Baguan-solar: AI Model for Fine-Grained Solar Irradiance Forecasting
Researchers have introduced Baguan-solar, a two-phase multimodal system that integrates predictions from Baguan, a global weather foundation model, with high-resolution imagery from geostationary satellites to generate 24-hour irradiance forecasts at a kilometer scale. This innovative two-phase approach initially predicts continuous day-night intermediates, such as cloud cover, before deducing irradiance levels. The fusion of modalities effectively retains detailed cloud structures from satellite data while incorporating broader constraints from Baguan forecasts. Tested over East Asia using CLDAS data, this method tackles the difficulties of accurate day-ahead solar irradiance forecasting, which are influenced by significant diurnal cycles and intricate cloud behavior. Existing techniques either fall short in fine-scale resolution or suffer degradation over extended lead times.
Key facts
- Baguan-solar is a two-stage multimodal framework for solar irradiance forecasting.
- It fuses forecasts from Baguan, a global weather foundation model, with high-resolution geostationary satellite imagery.
- Produces 24-hour irradiance forecasts at kilometer scale.
- First stage forecasts day-night continuous intermediates like cloud cover.
- Second stage infers irradiance from those intermediates.
- Modality fusion preserves fine-scale cloud structures from satellite and large-scale constraints from Baguan.
- Evaluated over East Asia using CLDAS data.
- Addresses limitations of numerical weather prediction and satellite extrapolation.
Entities
Locations
- East Asia