwhite (Colinus virginianus) in the United States

• Data integration improves abundance estimation for a species distribution model. • Northern bobwhite abundance in the US is concentrated in three distinct regions. • Effects of environmental covariates on abundance varies regionally across the range. • Northern bobwhite abundance in the eastern US was estimated at 8,577,291 birds. Species distribution models (SDMs) have traditionally focused on occupancy despite abundance potentially being a more useful metric for informing conservation initiatives. Integrating m ultiple species abundance datasets could retain the strengths of each data type and, at least partially, offset their weaknesses, potentially improving the performance of abundance-based SDMs. We developed spatially-non-stationary, abundance-based SDMs to assess the environmental drivers of spatial variation in abundance and to predict the abundance and distribution of northern bobwhite ( Colinus virginianus ) across the United States. We fitted Bayesian SDMs with regionally-partitioned coefficients by integrating structured North American Breeding Bird Survey (BBS) and semi-structured eBird count data. We found that bobwhite abundance was concentrated in three main regions: southern Texas, the Great/Midwestern Plains, and the southeastern coastal plain. Total abundance across the range was estimated at 8,577,291 (8,292,554 - 8,933,202). While the spatial extent of the predicted bobwhite range was generally similar across models, models fit with single data sources appeared to vastly underestimate (eBird) or overestimate (BBS) abundance, though abundance estimation was improved through data integration. Most covariate effects exhibited non-stationarity across the range, potentially leading to inappropriate inferences or management decisions from a spatially-stationary model. Our study provides an important example of how datasets collected at different spatial scales under different observation protocols can be integrated via SDMs to improve abundance-based modeling and correct for weaknesses of individual datasets. Our modeling framework provides regional estimates of the drivers of bobwhite abundance and range-wide estimates of abundance for guiding both local and range-wide bobwhite conservation.