The design and estimation of the performance of any solar energy system requires knowledge of solar radiation data obtained over a long period of time. The use of solar photovoltaic (PV) energy in Ireland is growing, leading to more interest in accurate solar shortwave radiation (SW) climatology. Reanalysis models use observational data from the past to simulate climatology. The network of stations measuring radiation is sparse, therefore, reanalysis datasets are used as a representation of climatology. The accuracy of reanalysis SW data can in part be explained by linking it to cloud amount in reanalysis. In this study, time-series analysis is performed to identify links between errors in SW and cloud structures at different spatial scales, by making use of satellite imagery and reanalysis cloud data. This study examines two popular reanalysis datasets, MERRA2 from NASA and ERA-Interim from ECMWF, with the aim of establishing the skill of reanalyses when compared to ground measurements to find which is more suitable for SW over Ireland. Reanalysis datasets are compared with a representative selection of Irish pyranometer data for time periods of up to 35 years, and standard skill scores (bias, RMSE and Pearson’s correlation) are calculated. Scores relative to climate (Anomaly Correlation Coefficient (ACC)) are also calculated to compare the performance in different seasons.
Error is defined by subtracting pyranometer measurements from reanalysis model estimations, accordingly a positive error signifies the model SW is greater than the observed SW from the pyranometer. The standard skill scores suggest that ERA-Interim is a better overall representation of SW than MERRA2. Both reanalyses overestimate total daily SW compared to ground observations, as seen in previous studies with different reanalysis datasets [1, 2].
Other work [1] has found that the original MERRA reanalysis (before MERRA2) and ERA-Interim often simulate clear sky conditions when actual conditions are cloudy. The opposite was also true, although less pronounced: where actual clear sky conditions are simulated as cloudy. Variations in SW occur primarily from interception by clouds between observation stations and the sun. This is also true in this study as all stations show a larger absolute value for positive error events compared to negative error events.
The mean error was used to identify individual events with large errors in SW. These events were analysed to find the prevailing cloud structure using both satellite imagery and the cloud data from reanalysis datasets. By linking cloud structure and errors in this way, initial analysis suggests that convective clouds are a source of negative bias in MERRA2 radiation, whereas frontal clouds are a source of positive bias. The next step is to develop objective classification of cloud structures in numerical weather prediction models. Knowledge gained will be used to help reduce errors in SW forecasts.
[1] A. Boilley and L. Wald, Comparison between meteorological re-analyses from ERA-Interim and MERRA and measurements of daily solar irradiation at surface. Renewable Energy 75:135-143, 2015.
[2] X. Zhang, S. Liang, G. Wang, Y. Yao, B. Jiang and J. Cheng, Evaluation of the reanalysis surface incident shortwave radiation products from NCEP, ECMWF, GSFC, and JMA using satellite and surface observations. Remote Sensing 8(3):225, 2016.