Abstract
This paper evaluates the generalization potential of a classification approach for the task of mapping tropical secondary forest across the highlands of mainland Southeast Asia. The approach applies linear mixture modelling to atmospherically and topographically corrected Advanced Space‐borne Thermal Emission and Reflection Radiometer data, and the resulting fractional images of green vegetation, background and shade are classified into four major land covers using a decision tree classifier. The results indicate a potential for developing a generic linear mixture model. However, the decision rules by which end‐member fractions are classified into land cover classes are site‐specific. Therefore a regional applicable and fully automated mapping approach is not realistic. This study concludes, however, that an approach which couples linear mixture modelling and decision tree classification is both accurate and robust for mapping tropical secondary forests across the region.
Acknowledgements
The authors wish to thank the University Support to Environmental Planning and Management Project (USEPAM) for facilitating the fieldwork in Vietnam and Laos. CARE Thailand, the Science, Technology and Environment office in Luang Prabang and the board and staff at Tam Dao National Park are also thanked for their support during the fieldwork. Moreover, the authors appreciated the comments and suggestions from the two anonymous referees which helped to improve the manuscript. The IKONOS data were acquired through the NASA Scientific Data Purchase Project by the Tropical Rain Forest Information Center (TRFIC), a member of NASA's Federation of Earth Science Information Partners (ESIP). These data were provided in partnership with Dr Skole and Michigan State University.