Bachman's Sparrow Species Distribution Model
Bachman's Sparrow locations were recorded in three distinct survey projects conducted by the North Carolina Wildlife Resources Commission (NCWRC) (n = 544; 2006-2013), Paul Tallie et al. (Taillie et al. In review) (n = 99; 2012), and a Virginia Tech study in cooperation with the U.S. Marine Corp at Camp Lejeune (n = 84; 2010). Paul Taillie (NCWRC), Jeffrey Marcus (The Nature Conservancy), Scott Anderson (NCWRC), and John Carpenter (NCWRC) collected data and summarized existing data sources. Data were collected as part of species-specific point count surveys with call-backs, general point counts, and as incidental observations. Points within 250 m of another presence location were removed from the dataset. Fires were quantified from the LANDFIRE program <http://www.landfire.gov/>. The National Land Cover Database classifications of evergreen forest and a combined layer of shrub/scrub and grassland/herbaceous depicted coarse land cover classes. Tree canopy cover was quantified with the NLCD. All of these variables were calculated with 3x3 and 9x9 neighborhood statistics. Connectivity was quantified using local connectedness from The Nature Conservancy's southeastern Terrestrial Resilience dataset (Anderson et al. 2014). This connectivity measure emphasized fragmentation caused by agriculture and urban development. After initial model development, we calculated the amount of habitat within a 5-km radius and included this result in the model.
A resource selection function modeling approach was used (see Boyce et al. 2002) with pseudo-absences dispersed throughout the survey extent; row crop agriculture and urban classifications (NLCD: land cover classes 21-24) were excluded. A logistic regression compared Bachman's Sparrow presence with pseudo-absence points. This compared Bachman's Sparrow habitat use vs. availability. As results of logistic regression are not directly relevant to assessing a resource selection function, data were divided into training data (60%) and validation data (40%). RSF's are relative measures of habitat use, but we wanted to determine a threshold for estimating presence/absence. Similar to other studies (Wilson et al. 2013; Fedy et al. 2014) we determined the optimal classification threshold with the training data, and applied it to the validation data. We were primarily interested in how well the model performed in predicting presence and the proportion of area predicted.
The results showed canopy cover, evergreen land cover, canopy cover SD, fires, and connectedness were related to Bachman's Sparrow. After projection of these variables, we also discovered an association with habitat within a 5-km radius of a cell. The confusion matrix of the validation results showed 89% classification accuracy and further results are presented below:
- BACS Resource selection function model predicted 13% of survey extent as "present" and captured 84% of BACS presence locations in North Carolina.
Questions about Bachman's Sparrow modeling may be directed to Bradley Pickens, bapicken@ncsu.edu
Known Issues:
For Bachman's Sparrow, the model is trained and validated with North Carolina data and is extrapolated to the whole South Atlantic LCC. More data is needed to validate this extrapolation. However, habitat variables were chosen with specific reference to Bachman's Sparrow ecology. Fires, canopy cover, evergreen trees, and fragmentation are all known to be factors affecting Bachman's Sparrow throughout the southeast. Definitive absence data would be helpful for verifying the presence/absence threshold chosen here.
Literature Cited
Anderson MG, Barnett A, Clark M, Ferree C, Sheldon AO, Prince J (2014) Resilient Sites for Terrestrial Conservation in the Southeast Region. The Nature Conservancy, Eastern Conservation Science. 127 pp.
Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA (2002) Evaluating resource selection functions. Ecol. Model. 157:281-300
Fedy BC, Doherty KE, Aldridge CL et al (2014) Habitat Prioritization Across Large Landscapes, Multiple Seasons, and Novel Areas: An Example Using Greater Sage-Grouse in Wyoming. Wildl. Monogr.:1-39
James FC, Hess CA, Kicklighter BC, Thum RA (2001) Ecosystem management and the niche gestalt of the red-cockaded woodpecker in longleaf pine forests. Ecol. Appl. 11:854-870
NatureServe (2014) NatureServe Central Databases. Arlington, Virginia. U.S.A.
Taillie PJ, Moorman CE, Peterson MN (In review) The relative importance of multi-scale factors on the distribution of Bachman’s Sparrow and the implications for ecosystem conservation. The Condor: Ornithological Applications
The National Bobwhite Technical Committee (2011) The National Bobwhite Conservation Initiative: a range-wide plan for re-covering bobwhites. National Bobwhite Technical Committee Technical Publication, ver. 2.0 In: Palmer W. E., T.M. Terhune, and D.F. McKenzie (ed). Knoxville, TN,
U.S. Fish and Wildlife Service (2003) Recovery plan for the red-cockaded woodpecker (Picoides borealis): second revision. U.S. Fish and Wildlife Service, Atlanta, GA, pp. 296
Wilson JW, Sexton JO, Jobe RT, Haddad NM (2013) The relative contribution of terrain, land cover, and vegetation structure indices to species distribution models. Biological Conservation 164:170-176
