| dc.description.abstract | Large-scale land acquisitions (LSLAs) for agricultural sector have grown significantly in the past decade, and are mostly prevalent in developing countries. Because LSLAs are not without negative effects on the environment and local communities, and because information about them is scarce and difficult to obtain, systems allowing LSLAs detection, characterization and monitoring in space and time are needed. With the increasing availability of global satellite data products, technological development in cloud computing, image and data mining analysis, remote sensing has evolved to an interesting tool for the detection and characterization of changes in land use systems.
This study presents a novel approach to generically detect and characterize LSLAs at regional spatial extents. In order to capture and analyze the full range of land use spectral and spatial signatures related to agricultural LSLAs, this study is based on a 2-level data driven approach (Self-Organizing Maps followed by a clustering algorithm), consisting of two phases: 1) land use/land cover change detection at regional scale within dense temporal stacks of vegetation indices (MODIS-NDVI, 250m) and 2), discrimination of different land use/land cover classes using a set of spectral vegetation indices, textural features and shape metrics computed from landscape-extracted objects (Landsat-8, 30m). Evaluation of the methodology is performed against a ground truth database on LSLAs in Senegal.
Results obtained during this exploratory research, are promising and provide some insights in agricultural LSLAs in the northern half of Senegal. With a very limited number of discriminative features (consisting of two Vegetation Indices and two textural features), detection of agricultural LSLAs is possible. Recommendations are given for enhancement of the generalization performance of the unsupervised classifier. | |
