A Multi-factor Weighting Method for Improved Clear View Compositing using All Available Landsat 8 and Sentinel-2 Images in Google Earth Engine

Abstract

The increasing availability of freely accessible remote sensing data has been crucial for improved global monitoring studies. Multisource image combination is a common approach for overcoming a major limitation associated with single-sensor data sources, which cannot provide adequate observations to fill data gaps arising from cloud contamination, shadows, and other atmospheric effects. In particular, image compositing is often used to generate clear view images over a large area. For example, the best available pixel (BAP) method has been proposed to construct clear view and spatially contiguous composites based on pixel-level quality rules. For any location with a bad observation, this method searches observations acquired in other dates and uses the one with the highest score to replace the contaminated observation. This, however, can lead to artificially large discontinuities along the edge of a filled area, which is typically caused by large phenological differences among the observations considered. To mitigate this issue, we developed a multifactor weighting (MFW) method for constructing clear view composites with a higher level of spatial continuity and radiometric consistency than those produced using the BAP method. Assessments over 4 study sites selected from different climate zones in China demonstrated that the composites produced using the MFW method were more consistent with reference images than those generated using the BAP method. Spectral agreements between MFW composites and the reference ( R = 0.78 to 0.95) were generally higher than the agreements between BAP composites and the reference ( R = 0.65 to 0.93). These results demonstrated that the proposed MFW method can provide a promising strategy for constructing clear view, seamless, and radiometrically consistent image composites for large-scale applications.