Development and Validation of an Operational, Cloud-Assimilating Numerical Weather Prediction Model for Solar Irradiance Forecasting

For solar irradiance forecasting, the operational numerical weather prediction (NWP) models (e.g. the North American Model (NAM)) have excellent coverage and are easily accessible. However, their accuracy in predicting cloud cover and irradiance is largely limited by coarse resolutions (> 10 km) and generalized cloud-physics parameterizations. Furthermore, with hourly or longer temporal output, the operational NWP models are incapable of forecasting intra-hour irradiance variability. As irradiance ramp rates often exceed 80% of clear sky irradiance in just a few minutes, this deficiency greatly limits the applicability of the operational NWP models for solar forecasting. To address these shortcomings, a high-resolution, cloud-assimilating model was developed at the University of California, San Diego (UCSD) and Garrad-Hassan, America, Inc (GLGH). Based off of the Weather and Research Forecasting (WRF) model, an operational 1.3 km-gridded solar forecast is implemented for San Diego, CA that is optimized to simulate local meteorology (specifically, summertime marine layer fog and stratus conditions) and sufficiently resolved to predict intra-hour variability. To produce accurate cloud-field initializations, a direct cloud assimilation system (WRF-CLDDA) was also developed. Using satellite imagery and ground weather station reports, WRF-CLDDA statistically populates the initial conditions by directly modifying cloud hydrometeors (cloud water and water vapor content). When validated against the dense UCSD pyranometer network, WRF-CLDDA produced more accurate irradiance forecasts than the NAM and more frequently predicted marine layer fog and stratus cloud conditions.