Increasing Solar Energy Forecast Skill Using a Mesoscale Ensemble

As solar energy installation accelerates, so does the need for more accurate and reliable weather forecasts. Accurate prediction of incoming solar radiation that includes forecast uncertainty is critical for managing energy reserves, effective market trading strategies, and most importantly delivering reliable power. Desirable aspects of solar forecasts include those that are best able to capture solar variability (high accuracy and forecast-observation association) and are probabilistically reliable with the smallest possible range in uncertainty (calibrated and sharp). Our results use a 15-member Weather Research and Forecasting (WRF) model ensemble to demonstrate the gains in forecast accuracy skill over single-model WRF forecasts at numerous Vermont locations for a five-day forecast horizon. Forecasts are also evaluated against a 10-year climatology-based forecast. Probabilistic aspects of the WRF ensemble forecasts are also discussed.

When compared to climatology, individual deterministic WRF global horizontal irradiance (GHI) forecasts are more accurate than climatology through a three-day horizon, after which the highest-resolution WRF forecasts with grid lengths between 1.6 and 7 km become less skilled than climatology, on average. Ensemble forecasts, however, remain more skilled than climatology throughout the five-day horizon, with mean absolute error skill scores over climatology ranging from 33% to 21% for the one day to five day horizons, respectively. Our results suggest that an ensemble of WRF forecasts increases predictability by at least two forecast days over single (deterministic) high-resolution WRF models, and extends the range of actionable solar predictability beyond the day-ahead market into the medium-range forecast regime.