Evaluation of numerical weather prediction for solar irradiance forecasting in the CONUS

The demand for solar energy is as large as ever. A primary advantage of solar power is in its relatively predictable nature. However, clouds associated with dynamic weather systems reduce solar radiation at the earth's surface. The weather patterns affecting solar radiation can be resolved through numerical weather prediction (NWP) tools. However, these models contain inherent biases, regional or otherwise, limiting their efficacy.

This study validates the ECMWF, NAM, and GFS solar irradiation forecasts for the southwestern U.S. using SURFRAD and California Irrigation Management Information System (CIMIS) measurements in 2009. By comparing NWP forecast data to ground measurements, innate bias errors were discovered and synthesized into a model output statistics (MOS) correction scheme. Applicable to future forecasts, the MOS correction minimizes model bias producing a more accurate solar forecast.

Specifically, the NAM solar forecast was shown to over predict GHI during clear sky situations and under predict GHI for overcast conditions. Since the bias error of the NAM forecast was consistent and predictable, MOS could be applied to design a correction scheme relating expected bias error to forecasted conditions. After application of this scheme, NAM mean monthly bias error across all ground measurement sites was reduced from up to 25% to less than ±5%.