A revised RTM-based algorithm was developed coupled with the classification of land cover types. The study area was near Cedar Island National Wildlife Reserve, North Carolina, USA. A cloud-covered Sentinel-2A image of 1000 (rows) × 1000 (columns) acquired on 2 February 2017 was studied. The area was classified into three land cover types of vegetation, bare soil, and water. The assumption that the linear relationship of the TOA reflectance of any two visible bands was independent of the terrain types in the RTM-based algorithm was not valid. Thus, the impact of the land cover types on the algorithm was analyzed. The cloud-removal algorithm should be separately developed based on each land cover type removing the clouds in each land cover type accordingly. This assessment was evidenced by the spatial correlation coefficients of the cloud-covered image and the cloud-free reference image, of the cloud-removed image using the revised RTM-based algorithm and the reference image, and of the cloud-removed image using the original RTM-based algorithm and the reference image. The revised algorithm was not only effective in the removal of thin clouds land cover type-by-cover type but also outperformed the original RTM-based algorithm.