Land use / land cover map of Vavatenina region (Madagascar) produced by object-based analysis of very high spatial resolution satellite images and geospatial reference data.

International audience

إنجليزي. Hereby presented data consists of a land use / land cover map of an 84 km2 part of the Vavatenina district, in the Analanjirofo Region of the east coast of Madagascar, where the landscape is dominated by woody vegetation. This map was obtained by processing very high spatial resolution multispectral images acquired by the Pleiades satellite sensor in 2018. Pleiades data were produced by Airbus defence & Space (France), and distributed by the French National Space Agency (CNES, France). An object-based approach was chosen to exploit the advantages of such images, allowing the use of various texture indices derived from the image to discriminate between different surfaces sharing similar radiometric properties while being structurally different. The object-based image analysis consists of iterative segmentations of the image into homogeneous regions, called objects, which can afterwards be classified using radiometric and textural variables. The reference database, computed out of field knowledge and used to train the mapping methodology, is also described here. It includes 2105 georeferenced points labelled with one of the 9 following classes: Built up/road/bare areas, Annual crops/pasture/short vegetation, Clove dominated park, Clove monoculture, Diversified agroforest, Diversified park, Plantation of woody species, Shrubby fallow, Woody fallow. It is formatted as a geographical information system, accessible for any other purpose like further methodology development. These ground-truth data also helped to the definition of the 13 thematic land use / land cover types to be distinguished in the area. Multiple segmentation levels were thus necessary to gradually derive these classes, with different classification algorithms depending on the level. The final map provides an exhaustive description of the area with a good overall accuracy, reaching 80%. It is openly available as an ESRI shapefile to allow quantitative spatial analysis in key applications such as agriculture, environment, or food security.