All-Sky Image Benchmark Dataset for Solar Forecasting at the Plataforma Solar de Almería in 2019

This dataset provides data for short-term solar irradiance forecasting using ground-based all-sky imagery during 2019, acquired at the Plataforma Solar de Almería (PSA), Spain, owned and operated by CIEMAT. The data comprises three months (September to November 2019) from the METAS station at PSA, inlcuding one-minute averaged irradiance measurements (GHI, DNI, DHI) and 30 second resolution RGB all-sky imagery. In addition it provides ambient temperature and air pressure from a neighboring meteorological station. From this period, 28 days have been manually selected to form a test set of a benchmark study for evaluating all-sky-imager (ASI)-based intra-hour solar forecasting systems in 2019. The image data are organized in daily TAR archives containing JPEG images and per-image JSON metadata. An accompanying Parquet index enables efficient access to image metadata without extracting the archives. Meteorological measurements are provided as a CSV file using UTC+01 timestamps. In addition, the dataset includes the official benchmark test split in a separate CSV file. Additional dataset metadata are provided in a YAML file, including the camera-specific intrinsic and extrinsic parameters obtained through geometric calibration (see also 10.1016/j.solener.2025.113476). The dataset is intended to support the development, benchmarking, and reproducible evaluation of machine learning and computer vision methods for all-sky image analysis and intra-hour solar irradiance forecasting.