Regional scale land use/land cover classification using temporal series of MODIS data
Résumé
This paper describes a methodology for systematic land use/land cover classification on a regional scale, with emphasis on a low cost and highly automatized approach. This methodology is based on multitemporal analyses of surface reflectance data from the Moderate Resolution Imaging Spectroradiometer (MODIS), which is located on board NASA's Terra and Aqua satellites and features high temporal frequency, extensive coverage, and extremely low costs for data acquisition. A sequence of automatized procedures were developed for MODIS data pre-processing, as well as for the training and execution of a supervised classification algorithm, where temporal profiles are fitted to smooth polynomial curves and intelligent curve features are then computed in order to reduce data dimensionality and improve profile interpretability, thus providing a more robust classification approach. A case study was performed in the High Taquari Basin, in the states of Mato Grosso do Sul and Mato Grosso, Brazil, which showed that the method was indeed capable of generalizing well over the entire region of study (over 25,000km2), effectively discriminating between areas of agriculture, pasture, and savannah. The methodology was also seen to be quite successful in identifying areas of deforestation, which is of particular interest for the monitoring of land use and land use change in the region.