Development and Testing of Linear Regression Models Predicting Bird-Habitat Relationships

Abstract

We used an existing forest inventory data base to develop models predicting the abundance of birds collected during the summers of 1983-85 in a mixed-conifer forest of the western Sierra Nevada. Stepwise multiple linear regression was used to develop models for 21 species of birds. Adjusted coefficients of multiple determination (R2) were low, ranging from 0.02 to 0.24. We used 1984 count data to validate models developed during 1983 (same place, different time validation). Most predictions ranged from 2550% underestimates of observed values. We combined 1983 and 1984 data to produce models used to predict count data collected during 1985 from different locations (different place, different time validation). Predictions were about 50-75% underestimates of observed values. Most observed values were, however, within the confidence intervals generated from the predictive equations. Although our final regression models were successful in predicting presence-absence of most species, it is doubtful that forest inventory systems can be used to predict bird abundance. J. WILDL. MANAGE. 51(1):247-253 The development of easily interpretable models that predict the distribution and abundance of wildlife is necessary for decisions regarding the management of natural resources. Such models can be developed using literature sources and/or expert opinion, such as the Habitat Suitability Index Model Series of the U.S. Fish and Wildlife Service (U.S. Fish. and Wildl. Serv. FWS/OBS-82/10) and from empirical data that has biological importance to the animal. In either case the models may, or may not, be subjected to a validation procedure. Validation allows confidence levels of model predictions and use for management decisions, testing of the adequacy of available habitat inventory procedures for predicting species' reThis content downloaded from 157.55.39.163 on Wed, 21 Sep 2016 05:51:23 UTC All use subject to http://about.jstor.org/terms 248 BIRD-HABITAT RELATIONSHIPS * Morrison et al. J. Wildl. Manage. 51(1):1987 sponses to management activities, refining of habitat inventory procedures to increase model reliability, and identification of fallacious model assumptions (Marcot et al. 1983). The intent of our study was to develop and test, through multiple linear regression analysis, bird-habitat models that serve as descriptors of the ability of a habitat variable to predict the abundance of a species. We used a previously existing vegetation data set, which was collected to predict and guide timber harvest operations, to determine if such data sets can be successfully used to analyze bird abundances and habitat relationships. We thank P. N. Manley, W. M. Block, K. A. Milne, S. A. Laymon, R. Etemad, and J. M. Anderson for assistance with field work during part of the study. D. E. Capen, R. N. Conner, and B. G. Marcot reviewed earlier drafts, and L. M. Merkle helped with manuscript preparation. The assistance of R. C. Heald and other members of the Blodgett For. Res. Stn. was appreciated.