97th ESA Annual Meeting (August 5 -- 10, 2012)

COS 28-4 - Modelling bird populations in the pacific northwest: Implications for species responses to recent climate change

Tuesday, August 7, 2012: 9:00 AM
B113, Oregon Convention Center
Javier Gutierrez Illan, Department of Biology, University of York, York, OR, United Kingdom, Chris D. Thomas, Department of Biology, University of York, York, United Kingdom, Julia A. Jones, College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, OR, Barbara J. Anderson, Department of Biology, UKPopNet, York, United Kingdom, Susan M. Shirley, Department of Forest Ecosystems and Society, Oregon State University, Corvallis, OR and Matthew G. Betts, Forest Ecosystems and Society, Oregon State University, Corvallis, OR
Background/Question/Methods

Climate change is predicted to lead to widespread changes in population dynamics, but quantitative predictions have rarely been tested empirically. We analysed historical abundance data of bird species continuously sampled during breeding season from 1970 to 1974 and from 1998 to 2002 over 756 routes from the Breeding Bird Survey (BBS) covering western North America. In total, 151 species satisfied the criteria for analysis. We selected monthly climate variables interpolated at 1 km resolution to derive a standard set of seven biologically meaningful climate predictors, derived from PRISM climate data. We then ran Boosted Regression Trees to test for the effects of the selected climate variables on bird population density. To control for the influence of climate in the surrounding areas of each route, models were developed at two spatial scales (1 km and 10 km buffer). Models were evaluated both by internal evaluation (testing model fit to the data used to build models) and testing against data from other time periods (i.e., backcasting, forecasting).

Results/Conclusions

Climate envelope models are able to predict bird species´ abundances in the study system. Approaching 90% of species produced highly significant models (P<0.001), based on model evaluation, but the explanatory power of models varied greatly in their predictive power (up to half of the variation in abundance). Model backcasts and forecasts also predicted spatial variation in abundance for more than 75% of species (at P<0.001), showing that abundance patterns can be predicted through time. However, as expected, tests against data from the building time period performed better (by ~40%) than predictions between years. Whilst climate-abundance models performed well for some species, they were quite ineffective for others. Relative model performance was consistent across species between spatial scales (all R2>0.72, P<0.001, n=151), using both verification and cross-validation, indicating that the conclusions are robust to variation in the spatial resolution of the environmental data.

We discuss the potential causes of variation in model fits among species, including the possible impacts of land-use change and changes in interspecific relations. The consequences for future responses by birds to climate change are considered.