Using prior probabilities in decision-tree classification of remotely sensed data

Land cover and vegetation classification systems are generally designed for ecological or land use applications that are independent of remote sensing considerations. As a result, the classes of interest are often poorly separable in the feature space provided by remotely sensed data. In many cases, ancillary data sources can provide useful information to help distinguish between inseparable classes. However, methods for including ancillary data sources, such as the use of prior probabilities in maximum likelihood classification, are often problematic in practice. This paper presents a method for incorporating prior probabilities in remote-sensing-based land cover classification using a supervised decision-tree classification algorithm. The method allows robust probabilities of class membership to be estimated from nonparametric supervised classification algorithms using a technique known as boosting. By using this approach in association with Bayes’ rule, poorly separable classes can be distinguished based on ancillary information. The method does not penalize rare classes and can incorporate incomplete or imperfect information using a confidence parameter that weights the influence of ancillary information relative to its quality. Assessments of the methodology using both Landsat TM and AVHRR data show that it successfully improves land cover classification results. The method is shown to be especially useful for improving discrimination between agriculture and natural vegetation in coarse-resolution land cover maps. D 2002 Elsevier Science Inc. All rights reserved.