A Data-Driven Model for Estimating Clear-Sky Surface Longwave Downward Radiation Over Polar Regions

Polar regions play a crucial role in global climate change. Surface longwave downward radiation (SLDR) is a primary energy source for the polar surface and plays an essential role in studies of polar hydrology, temperature, and climate. Therefore, accurately estimating the SLDR over polar regions is highly important. However, the accuracies of existing polar SLDR datasets and SLDR inversion methods are insufficient to meet the requirements of relevant research. In this study, we developed a data-driven model for high spatial resolution clear-sky SLDRs estimated from Moderate Resolution Imaging Spectroradiometer (MODIS) imagery in polar regions. The model comprises two layers: the first layer incorporates three machine learning models, namely, eXtreme gradient boosting (XGBoost), convolutional neural network (CNN), and transformer, while the second layer consists of a stacking meta-model. The ground measurements collected from 51 sites were used to train and validate the developed model. The bias, RMSE, and R-2 of the model training are zero, 14.15 W/m(2), and 0.95, respectively, whereas the values for the validation are 0.49, 15.35 W/m(2), and 0.9, respectively. We also compared the accuracies of the ERA5 and CERES-SYN SLDR data with the SLDR estimated by the developed model. The results indicate that the developed model is superior to the ERA5 and CERES-SYN SLDR models when evaluated at the validation sites. In addition, we analyzed the performance of the developed model under different elevations and seasons, demonstrating its robustness in different situations.