Predicting habitat quality for grassland birds using density- habitat correlations.

Predictive models relating densities of 5 grassland bird species to habitat variables were developed and tested using stepwise multiple regression and principal components regression. For most species the models gave poor quantitative estimates. Because there were distinct habitat differences among the study sites used to generate and the study sites used to verify the models, some models were required to extrapolate beyond the data used to generate them and thus failed to give accurate quantitative predictions. Other models produced predictions of low precision, whereas some may not have taken into account limiting factors not related directly to the habitat variables measured. Such problems will be encountered in most model building attempts and a strategy of continual model updating may provide a way to develop empirical regression models with more reliability. The scale at which habitats are measured to develop regression models relating density to habitat features may be too coarse to account for significant variation among individual breeding pairs in a given habitat. Nesting success, fledgling weights, and properties of nestling growth curves of individual nesting attempts can be related to habitat variables and may provide more information on habitat quality than density-habitat regressions. J. WILDL. MANAGE. 50(4):556-566 Habitat quality can be defined as the suitability of an area to support a reproducing population of a given species or group of species. The evaluation of habitat quality is a prime concern in development of management plans, impact assessments, mitigation studies, and multiple use strategies. It has long been realized that habitats have many effects on populations (e.g., Leopold 1933), so many researchers have attempted to approach the assessment of habitat quality using multivariate statistical methods (James 1971; Whitmore 1975, 1977; Smith 1977; Capen 1981; Collins et al. 1982). Multivariate habitat models are data intensive and require a substantial investment of time and money to collect and analyze the necessary data (Marcot et al. 1983). Less rigorous procedures can circumvent such expenditures. A widely used method developed by the U.S. Fish and Wildlife Service (USFWS) is the Habitat Evaluation Procedure (HEP). This method attempts to develop qualitative models reflecting habitat quality using a less data-intensive approach (Flood et al. 1977, Division of Ecological Services 1980). For any given species the HEP model produces a single index, the Habitat Suitability Index (HSI), which is a quantitative index of the quality of a given habitat for the species based on qualitative information regarding habitat requirements. The performance of HEP models when tested with actual fi ld data has been poor at best (Lancia et al. 1982, Bart et al. 1984), and it has been suggested that more rigorous procedures are neces ary (Bart et al. 1984). If quantitative habitat models are to be used in asses ing habitat quality, it is necessary to evaluat their reliability. It is not unreasonable to expect that quantitative models relating habitat variables to density will be of limited generality because they depend on a specific set of data taken in a restricted set of conditions. Because quantitative models are data specific, it s important to view them as empirical statements that describe correlations rather than causal relationships (Johnson 1981). The purpose of this paper is to evaluate the reliability of the predictive statements made by multiple regression models relating densities of 5 bird species to quantitative measurements of habitat and to demonstrate the sensitivity of such models to the specific data set used to generate them. Fieldwork for this research was made possible by support from the U.S. For. Serv. Rocky Mt. For. and Range Exp. Stn., the AppletonWhittell Res. Ranch Found., and the Ariz. Coop. Wildl. Res. Unit. C. and J. Bock were very helpful in making available the facilities of the Research Ranch (RR), and S. C. Martin provided information and unpublished data regarding the Santa Rita Exp. Range (SRER). R. W. Mannan, 'Present address: Department of Zoology, Brigham Young University, Provo, UT 84602.