SolarBench

SolarBench aims to reduce barriers for researchers and practitioners to access quality-controlled, standardized environmental data and machine learning tools to advance solar energy nowcasting for large-scale solar integration.

This repo contains multiple data modalities and is structured as follow:

• ground_full: Ground-based sensor data include sky images, a variety of meteorological measurements, and labels for solar forecasting tasks, namely different solar irradiance components (i.e., GHI, DHI, and DNI) or power generation from solar panels.
• ground_no_skycam: A lite version of ground_full by excluding sky images. Useful for fast analysis and training non-vision models or performing ablation studies.
• satellite: Geostationary satellite imagery including GOES, Himawari, MSG.
• weather_forecasts: Numerical weather predictions (e.g., GFS).
• climate_reanalysis: Retrospective climate reanalysis datasets (e.g., ERA5).

These data can be easily accessed and made into use through the machine learning toolbox we developed. For more details, check it out here: