Developing a Cloud Removal Tool by Combining Optical and Microwave ALOS Data

Abstract

In optical remote sensing, clouds have always been a major problem because they obstruct the details underneath them owing to the simple reason that the visible band of electromagnetic spectrum cannot penetrate clouds. This study presents a novel approach to address the problem of clouds and their shadows in optical images by combining optical and microwave data of Advanced Land Observation Satellite (ALOS) by taking advantage of the ability of microwaves to penetrate clouds, fog, haze and mist. The study focuses on development of an algorithm that is able to detect clouds, cloud shadow, mask both cloud and cloud shadow contaminated pixels and by use of interpolation technique between the optical and the microwave image, re-establish pixel values for the masked pixels. The detection of thick clouds is done by computing Total Reflectance Radiance Index (TRRI) using the four Bands of ALOS optical image (AVNIR-2 Image) and defining thresholds that accurately define thick clouds. Thin Clouds are detected using Cloud Soil Index (CSI) and also defining thresholds that define the thin clouds. Cloud shadows are detected by projecting the detected clouds by the average distance and bearing measured from a sample of clouds to their corresponding cloud shadows. In all cloud and cloud shadow contaminated pixels, their corresponding pixels are identified in the microwave image and using nearest search, other pixels with similar or nearest pixel values are searched for and once identified, their corresponding pixels in the optical image are identified and their pixel values are used to fill the masked pixels. The results show that the developed algorithm for detection of thick clouds using the defined TRRI thresholds worked very well and all the thick clouds were detected and masked out while the developed algorithm for detecting thin clouds using defined thresholds of CSI worked very welltoo with most of the thin clouds being detected and masked out. However, there were some remnants of thin cloud patches which still went on undetected and further research needs to be done for more precise methods of detecting thin clouds. The developed program function for detection and masking of cloud shadow worked very well with all the cloud shadow pixels being detected and masked out. The developed program for interpolation to fill the missing pixel values in the already masked cloud and cloud shadow pixels worked well with the missing information being revealed. Therefore, the algorithm for cloud and cloud shadow removal was successfully developed with all the subcomponents working well. However further research needs to be done for improved methods of thin cloud detection and future satellite developments needs to consider incorporation of both optical and microwave sensor in the same platform.