Predicting potential natural vegetation in an interior northwest landscape using classification tree modeling and a GIS

Integration of a GIS with statistical predictive models facilitates mapping the likely spatial distribution of plant associations and modification of maps as new data or vegetation-environment relationships are discovered. In this study, data for classified plant communities were used to develop a georeferenced database representing 39 plant associations and environmental variables at 1,249 plot locations. This database was used to develop models predicting the occurrence of plant associations. These predictive models were implemented in a GIS to render maps of predictable plant associations, plant association groups, and overstory series. Overall model accuracy ranged from 30% for the model predicting plant association to 63% for the model predicting series. However, several associations, groups, and even series could not be predicted, and model performance for those that were predictable often differed from overall model accuracy. Association-level accuracy of model predictions ranged from 18 to 84% while series-level accuracy ranged from 41 to 85%. Model selection for management applications should be based on specific management objectives. Expansion of the regional sample of reference plots and database augmentations, including documentation of disturbance histories, should provide useful enhancements for future modeling efforts.