In this 2-phase study, we developed field-validated site and landscape-level predictive models for identifying potential rare and endemic plant habitat in the Great Basin of western North America. Four species were chosen to include a range of environmental variability and plant communities. Herbarium records of known occurrences were used to identify initial sample sites. The geographic coordinates, environmental attributes, and vegetation data collected at each site were used to develop 2 predictive models for each species: a field key and a probability-of-occurrence or predictor map. The field key was developed using only field data collected at the sites on environmental attributes and associated species. Predictive maps were developed with a geographic information system (GIS) containing slope, elevation, aspect, soils, and geologic data. Classification-tree (CT) software was used to generate dichotomous field keys and maps of occurrence probabilities. Predictions from both models were then field-validated during the 2nd phase of the study, and final models were developed through an iterative process, in which data collected during the field validation were incorporated into subsequent predictive models. Cross-validated models were >96% accurate and generally predicted presence with >60% accuracy. These models identified potential habitat by combining elevation, slope, aspect, rock type, and geologic process into habitat models for each species.
Aitken, M.; Roberts, D. W.; and Shultz, L. M.
"Modeling distributions of rare plants in the Great Basin, western North America,"
Western North American Naturalist: Vol. 67
, Article 4.
Available at: http://scholarsarchive.byu.edu/wnan/vol67/iss1/4