Evaluating the Accuracy of Single-Camera Irradiance Forecasting

We present results from a newly established all-sky imager (ASI) testbed at DTU Risø Campus. Three calibrated ASIs capture hemispheric sky images every 30 seconds and provide input for cloud detection and irradiance mapping. Each camera was calibrated using the Scaramuzza fisheye model, yielding lookup matrices that map pixels to azimuth and zenith angles. Based on these, equidistant projections were created, and cloud transmittance maps were derived using a normalized color index. Ground-level irradiance was then estimated by projecting cloud transmittance onto the terrain using solar geometry. The estimated Global Horizontal Irradiance (GHI) was compared to pyranometer measurements at a meteorological station 1̃ km from one of the cameras. Results show good agreement in daily GHI dynamics and highlight challenges in accurately reproducing irradiance ramps near solar noon and under broken cloud conditions. This testbed provides a basis for systematic evaluation of ASI-based irradiance forecasting methods. Data collected in this testbed will be made available to the community to facilitate further research and development of ASI-based solar forecasting methods, and we aim to release a larger dataset in the future.