A CERES-based dataset of hourly DNI, DHI and global tilted irradiance (GTI) on equatorward tilted surfaces: Derivation and comparison with the ground-based BSRN data

Abstract

The NASA CERES SYN1deg(Ed4.1) satellite-based products include hourly global horizontal irradiance (GHI), diffuse horizontal irradiance (DHI) and direct horizontal irradiance (DirHI) as well as hourly solar zenith angle (SZA), and the direct normal irradiance (DNI) can thus be calculated by dividing the DirHI by cos(SZA). The spatial resolution of the dataset is 1° latitude by 1° longitude, and the time span is from March 2000 to near present (October 2023 as of this writing). While the GHI of the dataset agrees well with the ground-based Baseline Surface Radiation Network (BSRN) data, the DHI and DirHI, and thus DNI, show appreciable biases against the BSRN data. Nevertheless, the uncertainty of the DHI and DNI, as exhibited in the standard deviations of their differences from the BSRN data, are smaller than results from the DirIndex model, a global-to-beam model, applied to the CERES data, albeit the model is one of the two best out of 140 models according to a study. This means that the original CERES hourly DNI, though biased, renders the spatiotemporal variability better than does the DirIndex model. For this reason, we decided to perform a bias correction on the hourly DHI and DNI. With the corrected hourly DHI and DNI, we are able to use the isotropic model to calculate the global tilted irradiance (GTI) and the global tracker irradiance (GTrI) on an hourly basis. The BSRN DHI and DNI, on the other hand, are available on near-instantaneous time scales (1-, 2-, 3, or 5-minute intervals) and thus are more conducive to the isotropic model, and for the first time, we use the BSRN data to validate the GTI and GTrI, as opposed to GHI and DHI that have previously been used for validation purpose. For comparison and validation, we also use the monthly-mean-based LJCR method, an independent method as used by RETScreen®, which requires monthly mean GHI and DHI, to directly calculate the monthly mean DNI, GTI and, to extend the model, GTrI, and the results agree well with the BSRN data, which proves the LJCR method itself and corroborates the corrected DHI and DNI as well as the derived GTI and GTrI. The monthly mean DHI and DNI from the Whitlock Method, also monthly-mean-based, are evaluated as well, and we find that minor modification can dramatically improve the method.