Mapping Land use/land cover using a combination of Radar Sentinel-1A and Sentinel-2A optical images

Abstract

Abstract The current study used a combination of Sentinel-1A (SE-1A) radar and Sentinel-2A (SE-2A) optical images in mapping land use/land cover (LULC) in Dak Nong province in 2018. The Random Forest (RF) algorithm was adopted to digitally categorize Landsat images into LULC maps according to ten different LULC classes included: evergreen forest, semi-evergreen forest, deciduous forest, plantation, rubber, industrial plants, crop land, residential area, water surface and others. The results indicated an overall accuracy (OA) and kappa coefficient of 81.40%, Kappa = 0.79, respectively. Based on the results of classified image, a 2018 LULC map of the study area was simulated. Accordingly, the natural forests account for 34.27% of the total area of the province, distributed scattered in districts. In which, the evergreen forest occupies the highest area with more than 166.600 ha, equivalent to 74.54% of the total natural forest area, and concentrated in the high mountain areas. Non-forest covers occupy more than 63% area of Dak Nong province. The industrial and agricultural cropland indicated a high area in the study area with a rate of 34.82% and 11.38%, respectively. This shows a strong development in the scale of industrial and agricultural crops in the study area. The objective information on current land use/land cover in this study can serve as the basis for policymakers to orient the local forest resource sustainability strategies. Besides, the study also shows that the use of a combination of Sentinel-1A radar and Sentinel-2A optical image to classify and construct the LU/LC map is a high-efficiency research direction.