The increasing penetration of solar power into the electricity grid has created the need for accurate solar power forecasts for facilitating safe and reliable electricity grid management. Numerical Weather Prediction (NWP) is a common method for forecasting solar irradiance beyond several hours ahead, which is crucial, particularly for applications in day-ahead energy markets as well as for energy management at different scales, ranging from utility-wide to residential level. In this study, we evaluated the ability of the Australian Bureau of Meteorology's (BoM) numerical weather prediction system, the Australian Community Climate and Earth-System Simulator (ACCESS) model, to forecast solar irradiance. The global horizontal irradiance (GHI), direct horizontal irradiance (DIRHI) and diffuse horizontal irradiance (DHI) forecasts from the APS2 version of the regional (ACCESS-R) and the city scale (ACCESS-C) mesoscale models of ACCESS were validated against irradiance measurements from 11 ground stations across different climatic conditions in Australia. While the high resolution (1.5 km grid) city models were found to better capture the cloud variability in comparison to the coarse resolution (12 km grid) regional model, the forecast performance for both models were comparable in terms of Root Mean Square Deviation (RMSD). The annual average RMSD for both models ranged between 105 W/m(2)-156 W/m(2), 131 W/m(2)-181 W/m(2) and 71 W/m(2)-95 W/m(2) for horizontal global, direct and diffuse components of irradiance, respectively, across different locations. Moreover, it was found that both models have a tendency to overpredict DHI up to 50% during clear skies, provided the sky conditions are correctly forecasted. Further, selection criteria based on forecasted sky conditions are presented to help identify the more accurate forecasts among the city and the regional model.