Modelling photovoltaic system output using ERA5 reanalysis data validated with high-resolution actual measurements

Modelling photovoltaic (PV) system output is crucial for designing optimal systems, predicting system performance, and efficiently managing the system. It also helps assess the solar potential of a site, predict energy yield, and optimize system configuration.  Irradiance components such as global horizontal irradiance (G_h) diffuse horizontal irradiance (G_d), and direct normal irradiance (G_bn) are required for modelling PV output. However, in many locations, the availability of these components is limited, with some stations only providing G_h, which is insufficient for accurate PV modeling. To address this, gridded reanalysis datasets like ERA5 can be utilized, as it offers comprehensive irradiance components. Reanalysis data integrate historical weather and climate observations from multiple sources with a consistent weather model over an extended period. ERA5, provided by the European Centre for Medium − Range Weather Forecast (ECMWF), offers comprehensive hourly global data, available for public download. This dataset not only aids PV modelling but also enables the study of past and current solar energy trends for any given area, with data available from 1940 to the present. Despite the extensive use of reanalysis data, it is essential to evaluate their quality and suitability for specific study areas, given the varying climatic conditions.  This study assesses the quality and applicability of ERA5 irradiance data for modelling PV power output at a rooftop PV system installed at the CES Solar Cells Testing Canter (CSSC) in Thailand, located at 13.575° N, 100.443° E. The site's instruments measure G_h and in plane irradiance (G_t) at the same tilt angle as the PV array, along with high temporal resolution PV power output data recorded every minute. The data collected from ERA5 consist of G_h, beam horizontal irradiance (G_b), ambient temperature (t2m), wind speed (calculated from 10 m u−component and v−component of wind). To estimate the PV system's power output, pvlib-python (PVLIB), an open-source tool for modelling and simulating the performance of PV systems, is employed. The overall configuration of PVLIB for this study shown in figure 1. PVLIB takes in input such as irradiance components, meteorological parameters (wind speed and t2m), solar positions, PV module orientation (tilt and azimuth angle), and PV module specifications. The irradiance components include G_h, G_bn, and G_d, where G_d and G_bn were calculated. PVLIB employed transposition models to transform irradiance components into plane–of–array irradiance (POAI) and subsequently estimated the power output. The result of comparing hourly G_h from ERA5 and CSSC pyranometer using data for the year 2020 is shown in figure 2. The results show a strong relationship. Despite some underestimation by ERA5 and moderate error as indicated by normalized mean bias (NMB), and normalized mean error (NME), ERA5 reanalysis data appears to be reasonable. This supports the potential of use of ERA5 to be use as input in PV modelling with the need for correction or calibration to improve accuracy.  The power output from PVLIB is also expected to have reasonable agreement with the actual power of PV array. It should be noted that the results presented here are part of an ongoing research; thus, they are subject to improvement, and additional results will be presented.