Cirrus Cloud Correction in Landsat 8 Image Using the Image-Based Approach: A Case Study in Sumba Island, Indonesia

摘要

Satellite observations show that cirrus cloud can cover 30% of Earth's surface with 70% of its global coverage in the tropics. Indonesia as a tropical country has cirrus coverage all year round. Cirrus clouds are typically thin which form at high altitude making it difficult to detect on optical imageries. However, with the availability of cirrus band in Landsat 8, cirrus cloud effects can be estimated and corrected. This paper demonstrates the image-based approach by utilising cirrus band to estimate and correct cirrus clouds which cover Landsat 8 image of Indonesia. We use Landsat 8 image of Sumba in this study to perform cirrus correction. Cirrus effect on each pixel in visible bands is estimated using linear regression model involving contaminated pixel samples over homogenous area with variety of cirrus intensity. Cirrus cloud correction is performed using band math based on the slope regression coefficient which corresponds to the highest determination coefficient. The cirrus corrected image is also statistically validated using cirrus-free image as reference. This study shows satisfactory result of cirrus corrected image. The quality of cirrus corrected images of blue, green and red band statistically improves with the increase of determination coefficient by 2.17%, 10.36%, 7.44% respectively.