A STATISTICAL APPROACH FOR CLOUD-FREE MOSAIC OF LANDSAT-8 IMAGERIES (CASE STUDY: INDONESIA)

摘要

Monitoring forest and land use dynamic are crucial for better policy making processes. As one of tropical countries, parts of Indonesia are mostly covered by cloud almost all year round. It caused a challenge to acquire cloud-free images in a single acquisition data, as we need 255 Landsat-8 scenes to cover the entire country. Combining multi-temporal data acquisitions could be used to generate mosaic cloud-free satellite data. This research uses the statistical approach to twenty three available Landsat-8 data in 2017 for Indonesia region. Pre-processing steps include Top of Atmosphere (TOA) correction. Bidirectional Reflectance Distribution Function (BRDF). and cloud masking. These processing are applied to each of LIT Landsat-8 data to produce consistent reflectance data. Next, statistical minimum, maximum, and mean are extracted from any cloud-free pixel of the reflectance data for Red. Near Infrared (NIR), and Short Wave lnfrared-1 (SWIR-1) bands. The results are several mosaic layers of minimum, maximum and mean from Red. NIR, and SWIR-1 bands. Because of the sensitivity of the SWIR-1 spectral in detecting open areas, large numbers of these pixels are detected by high value of SWIR-1 time series pixels, while high NIR spectral reflectance shows mostly vegetated pixels. We observed that the RGB combination of the mosaic image can be accurately used to detect vegetated or opened areas. Most effective RGB combination for detecting vegetated pixels are: SWIR1-maximum. NIR-minimum, and Red-mean. Whereas open areas can be best detected using the RGB combination of SWIRl-minimum. NIR-maximum. and Red-mean. The comparison between both RGB combinations can be used to investigate land cover change across the years.