Finding Reliable WRF-Solar Physics Options for Forecasting Clouds and Solar Irradiance Using the NSRDB

Representing clouds in numerical weather prediction (NWP) models is the main challenge when forecasting solar irradiance. In addition, NWP users need to select the right combination of physics parameterizations among various choices of physics options within the NWP model to generate accurate solar forecasts. However, addressing this challenge is not straightforward for users because of the limited computational resources to test various physics options and the need for high-quality gridded observations to evaluate the model results for different types of clouds. This study focuses on finding the best WRF-Solar physics options that can accurately provide cloud mask and solar irradiance forecasts. In this work, we tested WRF-Solar model with 10 different combinations of the key parameterizations that represent radiation, microphysics, cumulus, shallow cumulus, aerosol effects, planetary boundary layer and land surface. 363 sets of day-ahead forecasts covering the year of 2018 were simulated at 9-km and 15-minute spatiotemporal resolution across the contiguous United States. To evaluate the day-ahead forecasts, we used the National Solar Radiation Database (NSRDB) as the satellite-derived solar radiation data, which offers the opportunity to conduct an exhaustive analysis of clouds and solar irradiance for large regions. The first part of this study evaluates cloud mask forecasts from 10 WRF-Solar experiments for different types of clouds using cloud detection metrics based on the information about clear-/cloudy-sky conditions from WRF-Solar and NSRDB. In the second part, we analyzed statistical metrics to evaluate the performance of each member in forecasting solar irradiance. Preliminary results indicate that the best configuration of WRF-Solar shows significant reductions in the Mismatched Cloud Frequency (MCF) for thin (60%–65%), middle-thickness (68%–74%), and thick (66%–77%) clouds when compared to WRF-Solar reference configuration. Consistent with the results in MCF, the best configuration reduces the bias in the day-ahead GHI forecasts from WRF-Solar reference configuration by 33% during 2018.