Evaluation of the ERA Reanalysis Data Accuracy for Temperature and Humidity Parameters Based on Upper-Air Soundings in Iran (1990–2020)

In this study, upper-air sounding observations from nine radiosonde stations across Iran were used to evaluate performance of the ERA5 reanalysis data in reproducing temperature and humidity parameters during the period 1990-2020. Observations were taken twice daily at 00:00 and 12:00 UTC. Temperature-related parameters included air temperature, dew point temperature (at 850, 700, and 500 hPa), and lapse rate (in three layers: the first 1-km, the first 3-km, and the 3–6 km layer from the surface). Humidity parameters include specific humidity (from 1000 to 200 hPa) and mixing ratio (at 100, 300, and 500 meters above ground level). Four statistical indicators were employed including bias, correlation at the 95% confidence level, root mean square error and Index of Agreement (IoA). Results indicate that ERA5 reanalysis data generally overestimate temperature at the lower tropospheric level at night time, while underestimations are found at daytime across most stations and seasons. At 700 hPa, Tehran, Kermanshah, and Isfahan show warm bias at daytime in all seasons, whereas Mashhad show a cold bias. At night times, all stations exhibited warm biases.
At the mid-troposphere, warm biases were observed across all stations and both times and seasons. Despite these biases, the correlation coefficients and IoA values for temperature across all levels exceeded 0.9, indicating generally good agreement. In addition, monthly mean lapse rate trends from observations and ERA5 reanalysis data showed distinct station-specific and climate-related variability. Although the frequency of surface cooling and temperature inversions-which ERA5 reanalysis fails to represent-decreased with altitude, discrepancies between observed and reanalyzed lapse rate trends were evident even in the 3–6 km layer. For example, observational and ERA5 reanalysis trends of lapse rate in 1-km layer differed at Shiraz, and also that for 3–6 km layer in Ahvaz, and Zahedan differed. Only in the 0–3 km layer both trends aligned across all stations. On a seasonal scale, reanalysis data underestimated lapse rates at all three layers at night time, while at daytime it overestimated lapse rates in the 1-km and 3-km layers and underestimated them in the 3–6 km layer. RMSE calculations revealed consistent overestimation of lapse rates across all stations, seasons, and times, with errors more pronounced in the lowest layer. IoA values indicated reasonable agreement between reanalysis and observations, particularly at daytime. Seasonal bias distributions of variables at 850 hPa were not deemed climatologically robust due to the limited number of soundings reaching that level at some stations with higher elevation. However, at 700 and 500 hPa—where more observations were available—88% of reanalysis dew point temperature estimates showed overestimation, with cold bias occurring in only 1% of cases. RMSE values of dew point temperature were higher at 500 hPa than that at 700 hPa, likely due to limited observational data at mid-troposphere levels, insufficient representation of upper-air dynamics in boundary layer schemes, and greater dependency on model-based estimates at higher altitudes. Nonetheless, IoA values showed high accuracy across all stations for reanalysis dew point temperature at 700 and 500 hPa at night and exceeded 0.8–0.9 in 89% of cases at daytime. Vertical profiles of seasonal specific humidity showed that, with the exception of Ahvaz station, the highest bias occurred near the surface and diminished with altitude. Larger biases were observed at night compared to daytime, particularly in the lower troposphere. Comparing mixing ratio estimates from reanalysis with observational data revealed overestimation at all three levels (100, 300, and 500 meters) in all stations except Ahvaz. The overestimation was more significant at night. Ahvaz was the only station where reanalysis consistently underestimated the mixing ratio during both day and night. The findings indicate that reanalysis data provide relatively acceptable performance in estimating mid- to upper-tropospheric temperature parameters (air temperature, dew point, and lapse rate), particularly during daytime and at higher atmospheric layers. However, near-surface layers, especially during nighttime, reanalysis data exhibited larger biases and errors, mainly due to reanalysis limitations in capturing surface cooling, temperature inversion, and moist boundary-layer conditions. Humidity parameters, such as specific humidity and mixing ratio, showed more significant overestimations near the surface and were generally less accurate than temperature parameters, with the exception of stations with unique climate characteristics, such as Ahvaz. In conclusion, although, reanalysis datasets cannot fully replace observational data, they offer valuable insights for climate studies, particularly in data-sparse regions and for mid- to long-term analyses. It is recommended that future studies integrate reanalysis with surface observations and satellite data to reduce uncertainties and gain a more comprehensive understanding of atmospheric processes.