Tropical forest monitoring by object-based change detection : towards an automated method in an operational perspective

Deforestation still nowadays occurs at an alarming rate in tropical regions. Forest monitoring is required to delineate the extents of deforested areas based on high resolution satellite images (SPOT). But classical change detection techniques have failed to detect small clearing spread over the landscape as occurring in African forests. Developed initially for temperate forests, the automated object-based change detection method using segmentation and statistical algorithm was extended to tropical regions. This approach consists in three phases: (1) multidate segmentation and object signature computation, (2) forest/non-forest classification and (3) forest change detection. First, the multidate image was partitioned into objects using segmentation and several summary statistics were derived from the within-object reflectance differences. Second, a automated forest/non-forest classification was applied on the first image to define the initial forest mask. Finally, focused on these regions, the forest change detection algorithm detected deforestation thanks to a statistical test using a multivariate iterative trimming procedure. Tested over a protected area located at the eastern border of the Democratic Republic of Congo, this method produced a deforestation map with an overall accuracy of 84 % as assessed by an independent aerial survey. Given its efficiency to detect complex forest changes and its automated character, this method is seen as adequate operational tool for tropical forest monitoring. * Corresponding author