Rule-based classification models: flexible integration of satellite imagery and thematic spatial data
A framework for automated land-cover classification based on a concept of a classification model was developed and tested. The framework employs a user-specified rule base to describe a classification model, defined as the series of spatial data operations and decisions used in landcover classification.’ Both evidential and hierarchical inference are supported utilizing a set of spatial data operators. The concept was tested through the development and application of a set of computer programs which support classification models. A rule base, thematic spatial data, and satellite image data were then used to define a classification model for conditions in northeastern Wisconsin. The test model incorporated Landsat Thematic Mapper data, soil texture data, and topographic position data. Classification accuracies and efficiencies using the developed system were then compared to those for supervised maximum-likelihood classifications. The classification model approach resulted in statistically significant, 15 percent improvements in classification accuracy when averaged across different analysts, geographic areas, and years.