Regional validation of the solar irradiance tool SolaRes in clear-sky conditions, with a focus on the aerosol module

The Solar Resource estimate (SolaRes) tool based on the Speed-up Monte Carlo Advanced Radiative Transfer code using GPU (SMART-G) has the ambition to fulfil both research and industrial applications by providing accurate, precise, and high-time-resolution simulations of the solar resource. We investigate the capacity of SolaRes to reproduce the radiation field, relying on 2 years of ground-based measurements by pyrheliometers and pyranometers acquired in northern France (Lille and Palaiseau). Our main objective is to provide, as a first step in clear-sky conditions, a thorough regional validation of SolaRes, allowing us to investigate aerosol impacts on solar resource. We perform comparisons between SolaRes-simulated and clear-sky-measured global horizontal irradiance (GHI), direct normal irradiance (DNI), diffuse horizontal irradiance (DifHI), and global and diffuse irradiance on a tilted plane (GTI, DifTI), and we even consider the circumsolar contributions.Using spectral aerosol optical thickness (AOT) data sets as input, which are delivered by the AErosol RObotic NETwork (AERONET) and the Copernicus Atmosphere Monitoring Service (CAMS), we examine the influence of aerosol input data sets in SolaRes on the comparison scores. Two aerosol models are mixed to compute aerosol optical properties. We also perform a sensitivity study on the aerosol parametrisation and investigate the influence of applying more or less strict cloud-screening methods to derive ground-based proof data sets of clear-sky moments.SolaRes is validated with the (relative) root mean square difference (RMSD) in GHI as low as 1 % and a negligible mean bias difference (MBD). The impact of the cloud-screening method in GHI is 0.5 % of RMSD and 0.3 % of MBD. SolaRes also estimates the circumsolar contribution, which improves MBD in DNI and DifHI by 1 % and 4 %, respectively, and RMSD by similar to 0.5 %. MBD in DNI is around -1 % and RMSD around 2 %, and MBD in DifHI is 2 % and RMSD around 9 %. RMSD and MBD in both DNI and DifHI are larger than in GHI because they are more sensitive to the aerosol and surface properties. DifTI measured on a vertical plane facing south is simulated by SolaRes with an RMSD of 8 %, comparable to that obtained for DifHI. Our results suggest a strong influence of reflection by not only ground surface but also surrounding buildings.The sensitivity studies on the aerosol parameterisation show that the spectral AOT contains enough information for high performance in DNI simulations, with low influence of the choice of the aerosol models on the RMSD. However, choosing a model with smaller aerosol single scattering albedo significantly decreases SolaRes DifHI and GHI. The best combination in Lille and Palaiseau consists of continental clean mixed with desert dust. Also, complementary information on angular scattering and aerosol absorption provided by the AERONET-inverted model further improves simulated clear-sky GHI by reducing RMSD by similar to 0.5 % and MBD by similar to 0.8 %. Eventually, the choice of the data source has a significant influence. Indeed, using CAMS AOT instead of AERONET AOT increases the RMSD in GHI by similar to 1 % and MBD by similar to 0.4 % and RMSD in DNI by 5 %. The RMSD in GHI remains slightly smaller than state-of-the-art methods.