U.S. Department of Agriculture: Forest Service -- National Agroforestry Center


Date of this Version



Published in Landscape Ecology 19: 515–530, 2004.


Quantifying patterns is a key element of landscape analysis. One aspect of this quantification of particular importance to landscape ecologists is the classification of continuous variables to produce categorical variables such as land-cover type or elevation stratum. Although landscape ecologists are fully aware of the importance of spatial resolution in ecological investigations, the potential importance of the resolution of classifications has received little attention. Here we demonstrate the effects of using two different land-cover classifications to predict avian species richness and the occurrences of six individual species across the conterminous United States. We compared models built with a data set based on 14 coarsely resolved land-cover variables to models built with a data set based on 160 finely resolved land-cover variables. In general, comparable models built with the two data sets fit the data to similar degrees, but often produced strikingly different predictions in various parts of the country. By comparing the predictions made by pairs of models, we determined in which regions of the US predictions were most sensitive to differences in land-cover classification. In general, these sensitive areas were different for four of the individual species and for predictions of species richness, indicating that alternate classifications will have different effects in the analyses of different ecological phenomena and that these effects will likely vary geographically. Our results lead us to emphasize the importance of the resolution to which continuous variables are classified in the design of ecological studies.