Assessing model performance of daily solar irradiance forecasts over Australia

Abstract In response to the rapid solar power installation worldwide, the solar industry is calling for more accurate solar irradiance forecasts with finer temporal and spatial granularity. It is timely to investigate if this need has been properly met by recent advancements in numerical weather prediction modelling. In this study, we validate and compare the current ability of three leading numerical weather prediction models to forecast daily solar irradiance for Australia. We found that all three models investigated perform well in the middle and west of Australia where clear sky weather prevails but struggle with forecasting solar irradiance in climatologically cloudy areas to some extent. In particular, the Global Forecast System (GFS) tends to significantly overpredict solar irradiance in southeastern Australia including Tasmania whilst the Australian Community Climate and Earth-System Simulator (ACCESS) system systematically underpredicts solar irradiance in northern Australia. The recent ERA5 reanalysis, which employs the Integrated Forecast System (IFS) to forecast solar irradiance, performs relatively robustly across Australia without notable deficiency, earning an overall forecast skill (defined as relative improvement in RMSE against 1-day persistence of clear-sky index) of 0.38 for 12-hour ahead forecasts of daily solar irradiance. An increase of the forecast skill to 0.44 is observed by linearly blending the three models.