Extraction of remote sensing-based forest management units in tropical forests

As a spatial source of forest data, the forest stand is the basic unit for forest management and planning. Computerized segmentation of a diverse range of remote sensing material has been studied for delineating stands in various forest types, but is relatively rare under tropical conditions. In line with REDD+, which advocates sustainable forest management in tropical developing countries, we report here on the delineation of forest stands in Laos using data from Airborne Laser Scanning, Airborne CIR and ALOS AVNIR-2. Rather than using the spectral layers alone, the segmentation was applied to wall-to-wall maps of predicted values from empirical models that retrieve forest stem volume and basal area. The homogeneity of 96 sets of segmentation categorized according to the hierarchical mean sizes of the segments was evaluated quantitatively in terms of the AICvar index and qualitatively by eye. The results show that the quality and performance of the empirical models are duly reflected in the quality and performance of the segmentation, and that the most competitive delineations are provided by models that involve ALS. Finally, the future prospects for using this empirical model-based segmentation approach to detect and quantify deforestation, or even forest degradation, and thereby to provide further support for the development of REDD methodology, are discussed.