Tropical forest monitoring by satellite remote sensing : application for African dry forests at continental scale

A global and systematic monitoring of tropical forest cover is important to assess accurately impacts that tropical deforestation might have on our global environment, in particular through carbon emissions to the atmosphere and loss of biodiversity. We developed a methodology to estimate consistently tree cover and tree cover change at global and continental levels based on a systematic sampling of medium spatial resolution (~30 m) satellite imagery. Standardized approaches were developed to process and analyse a large amount of multi-date and multi-scene imagery in an operational and consistent manner. The selection of imagery and the pre-processing chain developed provided a consistent multi-temporal dataset that constituted the basis for the automatic classification. The tree cover mapping and change detection approach comprised two automated steps of multi-date image segmentation and supervised, object-based land-cover classification based on a spectral library. Image segmentation was done at two spatial scales, introducing the concept of a minimum mapping unit via the automated selection of a site-specific scale parameter. The two automated steps of segmentation and classification provided consistent results and reduced the efforts of visual check and manual correction by experts at the end of the process. Despite the socio-economic and environmental importance of dry forests and woodlands in Africa, their extensive coverage and high exposure to change, the status of African dry forest and woodland cover and its changes over time is poorly documented. To better understand these dynamics, land cover and changes in land cover between 1990 and 2000 in dry African ecoregions were estimated from remotely sensed imagery and the developed processing chain. The study area corresponds to the Sudanian and Zambezian ecoregions, between the humid forests and the semi-arid regions. A systematic sample of 1600 Landsat satellite imagery subsets, each 20 km × 20 km in size, were analysed for the two reference years. The method generated the first consistent estimates of tree cover and tree cover change in dry Africa with known statistical precision at continental and ecoregion scales. The rate of deforestation and the carbon content in forests are the two main variables influencing estimates of carbon emissions and removals resulting from deforestation. Their uncertainties control directly the uncertainty of emission estimates. Our results of land-cover change in dry forests and woodlands of Africa have been combined with different datasets of forest carbon content using three methodological approaches. Spatial overlay of our land-cover maps with available pan-tropical carbon density maps gave the most reliable estimates of carbon emissions and uptakes. The thesis contributed to improve the consistency and comparability of tree cover area and change statistics across temporal and spatial scales and reduced the uncertainty regarding vegetation-cover dynamics and their resulting carbon emissions in African dry forests and woodlands, both essential information for science and environmental policies.