This study presents reliable methods for estimating clear-sky land surface albedo (LSA) using machine learning (ML) and satellite data, aiming to improve climate models and environmental monitoring. Top-of-atmosphere (TOA) radiance data from the Ocean Colour Monitor-3 (OCM-3) sensor aboard the Earth Observing Satellite (EOS-06) satellite containing 13 spectral bands were used, supported by 2.4 million synthetic simulations generated via the 6S (Second Simulation of a Satellite Signal in the Solar Spectrum) Radiative Transfer Model (RTM). The simulations spanned diverse land covers, atmospheric states, sun and viewing geometries covering wavelengths from 0.4 to 2.5 µm. Three ML models namely Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Multiple Linear Regression (MLR) were tested. Models were trained on 70% of the simulated data and tested on 30%. Validation with actual OCM-3 data included additional aerosol and water vapor information from MODIS. LSA estimations were compared to the MODIS standard product (MCD43A3). Among the three models, RF achieved the best performance, with the lowest RMSE (0.00036) and strong agreement across various land types with MODIS data. The results confer the potential of ML models, especially RF, combined with radiative simulations, and can be used for operational estimation of LSA for OCM-3 data.
Dave et al. (Mon,) studied this question.