CSRFormer: An Instantaneous Global-ocean Clear-sky Radiative Flux Dataset Derived From CERES

CSRFormer_v1.0 provides instantaneous clear-sky shortwave and longwave radiative flux estimates over the global ocean at the observation time of CERES Single Scanner Footprint (SSF) overpasses. The dataset was developed to support more accurate estimates of Cloud Radiative Effect (CRE), which require clear-sky fluxes that are temporally consistent with the corresponding all-sky observations. Because passive satellite sensors cannot directly observe the clear-sky state beneath clouds, clear-sky fluxes in cloudy regions are often approximated using radiative transfer simulations or temporally interpolated products, both of which can introduce substantial uncertainty and temporal mismatch.

To address this issue, the dataset is generated with CSRFormer, a physically constrained deep-learning framework trained against CERES SSF clear-sky observations. The model uses ERA5 atmospheric profiles, MERRA-2 aerosol properties, and NOAA sea surface temperature as predictor variables to estimate instantaneous clear-sky shortwave and longwave fluxes over the ocean. Evaluation against independent clear-sky samples shows strong agreement with CERES SSF observations, with R² values of 0.81 for shortwave flux and 0.99 for longwave flux, and corresponding RMSE values of approximately 3.2 W m⁻² and 2.0 W m⁻².

A key advantage of this dataset is that the predicted clear-sky fluxes are matched to the exact CERES overpass time, avoiding the temporal inconsistency inherent in hourly mean products. When applied to cloudy scenes, CSRFormer provides an observationally anchored estimate of the theoretical clear-sky reference state beneath clouds. This makes the dataset particularly useful for CRE analysis over the global ocean, as well as for studies of how cloud radiative effects covary with cloud properties at satellite overpass time.