MAPPING AND RECOVERING CLOUD-CONTAMINATED AREA IN MULTISPECTRAL SATELLITE IMAGERY WITH VISIBLE AND NEAR-INFRARED BANDS

Cloud cover severely influences the accuracy of land use/cover mapping and biomass estimation with optical satellite imagery. This study integrated automated threshold selection algorithm (ATSA) and region growing to delineate unrecoverable thick cloud. Concerning hazy areas, Fourier analysis was used to generate haze filter to reduce haze interference and recover ground information. The result of thick cloud delineation shows the overall accuracy and kappa statistics are 94.75% and 0.883 separately. For the haze-off result, haze filter improves land cover classification and increases the overall accuracy and kappa statistics by about 4%. With NDVI results, the root-mean-square (RMS) between hazy and clear image is 0.21 while RMS between haze-off and clear image is 0.15. This study demonstrated that cloud processing only using Green, Red, NIR bands without cloud-free reference areas or imagery is sufficient for thick cloud delineation and can achieve some improvement in haze removal.