Bland-Altman Analysis of Open-Access Online Weather Data

Solar radiation data is essential for evaluating solar energy potential; designing, optimizing and developing predictive models for solar energy systems; and other applications. While weather stations provide reliable data, their high installation and maintenance costs lead to data gaps in many regions. Satellite-derived data presents a cost-effective alternative, offering broad coverage. However, satellite-derived data require continuous evaluation to prove them as reliable substitutes for ground measurements. This study compared satellite-derived (SD) solar irradiation data from two sources, NASA's POWER and PVGIS, against ground-measured (GM) data from the World Radiation Data Centre (WRDC). The comparison relied on data from 171 WRDC stations spanning 2005 to 2020. The Bland-Altman method was the primary statistical measure used because of its ability to determine agreement between data sources and identify systematic bias; this involved constructing Limits of Agreement (LOA), within which the most differences between the two data sources are expected to lie. Additional statistical measures, including r-correlation, root mean square error analysis, and t-value analysis, were employed to validate the BA findings and to investigate the influences of latitude, diurnal periods and annual seasons on the agreement between the SD and GM data. The results of the study showed that, when compared with the WRDC data, the POWER data exhibited limits of agreement (LOA) of 0.45±4.78 MJ/m²/day, while PVGIS data had LOA of 0.47±5.11, with the range of LOA being 9.55 and 10.23, respectively. In addition, distinct relationships between the range of LOA and latitude and season were visually identified from the plots, indicating that these factors affect the agreement between the SD and GM data. The narrower LOA range of the POWER data suggested it to be the more reliable substitute for GM data.