Using the BAM database to advance methods for estimating population trends for landbird species in North America. 

Project Summary

Given poor coverage of Breeding Bird Survey (BBS) data in the boreal region, we have looked to other sources of data to augment the BBS for the purposes of trend estimation. Although the BAM dataset is ad-hoc and temporally sparse, it has grown substantially in extent and duration, allowing us to explore several hybrid methods for using it — in conjunction with BBS data — to estimate population trends. In the last year, with input from Environment Canada and Climate Change (ECCC) statisticians and biologists, we have identified several options for further exploration, one of which is a direct extension of our generalized national models. In regions not well represented by BBS, this relatively robust approach can provide an alternative to estimates based solely on repeated BBS counts. It is based on the development of spatio-temporal abundance models that combine data from multiple years to quantify habitat relationships, while considering inter-annual variation in abundance. These new generalized national models (see previous section, page 10) were separately constructed for each BCR sub-region within Canada (south of the Arctic) and hemi-boreal portions of the United States, thus ensuring that annual density estimates are regionally relevant.

To balance spatio-temporal coverage of input data and thereby limit the influence of sampling bias, we spatially and temporally stratified and subsampled the data for modeling purposes. We also controlled for the effects of spatial and temporal variation on abundance by including sources of temporal (sampling year) and spatial (climate, terrain, and vegetation) variation as direct covariates in our models. The boosted regression tree modelling approach that we used captures non-linear and interactive habitat relationships, thus resulting in relatively fine-scale (1-km resolution), spatially heterogeneous, annual predictions. These annual, pixel-level estimates can then be “rolled up” to estimate trends for a variety of different geographies and time periods. Trends are based on a combination of direct predicted changes as a function of changes in “habitat supply” (vegetation) over time, and unexplained (“residual”) variation in abundance that may be attributed to a variety of potential (unmapped) predictors, including wintering ground and migration conditions.

Further testing is still required, but preliminary results suggest that BAM data provide the best available trend estimates in areas not well-sampled by the BBS. We are working to identify the specific areas for which BAM trend estimates constitute an improvement over BBS-based trends. 

This work is led by Diana Stralberg and Péter Sólymos. For more information please contact us.