A Cloud Radiative Flux Computation using Machine Learning Approach: Influence of cloud properties in modifying longwave and shortwave outgoing radiation

Atmospheric Top of the atmosphere (TOA) radiation flux is often calculated using satellite observations, reanalysis data, in the radiative transfer models, however cloud feedback, aerosol interactions, etc. may cause inaccuracies in the calculation due to the complex dynamics and nonlinear interdependencies. Machine learning (ML) offers a promising approach by capturing complex, non-linear atmospheric process relationships. This study develops XGBoost models to compute TOA IR window region and shortwave flux over Central India from cloud and geophysical data. The data used in the study is CERES Terra observations for the 2018–2023 period. The model presents over 95% of the variability in test data and considerably low MAPE and RMSE. SHAP (SHapley Additive exPlanations) analysis reveals, cloud phase and cloud optical depth as key drivers of TOA shortwave flux, while cloud top temperature and surface parameters primarily influence IR window flux. This study highlights XGBoost’s effectiveness in cloud-radiation interaction research and demonstrates the value of ML-driven SHAP analysis in understanding TOA flux variability. Results also provide an insight on how each cloud property modifies the outgoing longwave and shortwave flux in a weighed manner along with their linear and non-linear dependence.