Maximizing land cover classification accuracies produced by decision treesat continental to global scales

Classification of land cover from remotely sensed data at continental to gl obal scales requires sophisticated algorithms and feature selection techniq ues to optimize classifier performance, We examine methods to maximize clas sification accuracies using decision trees to map land cover from multitemp oral AVHRR imagery at continental and global scales, As part of our anal, s is ne test the utility of "boosting," a new technique developed to increase classification accuracy by forcing the learning (classification) algorithm to concentrate on those training observations that are most difficult to c lassify, Our results show that boosting consistently reduces misclassificat ion rates by approximate to 20-50% depending on the data set in question, a nd that most of the benefit gained by boosting is achieved after seven boos ting iterations, We also assess the utility of including phenological metri cs and geographic position as additional features to the classification alg orithm, We find that using derived phenological metrics produces little imp rovement in classification accuracy relative to using an annual time series of NDVI data, but that geographic position provides substantial pou er for predicting land cover types at continental and global scales. However, in order to avoid generating spurious classification accuracies using geograph ic position, training data must be distributed evenly; in geographic space.