Using supervised learning to develop BaRAD, a 40-year monthly bias-adjusted global gridded radiation dataset.

Area of Science:

• Earth and Atmospheric Sciences
• Climate Science
• Radiative Transfer

Background:

• Diffuse solar radiation is a critical but under-archived component of Earth's radiation budget.
• Global climate models often lack diffuse radiation data, and ground-based observations are scarce.
• Accurate radiation data is vital for climate modeling and understanding biosphere interactions.

Purpose of the Study:

• To develop a comprehensive, global, 40-year (1980-2019) monthly database of solar radiation, including diffuse and direct components.
• To address the under-representation of diffuse solar radiation in climate models and research.
• To provide a valuable resource for studies on atmosphere-biosphere interactions, such as the diffuse radiation fertilization effect.

Main Methods:

• Development of the Bias-adjusted RADiation (BaRAD) dataset.
• Utilized a random forest algorithm trained on Global Energy Balance Archive (GEBA) observations.
• Applied the algorithm to the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) dataset at its native resolution (0.5° x 0.625°).

Main Results:

• The BaRAD dataset provides global monthly data for total, diffuse, and direct shortwave radiation from 1980-2019.
• The dataset preserves seasonal, latitudinal, and long-term trends present in MERRA-2 data.
• BaRAD demonstrates significantly reduced biases compared to MERRA-2, with near-zero mean bias error for diffuse radiation (RMSE = 10.1 W m⁻²) and -0.2 W m⁻² for total shortwave radiation (RMSE = 19.2 W m⁻²).

Conclusions:

• The BaRAD dataset offers a reliable, bias-adjusted record of solar radiation components.
• This dataset fills a critical gap in climate data, particularly for diffuse radiation.
• BaRAD is expected to enhance research in climate science and atmosphere-biosphere interactions, including the diffuse radiation fertilization effect.