Solar Power Forecasting with SunCast and GRAFS-Solar
NCAR’s philosophy in constructing SunCast is that to provide a seamless forecast across temporal scales requires multiple systems that leverage technologies at each scale. On the shortest timeframes, systems are based on real-time measurements in the field (StatCast), including specialized instrumentation such as total sky imagers (StatCast). As satellite data become available, short-range forecasts advect the observed cloud patterns (CIRACast, MADCast). The team is working together to advance numerical weather prediction for solar energy through building new modules and improving existing modules of the Weather Research and Forecasting (WRF) system, WRF-Solar. The forecasts from these models are blended and tuned to historic observations via statistical learning in a complex engineered system. Irradiance is converted to power and uncertainty is quantified with the Analog Ensemble approach. Much of this SunCast software will be OpenSource and distributed widely.
NCAR is leveraging this work to also build the OpenSource Gridded Atmospheric Forecast System (GRAFS), with the first implementation being GRAFS-Solar. This gridded system covers the continguous US and begins with NWP model output interpolated to a 4 km grid. The forecasts are tuned to historical observations using any of a host of artificial intelligence algorithms.