Upper Midwest Environmental Sciences Center
Our first task was to build statistical models and map predicted bird abundance across the Prairie-Hardwood Transition, also known as Bird Conservation Region (BCR) 23. This BCR lies entirely within the U.S. Fish and Wildlife Service (FWS) Region 3, overlaps three FWS ecosystem regions, and is virtually identical to Partners-in-Flight Physiographic Area 16 (Upper Great Lakes Plain). We chose this region because of its diverse land uses, both historical and current. The methodology we describe here was employed with only slight modification where necessary for subsequent modeling efforts in other regions.
We chose to model 13 species of high conservation concern in BCR 23. Twelve of the 13 species are counted by the North American Breeding Bird Survey (BBS). The 13th species, the American Woodcock (Scolopax minor), is more properly counted by singing ground surveys. We have expanded the focus our work to three additional BCRs comprising the majority of the breeding range of the Cerulean Warbler: the Appalachian Mountains (BCR 28), Central Hardwoods (BCR 24), and the Lower Great Lakes/St. Lawrence Plain (BCR 13).
To model species-habitat relations, we used all BBS routes within a BCR plus any additional routes within 50 km of the study area. We included these additional routes to minimize the influence of edge effects when predicting abundance at the edges of the region. For example, there were 124 BBS routes in BCR 23, with an additional 80 routes within 50 km of the study area. Each BBS route has 50 evenly spaced survey locations (stops) at which an observer counts all birds seen or heard in a 3-min period. The sum of counts from the 50 stops in a year's route survey was used as an index to abundance along the route for that year. We used counts collected from 1981 to 2001 to coincide temporally with our land cover data, derived from satellite imagery taken in the late 1980s and early 1990s.
We chose variables relevant to birds based on a review of literature and expert opinion (Regional FWS biologists and migratory bird coordinators). Variables included in our analysis were those that could be measured remotely across the region, including land use composition and configuration, climate, terrain, and human influence. Each variable was assessed at three logarithmically-related scales, 800-, 8,000-, and 80,000-ha. These extents correspond to the mean product of zones within 0.1-, 1-, and 10-km, respectively, of each BBS route.
To predict bird abundance at locations where surveys were not conducted, these environmental variables were related to BBS counts in a statistical model. This statistical model, described as a hierarchical spatial linear model, accommodates variation in bird counts due to differences among observers, similarity in counts over years and space, and other factors identified as nuisances obscuring the relevant relations between bird abundance and habitat. Coefficients from these linear models were used to combine relevant data layers in a geographic information system (GIS), yielding for each location in the region a predicted abundance.
We evaluated our models with independent sources of data as much as possible. We withheld from model construction random sets of BBS data and used these data in a model evaluation. For example, in BCR 23 we withheld 64 of the 204 BBS routes occurring within the study area. For each route used in validation, we calculated a mean predicted abundance. Abundance from the validation route was assessed against this mean predicted abundance. Additional sources of independent model assessment included the use of point counts of birds collected at 21 study areas within BCR 23 and data provided by the Cerulean Warbler (Dendroica cerulea) and Golden-winged Warbler (Vermivora chrysoptera) Atlas Projects (Rosenberg et al. 2000). The Atlas data only included known cerulean warbler and golden-winged warbler locations and, therefore, did not provide insight into how well the model predicted areas of absence. Model evaluation was most successful when bird survey data, independent from the BBS, was readily available. The Bird Point Count Database is a central repository for bird point count data collected across the United States and data can be selected and downloaded from the web site.
For more information on methods see:
Thogmartin, W. E., and M. G. Knutson. 2007. Scaling local species-habitat relations to the larger landscape with a hierarchical spatial count model. Landscape Ecology 22:61–75. URL: http://dx.doi.org/10.1007/s10980-006-9005-2.
Thogmartin, W. E., M. G. Knutson, and J. R. Sauer. 2006. Predicting regional abundance of rare grassland birds with a hierarchical spatial count model. Condor 108(1):25–46.
Thogmartin, W. E., J. R. Sauer, and M. G. Knutson. 2007. Modeling and mapping abundance of American woodcock across their breeding range in the United States. Journal of Wildlife Management 71(2):376–382.
A Hierarchical Spatial Count Model with Application to American Woodcock
Mapping Peaks in Predicted Abundance
We modeled patterns in rare bird abundance to identify areas in which to focus management and conservation effort. We were specifically interested in predicting peaks or 'hotspots' of relative abundance. These areas of concentrated abundance are more readily conserved than areas in which the species are more diffusely distributed, allowing for the efficient expenditure of meager conservation resources.
Another benefit of mapping hotspots is that they allow for assessing the private lands considerations necessary for conserving these species. We related the species-specific hotspots against federal, state, and tribal lands and identified areas of concentrated abundance occurring on private lands. These private lands hotspots may require a different course of management (e.g., easements, land acquisition, landowner subsidization) than hotspots on governmentally-managed lands. They also present an opportunity to understand how rare birds may co-exist outside of direct governmental protection. The figure below shows the original Bobolink predicted relative abundance model before hotspot mapping.
To identify peaks in predicted abundance, a "smoothed" model was created by applying the ESRI "Focal Statistics" tool to the original predicted relative abundance model for each species. This tool was used to calculate a mean value for each cell in the model based on a 21-cell radius, circular focal window with a 300 meter cell size. The 21-cell radius was selected based upon trial-and-error to create isopleths of data advantageous to highlighting hotspots of elevated values. The figure below shows the results of this smoothing process.
Next, areas determined to be non-habitat according to the National Land Cover Dataset (NLCD) 1992 were removed from the smoothed model and the remaining predicted relative abundance values within the model’s value attribute table were ordered from largest to smallest and the top 5% of these values (by area) were identified. The lower limit of these identified values was used as the cut-off to determine “Hotspots” within the smoothed model. In the case of the Bobolink example this was identified to be a predicted relative abundance value of 14.56. The following figure highlights those areas with predicted relative abundance values greater than or equal to 14.56. The figure below shows these hotspots with a black outline.
Next, an ESRI shapefile depicting boundaries around the calculated hotspots was created and used to calculate Zonal Statistics on the smoothed model using the "Zonal Statistics as Table" tool. This table was then used to calculate the top ten hotspots with a 50% weighting given to the MEAN of the predicted relative abundance values within the hotspot and a 50% weighting given to the SUM of the predicted relative abundance values within the hotspot. This was done to take into account not only relative intensity of predicted relative abundance but also relative size of the hotspot. The following figure displays the top 10 hotspots with a thick black outline.
The hotspots were then mapped in relation to their proximity to federal, state, and tribal management. Tabular summaries were also created using the "Zonal Statistics as Table" tool. The non-habitat values as determined using the NLCD 1992 were inserted into the model and given a value of zero for use in the tabular summaries. The summaries were done to compare individual states and hotspots according to their federal, state, and tribal management status. The percentages of managed and unmanaged lands were calculated using the area totals and also the sum totals for the cells in the model. The figure below shows the smoothed model with the non-habitat NLCD 1992 mask included.
Hamel, P. B. 2000. Cerulean Warbler (Dendroica cerulea). The Birds of North America 511 (A. Poole and F. Gill, Eds.). Academy of Natural Sciences, Philadelphia, Pennsylvania.
McNab, W. H., and P. E. Avers. 1994. Ecological subregions of the United States. U.S. Forest Service, WO-WSA-5, Washington, DC.
Rosenberg, K. V., S. B. Barker, and R. W. Rohrbaugh. 2000. An atlas of Cerulean Warbler populations. Cornell Lab of Ornithology, Ithaca, New York. 56 pp.