Localized Uncertainty Model of the CAMS Radiation Service Irradiance Product

Kurzfassung

The CAMS Radiation Service (CRS) provides operational historical irradiance estimates at ground level for Europa/Africa (since 2004) and for Asia/Oceania (since 2016). The provision of this data is an important service to the solar energy community as it allows an open access to high quality historical irradiance estimates to a wide variety of solar community stakeholders (project developers, energy traders, researchers, etc). The performance of the CRS irradiance is assessed quarterly in the publicly available CRS Evaluation and Quality Control (EQC) reports found in CRS-EQC (2025). In these EQC reports ground observations from high quality selected sites around the world are used to assess the error of the CRS irradiance in term of well established statistical metrics (MBE, RMSE, etc.). The fact that the measurement stations used in these reports are sparse and very far apart from each other limits our understanding of the spatial as well as the systematical error characteristics of the CRS. Indeed, the stations are subject to very different meteorological situations and climatic conditions. It is often difficult to compare the evaluations made for the different stations, as it is difficult to assess what proportion of the error is due to the specific situation of the station and what represents a deficiency in the physical model. Looking into a continuous improvement of the CRS service, one of the aims in the CAMs sErvice evOlution (CAMEO) project is to assess and quantify the error patterns of the CRS irradiance on a localized pixel scale. In order to do so, first the spatial error patterns of the CRS must be studied and understood. Then, the knowledge obtained from this spatial analysis is used to identify the variables that have a big impact on the error of the irradiance estimations. Finally, these variables are used to build an error model that gives the local expected error for every individual estimate (pixel) on the domain. In this abstract we present the work done on the spatial error analysis of the CRS. Using multiple years of data of two highly dense populated pyranometric networks (~300 stations) monthly statistical error maps were generated. These maps enable the visualization and analysis of the spatial error structures of the CRS. We show that most of the spatial patterns observed in these monthly error statistics can be explained by the mean solar irradiance of the pixel on consideration and by the station altitude. These results are currently being investigated from a modeling perspective in the CRS to verify whether this effect results from issues in the coupling with clouds inside the retrieval scheme. At the time of writing the quantification of the localized uncertainty model is being developed. This should result in a model that will be able to estimate the statistical uncertainty of the CRS irradiance per pixel. As a result, the CRS output will provide along the usual pixel irradiance estimate also the estimate of the irradiance uncertainty.