A Radiative Transfer Model-Driven Machine Learning Approach to Estimate the Snow Surface Albedo Over the Greenland Ice Sheet

Abstract

Accurate estimation of snow surface albedo is vital to investigate the energy budget and climate change of the Greenland ice sheet. However, the rapid changes in snow surface albedo cannot be well captured by the method based on accumulated multiangular data, especially for the ablation zone during the melt season. Here, we proposed a radiative transfer model-driven machine learning approach to estimate the snow surface albedo. First, a training dataset of snow directional reflectance and albedo was derived using the asymptotic radiative transfer (ART) model. Then, four machine learning methods were used to approximate the nonlinear relationship between directional reflectance and snow surface albedo. Finally, the daily snow surface albedo was estimated from the Moderate Resolution Imaging Spectroradiometer (MODIS) data and validated with in situ measured data. The results show that the temporal changes in snow surface albedo can be better estimated using a radiative transfer-driven machine learning approach. The extreme gradient boosting (XGBoost) method obtains the best estimation results, which are closely consistent with in situ measurements ( ${R} ^{2} =0.893$ and root-mean-square error (RMSE) =0.0545), and has improved computational efficiency. Thus, the proposed radiative transfer model-driven machine learning approach has great potential for generating long-term data records of snow surface albedo.