Cirrus Cloud Correction in Landsat 8 Image Using Combined Image-Based Approach and Various Classification Schemes

Abstract

The interaction of electromagnetic energy with the atmosphere causes the sensor to detect some of the elements found in the ozone layer such as ice crystals, dust, and clouds. Cirrus cloud in particular is often contaminating satellite imagery and yet relatively difficult to visually detect in visible spectrum. Indonesia as one of the tropical countries has highly cloud cover almost throughout the year. This condition causes land covers are contaminated by cirrus cloud which alters the digital numbers. The availability of cirrus band in Landsat 8 brings an advantage to eliminate cirrus clouds by performing cirrus cloud effect estimation and simple regression method. In this experiment, image-based cirrus correction was implemented in Landsat-8 over Palangkaraya city with high cirrus contamination. Cirrus cloud effect is estimated by using simple linear regression method involving samples or training area over homogeneous area cirrus contamination. Homogeneous areas were defined based on visual interpretation and statistical calculation. After estimating cirrus cloud effect on the pixel, cirrus cloud correction was performed by using arithmetic operations on images based on the slope regression coefficient which corresponded with the highest coefficient of determination. The quality of the corrected image was also statistically evaluated using reference image without cirrus contamination. Not only was the digital number evaluated but also Normalized Vegetation Index (NDVI) was compared in order to estimate the implication of cirrus correction in further image analysis.