Assessment of Solar Radiation Datasets for Building Energy Simulation

Accurate solar radiation data are essential for reliable building energy simulations, particularly for policy making. However, uncertainty in solar input, especially in regions with limited ground-based measurements, can significantly affect simulation outcomes. This study investigates the impact of different solar radiation datasets on building energy performance simulations across two climatically distinct years, 2015 and 2024, in a subtropical urban environment. Solar inputs from ERA5, CAMS, and NASA POWER were compared against a regional reference from the Brazilian National Institute for Space Research (INPE). In addition to energy simulations, the datasets were evaluated using statistical metrics including root mean square error (RMSE), mean bias error (MBE), and Pearson correlation. NASA POWER showed the best agreement with ground data and yielded simulation results that were reasonably aligned with observed cooling loads and thermal comfort in both years, with slight overestimations in cooling demand and overheating hours. CAMS maintained consistent performance across both years and produced the lowest cooling and overheating estimates, slightly underestimating results while closely matching monthly trends. ERA5 exhibited the largest positive bias in solar input, particularly in DNI, leading to substantial overestimation of cooling demand, up to 34% in 2024, especially during heatwaves. These discrepancies highlight the sensitivity of energy simulations to solar input selection and the importance of using validated high-quality datasets to ensure reliable modeling under increasing climate variability.