IEA-PVPS has published a report to help the solar industry select appropriate surface radiation models and data providers based on location and application requirements.
Now, IEA-PVPS Task 16 has produced a global benchmark of modeled solar radiation data to help the solar industry make better decisions about solar resource evaluation.
The study benchmarks model-based direct normal irradiance (DNI) and global horizontal irradiance (GHI) data using 129 high-quality ground-based irradiance measurement stations around the world from 2015 to 2020. It then compares the DNI and GHI values. GHI evaluated 10 solar radiation datasets, either public or commercial, against this benchmark. The datasets are ACCESSG3, DWDSARAH, CAMS v3.2, KNMISEVIRI, CAMS pre-v4, METEOTEST, CERES, NSRDBGOES, CSIROHIMAWARI and Solargis.
The comparison is performed with an hourly time resolution and the performance of the modeled data is analyzed for different regions and climate zones.
The IEA-PVPS says the quality of the reference database is ensured by selecting data based on a “comprehensive set of best practices and recently implemented quality control procedures.” These include automated and manual data quality control tests performed by a team of experts on all stations, resulting in a ticket describing the quality of each timestamp. Out of a total of 129 stations, 31 are located in Africa, 31 in Asia, 27 in North America, 20 in Europe, 13 in Australia, five in South America and two in Antarctica. The report calculates the mean deviation, standard deviation, and standard deviation between the benchmark and each station for the period 2015-2020.
“Comparison results have shown significant deviations in the performance of different modeled datasets,” the report says. “In particular, it was found that the most suitable data set actually depends on both the location and the climate or continent of interest. Some stations are particularly challenging for some models, as evidenced by the large deviation of several data sets in difficult environments (e.g., high mountains or coastal areas).
The report showed that model errors and differences between different modeled data sets are much larger for the DNI than for the GHI, due to the former’s greater sensitivity to aerosols, clouds, and altitude, among other factors.
In terms of dataset-specific results, the study shows that CERES had significantly larger biases than all other satellite-derived datasets, likely due to its coarser resolution. According to the IEA-PVPS, Solargis showed the lowest average deviation and was also the best at many individual stations.
“From a methodological point of view, this benchmark highlighted the importance of the quality of the reference data. Without a strict quality control procedure, no real validation can be done and there is a risk of erroneous results,” the researchers concluded.