A cost-effective, rule-based technique to improve forestry inventory on a national scale

Although South Africa´s current National Forestry Inventory (NFI) provides a good foundation for forestry planning on a national level, its accuracy and scale are inadequate for monitoring and management on a local level. Consequently, a more accurate, larger-scale NFI is urgently needed. However, updating the current NFI using traditional supervised classification techniques would be extremely costly. This article assesses the possibility of using an expert-system, rule-based remote sensing technique to map forests cost-effectively. A rule-set was created for SPOT 5 imagery in an object-orientated environment where a variety of different spectral and textural features were tested for potential use in forestry classification in two areas near Richards Bay, South Africa. Rule-set accuracies exceeded 90% for both areas and were comparable to an object-orientated supervised classification performed on the same areas. The supervised classification, however, required userintensive training area delineation for each image classified, while the rule-set classifier did not. It was concluded that the higher level of automation shown by the rule-set classifier rendered it more cost-effective for mapping forests on a national scale.