Many investigators have used bird survey data to develop contour maps of bird abundance based on mean counts on survey routes. Root (1988) provided a grid of smoothed relative abundances for species observed on Christmas Bird Counts. Sauer and Droege (1989) mapped relative abundances of Eastern Bluebirds (Sialia sialis) just after severe winters in the mid 1970's and after their populations returned to pre-winter levels. We have also used relative abundance maps to document the ranges of several species (e.g., Droege and Sauer 1990). See Sauer et al. (1995 a and b) for applications and discussions regarding mapping of survey data.
These maps are based on exactly the same data that are used in the BBS trend analyses, and route summaries are simple averages of counts on routes over time. Please note that these simple averages do not account for observer differences in counting ability or for other factors that could be controlled in more sophisticated analyses. We also note that the BBS data are edited to remove data that are of questionable quality or represent birds that are thought to be migrating rather than breeding. Please refer to the metadata for the BBS dataset for more information on editing and quality control of the BBS data.
We developed a map of starting locations of BBS routes. Latitude and longitude (degree-minute) of the starting locations were taken from topographic maps of the route path. Of course, the route is 24.5 miles in length, hence any point used to characterize the route is arbitrary.
We estimated average counts from the interval 2011 - 2015 on each route for each species, and copied them into database files. We developed contour maps of bird relative abundances, using the route relative abundances as input to smoothing procedures (Isaaks and Srivastava 1989, Cressie 1992).
We used inverse distancing (Isaaks and Srivastava 1989) to prepare a smooth of the data. This procedure estimates the abundance at a location as a distance-weighted average of counts from nearby survey routes. We used inverse distancing to estimate abundances for a grid of points overlaid on the survey area, then used Arc/Info to make a contour map from the estimated abundances (Environmental Systems Research Institute 1991). See below for more datails of the analysis.
Arc/Info provided an arc coverage of contours that connect points having the same value. Depending on the maximum relative abundance of the species, we used levels of 1, 3, 10, 30, and 100 for contours. The maps end at a minimum level of 0.1, which was chosen as a possible edge-of-range index after some comparisons of contours with known edges of ranges (S. Droege and D. Bystrak, Personal Communication), and the larger cutpoints were chosen as a series of powers of 3, rounded up for ease of presentation.
The maps presented here are quite similar to the maps in the those of earlier versions of the Home Page, but several differences exist between the procedures used to prepare the earlier maps and these maps. To make the 1966 - 1992 maps, we used Kriging, a procedure in which a variogram is estimated for the species and is used to define the distance-covariance relationship for the smooth (Cressie 1992). In theory, the Kriging should provide a more accurate surface than a procedure such as inverse distancing, which never uses information from the data to adjust the weighting. However, in our experience the variograms were not particularly informative, suggesting that at the scale of the BBS there is little advantage in using Kriging.
We acknowledge, however, that the maps provided here are designed to provide a large-scale summary of the data, and if a species is of particular interest, a more intensive analysis should be conducted using Kriging or some other smoothing procedure. Kriging is a model-based estimation procedure, and if the model is appropriate for the data we can put confidence intervals on the resulting surface. By developing a semivariogram model that more accurately portrays the spatial covariance among routes, the resulting Kriged surface will better reflect the patterns of change among the routes. Often features such as directionality of the semivariogram and trend in the data will require the use of more complex models. See Isaaks and Srivastava (1989) for a useful discussion of the technical details of fitting semivariograms to data, and Cressie (1991) for a more technical discussion of all aspects of spatial modelling.