Comparison of Solar Energy Nowcasting Techniques

The need for improved solar energy forecasting is increasing as more solar farms come online nationwide. With an increasing share of the energy portfolio for many electrical utility companies, incorporating the highly variable power output from solar arrays into the grid is becoming increasingly difficult. Accurate predictions of the irradiance received at individual solar farms are vital. Utilities require these predictions in the nowcasting time scale (minutes to hours ahead) with high temporal frequency in order to perform load balancing properly.

The SunCast solar energy forecasting system is designed to meet these challenges. Several forecasting components have been developed to predict irradiance in different time windows and at different spatial scales. These include various site-based statistical learning methods (StatCast), dynamic and advective forecasts from stereoscopic analysis of total sky imager photos (TSICast), satellite cloud advection forecasting (CIRACast), multi-sensor advective diffusion forecasting (MADCast), and a modified version of the Weather Research and Forecasting (WRF) numerical weather prediction model (WRF-Solar).

In this study we compare the performance of these components of SunCast for a partly cloudy day. Days with intermittent clouds are the greatest challenge for any solar forecasting method. This comparison will reveal the strengths and shortcomings of the various techniques, as we seek to continually improve them and blend them into a single solar energy forecasting product.