Seasonal and spatial variability in the accuracy of hourly ERA5 and MERRA-2 reanalysis datasets: A 14-year comparison with observed meteorological data in Türkiye

Abstract

This study comprehensively evaluates the performance of ERA5 and MERRA-2 reanalysis datasets in representing key meteorological parameters consisting of air temperature, mean sea level pressure, and relative humidity across Türkiye between 2010 and 2023. By comparing reanalysis data to observations from 116 meteorological stations, the analysis provides critical insights into their spatial and seasonal accuracy, revealing notable strengths and limitations. The findings confirm that ERA5 consistently outperforms MERRA-2 across all three parameters, exhibiting higher determination coefficients (R²), lower root mean square errors (RMSE), and reduced mean bias errors (MBE), particularly for air temperature and relative humidity. ERA5 demonstrated superior performance in temperature representation, with seasonal R² values ranging from 0.85 to 0.91, while MERRA-2 exhibited lower performance, ranging from 0.75 to 0.86. Similarly, ERA5 outperformed MERRA-2 in relative humidity estimation, achieving R² values between 0.50 and 0.75, compared to MERRA-2's lower range of 0.33 to 0.62. Both datasets performed comparably for mean sea level pressure; however, ERA5 achieved slightly lower RMSE and MBE values, particularly in winter and regions with complex topography, where MERRA-2 systematically underestimated pressure. The seasonal and spatial analyses highlight ERA5's finer spatial resolution and enable a more accurate representation of meteorological variability, particularly in topographically complex and coastal regions. Its reduced systematic biases and improved accuracy in transitional seasons underscore its suitability for high-resolution applications, such as urban climate modeling, air quality research, and hydrological simulations. Conversely, MERRA-2, with its coarser grid and pronounced biases in colder months, remains suitable for broader-scale climatological studies but may require bias correction for localized applications.