Surface Broadband Radiation Data from a Bipolar Perspective: Assessing Climate Change Through Machine Learning

Clouds influence net radiative flux by affecting both shortwave (SW) and longwave (LW) radiation, but their impact in polar regions is uncertain due to limited ground data and satellite challenges. This study examines sky conditions from 2010–2020 at six polar stations (two Arctic, four Antarctic) using BSRN radiation measurements. Cloud fractions were estimated via the RADFLUX method from SW and LW fluxes, classifying skies as clear, cloudy, or overcast. These labels trained machine learning models—Random Forest, KNN, and XGBoost—with features like LW radiation and temperature. XGBoost performed best, achieving balanced accuracy of 0.78. Results suggest clustering similar stations and feature normalization improve model generalization across sites.