Evolution Estimated by Remote Sensing of Forest Boundaries in Two Parks in the Rainforest of Madagascar Labeled UNESCO

In Madagascar, increasing poverty is driving the population along the protected areas into continue clearing. The annual deforestation rate is estimated at 1% in Malagasy protected areas, and at 4% in forest outside protection. The two parks in rain forest, Ranomafana (400 to 1 200 m) and Andringitra (600 to 2 600 m), which have been designated UNESCO in 2007, are not immune to this phenomenon.

The data allowed the listing of World Heritage in the Eastern forests have clearly mentioned the lack of monitoring of violations in the parks. To consolidate a state 0 at the label, details have to be gathered on the magnitude of the pressure before and after. The detection of land use change in this interval will help scheduling a future periodic evaluation every 5 years.

Since the under protected area status of the two parks Ranomafana (1991) and Andringitra (1927 as a nature reserve and 2001 as a national park), the pressure was reduced and controlled. However, in remote areas controls are virtually nonexistent and deforestation continues. Remote sensing evaluation may take advantage of SPOT improved spatial resolution images from 20 to 2.5 m and performance of processing tools (OTB from CNES). Land use changes were evaluated in 20 years between two dates, 1988-2009 in Ranomafana and 1986-2008 in Andringitra. The SPOT images (CNES-ISIS) were treated by ENVI 4.5 and ArcGIS 9.2; supervised classifications use the maximum likelihood algorithm; the detection of changes between two dates, is made with ENVI and OTB.

For each park, two scenes, north and south, have been assembled. In Ranomafana the difference of shooting dates in 1988 is solved by radiometric corrections. The delineation of a buffer zone of ten kilometers around the park boundaries includes their peripheral areas. Clouds, cloud shadows and shadows of the slopes are the difficulties encountered. Several fields transects were carried out for verification and validation at the same time than laboratory work.

For the two parks, 11 classes were identified among which 4 are common: water, bare soil, savanna, crop-fallow mosaic. Differences remain for the distinctions forest-afforestation, wetland-rice field, and levels of forest degradation. The detection of changes between two dates encounters two problems: the difference in resolution between the two dates and the confusion between degraded forest and fallows. The clearing rate in Andringitra is estimated at 10% in 22 years. The assumption is that better access leads to a higher rate in Ranomafana than in Andringitra.

The methodology followed could be used in humid eastern forests integrating corridors that connect parks between them and correcting distortions due to relief. These lessons would be recovered in the labeling procedure of dry forests of Madagascar.