Validation and Analysis of HRRR Insolation Forecasts using Surfrad

The challenges of short-term solar forecasting require a comprehensive use of observations, numerical models, and hybrid algorithms to optimize skill at different forecast windows. As forecast models continue to advance in spatial resolution and physical sophistication, assessments of their performance are required to understand forecast product evolution across space, time, and meteorological regime. Toward this goal, we present the results of hourly surface insolation forecasts from the NOAA High-Resolution Rapid Refresh (HRRR) model validated against observations from the NOAA-ESRL SURface RADiation (SURFRAD) network for the period April 2011 through the present. This validation seeks to assess the predictive skill of the HRRR model in the 1-6 hour forecast range for applications in solar power planning and management. Issues affecting the HRRR forecast are examined, particularly with regards to when radiative transfer routines in the HRRR model are called. In addition to the validation work, a two-dimensional histogram of model error as a function of forecast time is generated from the model forecasts and surface observations. Differences between the daily model and observation datasets are normalized by a clear-sky model, then mapped onto a histogram of model output valid time versus model forecast time (from 1-6 hours.) The analysis reveals patterns in model error for the daily observation as a function of forecast valid time and lead time -providing useful insight to both developers and users of the HRRR model. Based on these data, a k-means based regime analysis can be performed to determine trends in model forecast error for insolation over the analysis period.