Enhanced Statistical Post-Processing of Solar Irradiance Predictions Using Optimized WRF Forecasts of Cloud Cover Categories

It is increasingly important for a range of applications to accurately predict solar irradiance at the surface from weather forecasts. Often a combination of numerical weather prediction (NWP) modeling and statistical post-processing is employed for accurate predictions. Towards this end, we find that a reliable forecast of cloud cover categories (e.g. clear-sky, scattered-clouds, broken-clouds, overcast) by NWP models is a key to improving solar irradiances prediction accuracy via machine-learning based statistical post-processing. To improve the ability of NWP models to forecast cloud cover categories, we extensively search for optimal configurations of the Weather Research and Forecasting (WRF) model. We use a variety of WRF model configurations, including different microphysics packages, radiation transfer schemes, and data assimilation sources. An optimal WRF configuration is determined via validating forecasts against measurements from the highly instrumented SURFRAD sites.