Hourly 5-km surface total and diffuse solar radiation in China, 2007-2018

Area of Science:

• Earth and Environmental Sciences
• Atmospheric Science
• Remote Sensing

Background:

• Surface solar radiation is crucial for numerical modeling and ecosystem carbon uptake.
• Accurate, high-resolution solar radiation data is needed for climate and ecological research.

Purpose of the Study:

• To generate a 12-year (2007-2018) hourly dataset of surface total solar radiation (R_s) and diffuse radiation (R_dif) at 5-km resolution.
• To utilize deep learning techniques for improved spatial and temporal accuracy of solar radiation estimation.

Main Methods:

• A deep neural network combining convolutional neural network and multi-layer perceptron was employed.
• The network integrated spatial patterns and simulated complex radiation transfer using MTSAT satellite data.
• The generated dataset was validated against ground measurements.

Main Results:

• High validation metrics were achieved: hourly R_s (R²=0.94, MBE=2.48 W/m², RMSE=89.75 W/m²) and R_dif (R²=0.85, MBE=8.63 W/m², RMSE=66.14 W/m²).
• Correlation coefficients improved at daily and monthly scales for both R_s and R_dif.
• Spatially continuous hourly maps accurately reflected regional differences and diurnal cycles.

Conclusions:

• The generated dataset provides a valuable resource for regional climate change studies, terrestrial ecosystem simulations, and photovoltaic applications.
• Deep learning techniques effectively enhance the accuracy and resolution of satellite-derived solar radiation data.