Assessing the value of simulated regional weather variability in solar forecasting using numerical weather prediction

Numerical weather prediction (NWP) is currently the best tool to forecast solar radiation beyond several hours ahead. However, mainly due to the stochastic nature of clouds, spatial resolution used by NWP significantly affects the forecasting accuracy of solar irradiance and power. In this study, the effects of the simulated regional weather variability at a relatively fine spatial resolution on the forecasting accuracy of solar irradiance are systematically investigated using the Conformal Cubic Atmospheric Model (CCAM) and the Global Forecast System (GFS). Nudging from the US National Centers for Environmental Prediction (NCEP) global analysis, CCAM has been run to forecast solar radiation at a resolution of 4 km in horizontal space covering the whole Australia. For the prediction of Global Horizontal Irradiance (GHI), we find that the high-resolution CCAM generally produces more accurate forecasts than the low-resolution GFS for all nine observation stations we investigate, when using the nearest grid point approximation in combination with bias correction. Spatial averaging to a certain scale is able to enhance the performance of both NWP models in solar forecasting as measured by mean errors. However, spatial averaging, which is similar to a low resolution used in NWP models, tends to significantly and unrealistically reduce the extent of solar variability. The optimal scale of spatial averaging, when determined by the minimum of Mean Absolute Error (MAE), relies on the climatic characteristics of the location and ranges from about 100 km to about 400 km for the nine stations.