Comparison of Tropospheric Temperature Profiles from Ground-based Microwave Radiometer and Radiosonde at Xi’an, Central China

The reliability and accuracy of temperature profiles retrieved by ground-based microwave radiometer (MWR) using neural networks at Jinghe meteorological station (Xi’an, China) was evaluated by comparing the temperature retrievals against simultaneous radiosonde observations. For the comparison purposes, temperature measurements from MWR and radiosonde with altitude difference of less than 10 m and time difference of less than 5 minutes were matched with each other. Finally, a total of 1240 pairs of profiles have been compared for the period from June 2017 to July 2019. The MWR-derived temperature showed consistent negative bias of 0.14–2.44 °C throughout the retrieved profile except at 10 km where with a value of +0.22 °C. The root mean square error is less than 2°C for heights below 100 m, and rapidly increases to 4°C at 2 km and remained stable thereafter. Linear regression between temperature derived from MWR and radiosonde at different altitudes indicated that both the slope and intercept showed altitude-dependent behavior. Fairly good agreement between radiosonde and MWR-derived temperature is observed in the lowest 1 km, with linear regress slopes between 0.91–1.00 and correlation coefficients exceeding 0.96. However, both the slope and intercept decreased dramatically with height. The altitude variations of the slope and intercept suggest that the MWR tends to overestimate the temperature in cold season but underestimate the temperature in warm season in the middle and upper troposphere. Temperature gradient derived from radiosonde showed that the occurrence frequency of temperature inversions peaked at 28% at near surface at Xi’an, and declined to 8% at 2 km. MWR failed to distinguish the temperature inversion in most cases. The missed detection ratio is approximately 40% at surface and beyond 80% at altitude higher than 500 m. There is still big room for improving the accuracy of the temperature retrievals of MWR, especially in the aspect of temperature inversion detection. Our findings suggest that the MWR data needs to be used with caution in the air pollution study.