Transfer learning sharpens solar radiation estimates from China's Fengyun-4A satellite

A new study published in the Journal of Remote Sensing demonstrates how transfer learning can enhance solar radiation mapping from China's Fengyun-4A (FY-4A) geostationary satellite. The research addresses the challenge of accurately estimating surface solar radiation (SSR) and its components—global, direct, and diffuse—which are critical for solar power forecasting, climate research, and land-surface modeling. Ground-based radiometric networks provide reliable observations but are sparse, while reanalysis products may suffer from coarse resolution and simplified cloud-aerosol interactions. Satellite observations can fill these gaps, but many existing algorithms are sensor-specific and focus mainly on global radiation.

The research team, composed of scientists from the Aerospace Information Research Institute, Chinese Academy of Sciences; Sichuan University of Science and Engineering; and the Institute of Atmospheric Physics, Chinese Academy of Sciences, developed a deep neural network (DNN) model that uses a transfer learning strategy to adapt knowledge from Japan's Himawari-8 satellite. The model was pretrained on Himawari-8 Level 1 (L1) observations and the Cloud, Atmospheric Radiation and Renewal Energy Application (CARE) radiation product, then fine-tuned with FY-4A L1 data. This approach reduces reliance on auxiliary meteorological datasets, making the method more practical for near-real-time monitoring.

Validation using 33 ground stations from the Baseline Surface Radiation Network (BSRN), Bureau of Meteorology (BOM), and Global Tropical Moored Buoy Array (GTMBA) during 2018–2020 showed strong performance. At representative BSRN sites, FY-4A achieved instantaneous root mean square errors (RMSEs) of 102.2, 117.5, and 83.1 W m⁻² for global, direct, and diffuse radiation, respectively. At the daily mean scale, RMSEs dropped to 28.5, 30.1, and 22.6 W m⁻².

The ability to separately estimate direct and diffuse radiation is particularly important for solar energy applications. Direct radiation is crucial for concentrating solar power, while diffuse radiation affects photovoltaic output under cloudy or hazy conditions. By resolving these components, the new FY-4A radiation product could improve PV site assessment, power forecasting, and grid management. It also supports climate modeling and land-surface simulations.

The study highlights that transfer learning can overcome sensor differences and limited ground training data. This framework could be extended to other Chinese geostationary satellites, such as Fengyun-4B (FY-4B), to enhance solar-energy monitoring across East Asia and beyond. The findings provide a stronger data foundation for sustainable energy planning and the clean-energy transition.

Detailed information is available in the original study published in the Journal of Remote Sensing (DOI: 10.34133/remotesensing.1044).