Overview Of A Decade Of Yearly Land Cover Classifications Derived From Multi-Temporal Optical Satellite Images

This article presents a monitoring of land cover/use by satellite images over an 11-year period (2006-2016), over a study site located in southwestern France near Toulouse. Time series of optical data are acquired by Spot and Landsat, which deliver images in multispectral mode with high spatial resolution (10-30 m). The detection of the different types of land cover/use (crops, grasslands, water, urban and wood) is produced every year. It is based on national reference geographical data and a random forest algorithm. The classifications are characterized by a high level of performance, with an average kappa of 0.83 (OA=0.85). The performance by land cover/use type is related to their representativeness, dates and number of acquisitions, and the resolution of satellite images. The results allow analyzing the evolution of the three main crops (wheat, sunflower and corn).