GEONEX: A Deep Learning Approach to Prediction of Surface Spectral Reflectance

An ever-expanding volume of remote sensing data is available for research in the Earth sciences, with surface reflectance (SR) being among the most basic parameters needed for deriving higher-level land products. SR is traditionally calculated from Top Of Atmosphere (TOA) reflectance via atmospheric corrections and can be used for applications such as land use and land cover classification, burned area and vegetation indices. However, the parameters used for atmospheric corrections are difficult to measure and unstable across locations and atmospheric conditions. Additionally, NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) follows a polar orbit, capturing images of any given location intermittently. Due to the intrinsic computational complexity associated with atmospheric correction algorithms, there is a pressing need for a rapid method of SR prediction. Geostationary satellites, such as …