Land cover maps of Antananarivo (capital of Madagascar) produced by processing multisource satellite imagery and geospatial reference data

Abstract

We describe a reference spatial database and four land use maps of Antananarivo city produced over 2017 reference year using a methodology combining machine learning and object based image analysis (OBIA). These maps are produced by processing satellite images using the Moringa land cover processing chain developed in our laboratory. We use a single very high spatial resolution (VHSR) Pleiades image, a time series of Sentinel-2 and Landsat-8 images, a Digital Terrain Model (DTM) and the aforementioned reference database. According to the Moringa workflow, the Pleiades image is used to generate a suitable object layer at VHSR using a segmentation algorithm. Each object is then classified using variables from the time series and information from the DTM. The reference database used to train the supervised classification algorithm is here described, as well as the 4 land cover maps produced at four different hierarchically nested nomenclature levels. For a number of classes going from 2 to 20, overall accuracies range from 94% to 74%. Such land cover products are very rare in Madagascar, so we have decided to make them openly accessible and usable by land managers and researchers.