Irradiance modelling using satellite and ground based methods

This thesis investigates a key area of uncertainty in solar irradiance research: the level of accuracy of geostationary satellite-derived terrestrial solar irradiance estimates. A method for identifying sources of error via sky camera imagery is presented along with an assessment of irradiance enhancement and a method for producing GHI estimates from sky camera imagery. Pyranometer data from 2017-2020 and sky camera images from 2018-2020 were taken from Loughborough university and compared to Satellite data (2018-2020) for the Loughborough locale. Ancillary GHI data from BRSN and UK met office for 2018-2020 were used in chapter 4.

Validation data from a Secondary standard thermopile pyranometer at Loughborough University is used from 2017-2020 and reveals the frequency and magnitude of irradiance enhancements (0.2%-16.7% of measured data points during 2017 dependent on model parameters and timeframe). Irradiance enhancement is defined in this study as GHI values in excess of the Clear Sky Maximum GHI by 50W/m² or more when TL values are set to 1. These enhancements are identified using pyranometer data and corresponding sky camera imagery. Broken cumulus type clouds are shown to occur in sky camera images during 97.42% of enhancements recorded by the CMP11 pyranometer. Cirrus type clouds were shown to occur less frequently (1.99%) during enhancements than mean monthly occurrence (12.5% and 10.3%). Hourly time averaged data was shown to mask irradiance enhancement. A maximum instantaneous value of irradiance enhancement of 564 W/m² above clear-sky irradiance is observed.

A sky camera is used to identify weather conditions that lead to the greatest errors in irradiance estimates provided by EUMETSAT from the METEOSAT 11 satellite for the period of 2018-2020. Ground-measured and satellite-modelled irradiance values show consistent agreement on clear days. It is found that satellite irradiance estimates on days where sky camera imagery show clouds have considerable bias and random error when compared to pyranometer data and are dependent on cloud type. Cumulus type clouds produce the greatest daily bias (-38%) and random error (+/-32%) in satellite hourly values when compared to pyranometer values. Cirrus type clouds produce the opposite effect to a lesser extent. Satellite irradiance estimates are shown to fail to record irradiance enhancements seen in the pyranometer data and can fail to record occlusions, reducing irradiance by ~40% in sub-hourly time frames. Comparison of Loughborough spot measured pyranometer data with BRSN spot data from Cabauw (NL) show an annual bias of -12%.

In addition, the sky camera produces images that are of sufficient quality that a method of deriving irradiance estimates from these is presented. Camera-derived solar irradiance mean bias error (MBE) over the period 2018-2020 at Loughborough University was found to be -1.27% compared to that of the validation pyranometer (with a value of 13% for RMSE over the same period). Comparison of instantaneous camera and pyranometer spot data showed bias within the clear sky model, specifically around the use of monthly averaged TL values. In addition cloud enhancement and ice cloud signals within the pyranometer dataset could not be replicated by the sky camera model. However, solutions to these shortcomings were identified.