RADIOMETRIC COMPARISON OF TWO CONSECUTIVE LANDSAT 8 AND LANDSAT 9 SATELLITE SCENES FROM MATOPIBA FOR LAND USE AND COVER MAPPING
Land use and land cover. Remote sensing. Random Forest. GLCM.
Since 1972, successive series of Landsat satellites have shown improvements in their imaging acquisition modes. The objective of this study was to compare the potential of the Operational Land Imager (OLI) sensor on board the Landsat 8 (L8) and Landsat 9 (L9) satellites to map the following land use and land cover (LULC) classes in the MATOPIBA region: Gallery Forest; Savanna shrubland; Wooded savanna; Cultivated pastures; Green crop residues; Dry straw; and Bare soil. Two consecutive scenes from the dry season in the south of the Piauí State, obtained by the L8 and L9 satellites eight days apart, were classified by the supervised image classification method called Random Forest. The input attributes were six multispectral bands converted to surface reflectance; four spectral indices; and five texture attributes derived by the Gray Level Co-Occurrence Matrix (GLCM) technique. The overall classification accuracies were similar: 85.6% and 85.0% for L8 and L9, respectively. Despite this similarity, there was a change in the classification results in 7.1% of the total scene area. The main discrepancies occurred in areas where there were burn scars, areas of annual crops with some land cover change during the eight-day interval, and in areas with a predominance of LULC classes with high spectral similarity. The results obtained suggest that both satellites have similar potential for mapping the main LULC classes in MATOPIBA.