Semiautomatic approach for land cover classification: a remote sensing study for arid climate in southeastern Tunisia

The land use and land cover (LULC) classification has great potential to contribute to the monitoring of land degradation and climatic disasters. The purpose of this study was to assess the performance of parametric and non-parametric classification methods using remotely sensed Landsat satellite data of arid and semiarid areas, based on the computed producer’s accuracy, user’s accuracy, overall accuracy, and Cohen’s kappa coefficient. Three LULC classes were identified, and supervised classifications were applied to Landsat 8 imagery. The results show that the support vector machines (SVM) classification method produced more accurate results, using two different kernel functions, compared with the maximum likelihood classification (MLC) and the minimum distance classification (MDC). The basis radial function affords the highest overall classification accuracy of 91.20% and a mean kappa coefficient of 0.87. This classification method is very well suited to accurately map LULC in arid and semiarid regions where the main vegetation type is oasis or steppes.