A Comparison of Climate Models

Each of the models discussed here are based on different datasets and analytic approaches. None are “right” or “wrong” – they’re just different. When different approaches all indicate comparable results, this adds credence to all. When different models yield different results, this simply indicates that further research may be needed to clarify the results.

Some of the more important differences among the models are briefly summarized here. The primary issues revolve around (1) source of bird distribution data, (2) source of environmental (climate) data, (3) scale of the analysis grid, (4) extent of the region being analyzed, and (5) statistical procedures that were used.

Bird distribution data. Mass Audubon’s analysis used data submitted to eBird. Only a single observation was used whenever multiple eBird records fell within a single grid cell. Only records for sightings obtained in June and July were used. National Audubon used presence-absence data for 2000-2009 from the Audubon Christmas Bird Count (CBC) and North American Breeding Bird Survey (first 30 stops). US Forest Service  used abundance estimates based on the North American Breeding Bird Survey (abundance estimates).

Environmental (climate) data. Mass Audubon’s analysis used 19 bioclimatic variables, plus elevation, as provided by WorldClim 1.4 (http://www.worldclim.org/). National Audubon used data for 17 climatic variables, obtained from the Canadian Forest Service, which were annually matched to bird data. The US Forest Service used data from 7 climate variables (obtained from the Canadian Forest Service), 4 elevation variables, and 88 tree species distributions.

Scale of the analysis grid. Mass Audubon used an analysis grid of approximately 1 km X 1 km. National Audubon’s grid resolution was approximately 10 km X 10 km.  The US Forest Service used a grid of 20 km X 20 km.

Extent of the region being analyzed. Mass Audubon’s model was based on bird and environmental data from a 30° X 30° area that encompassed eastern North America from Nova Scotia and Labrador west to Ontario and south to northern Florida; results from this large scale analysis were then clipped to the state boundary of Massachusetts. National Audubon’s analysis encompassed Alaska, Canada, and the lower 48 states. The US Forest Service included states of the continental U.S. east of the 100th meridian.

Statistical procedures that were used. Mass Audubon used the default settings of Maxent 3.3.3k with 5 replications (https://biodiversityinformatics.amnh.org/open_source/maxent/). For information about the statistical approaches used by National Audubon see http://climate.audubon.org/sites/default/files/AudubonUSFWS%20LCC%20Climate%20Change%20FINAL%201.1.pdf. For information about the statistical approaches used by the US Forest Service, see https://www.researchgate.net/publication/216770865.