Downscaling of Land Surface Albedo Method Based on Stratified Linear Regression

Abstract

Due to limitation of observation angles and resolution of satellite sensor, the resolution of surface albedo products retrieved based on angle information is usually coarse, such as MODIS albedo products and GLASS albedo products. Downscaling method of stratified linear regression proposed in this study solved this problem. First, under the assumption of non-anisotropy surface, Landsat8 primary albedo is obtained by converting narrow albedo to broadband albedo. Under the resolution of 500m, the correlation degree of Landsat8 primary shortwave albedo and MCD43A3 shortwave albedo shows higher after classification. Therefore, a linear regression model for each land cover is established. By fusing Landsat8 data with MCD43A3 albedo, downscaled shortwave albedo with high-resolution is obtained. Finally, it is validated with data of four observation sites in the United States. The results show that downscaled albedo has high precision (bias is 0.01 and sd is 0.012) and rich details, and the algorithm is reliable for different land cover, indicating its potential to become an operational algorithm for high-resolution albedo product.