Direct normal irradiance modeling: Evaluating the impact on accuracy of worldwide gridded aerosol databases

Abstract

Clear-sky solar irradiance is required in multiple solar applications. In the general case, it has to be simulated with a clear-sky solar radiation model with inputs from synthetic data sources such as atmospheric reanalyses or similar numerical weather products. Given the high sensitivity of direct normal irradiance (DNI) to aerosol optical depth (AOD), it is expected that the use of such synthetic sources for AOD results in a non-negligible change in the DNI simulation error as compared to the use of AOD values retrieved at ground stations. This study evaluates the DNI simulation error of the high-performance SMARTS clear-sky solar radiation model using AOD from two well-known atmospheric reanalyses comparatively to the use of AOD values retrieved at ground sites. The study is conducted at four arid sites with one year of collocated DNI observations from the Baseline Surface Radiation Network (BSRN) and AOD observations from the Aerosol Robotic Network (AERONET) at each site. It is found that both DNI bias and dispersion using synthetic AOD databases may increase up to 6 times compared to when AOD values are retrieved from AERONET, thus confirming the general need for site-adaptation post-processing techniques to obtain low-uncertainty solar resource assessments.