Using Satellite Information to Evaluate Cloud Forecast from WRF-Solar EPS

Cloud forecasting is an enormous challenge in numerical weather prediction (NWP) models because of the complex physical processes, high variability in spatial and temporal scales, and lack of observations for model evaluation. The capability to represent cloud fields in the NWP models directly impacts the accuracy in predicting surface solar irradiance. Verifying cloud forecasts from NWP models is essential to investigate the source of uncertainty and error stemming from predicting various types of clouds. However, it requires high-quality observations containing various information of clouds and in-depth analysis over a wide range of regions to assess the cloud fields. In this study, we produce day-ahead cloud forecast over the contiguous United States (CONUS) for 2018 using the Weather Research and Forecasting-Solar Ensemble Prediction System (WRF-Solar EPS) which is a state-of-the-art ensemble NWP model specialized for solar applications. The strengths and limitations of WRF-Solar EPS in reproducing cloud fields is diagnosed using satellite observations from the National Solar Radiation Database (NSRDB). The frequency of clouds and various cloud detection metrics including the probability of detection (POD), the false alarm rate (FAR), the hit rate (HR), Kuiper’s skill scores (KSS), and mismatched cloud frequency (MCF) are calculated to assess the performance of WRF-Solar EPS. In the first part of this study, we focus on monthly analysis using the detection metrics to account for seasonal performance of WRF-Solar EPS. In the second part, the MCF classified by cloud top height and cloud optical depth is analyzed to investigate the model’s capability to predict nine different types of clouds. The study exhibits that the WRF-Solar EPS has difficulty predicting optically thin clouds; overall MCFs show 46% (cumulus), 34% (stratocumulus), 19% (stratus), 33% (altocumulus), 23% (altostratus), 16% (nimbostratus), 27% (cirrus), 13% (cirrostratus), and 8% (deep convective) for the nine cloud types.