Multi-scale accuracy evaluation and adaptability analysis of different atmospheric weighted mean temperature models in China

Abstract The atmospheric weighted mean temperature (Tm) plays a critical role in deriving precipitable water vapor (PWV) from zenith wet delay (ZWD) in GNSS meteorology. Although numerous empirical Tm models have been rigorously validated at regional and global scales, their applicability and reliability across China remain insufficiently characterized. To address this gap, this study systematically evaluated four representative Tm models – Hourly Global Pressure and Temperature 2 (HGPT2), Global Pressure and Temperature 3 (GPT3), Global Tropospheric Model (GTrop), and China Tropospheric Model (CTrop) – using stratified meteorological data from 87 radiosonde stations in China (2010–2019). Model performance was assessed through bias and root mean square error (RMSE) metrics. Key findings reveal significant spatial heterogeneity in model accuracy across China’s four geographic regions. The GTrop and CTrop models demonstrated superior accuracy and nationwide stability, whereas HGPT2 and GPT3 exhibited pronounced limitations in mid-to-high latitudes. Notably, the CTrop model outperformed GTrop at select stations, achieving the lowest overall bias (0.39 K) and RMSE (3.95 K) among all models. Altitudinal analysis indicated that all models performed optimally below 1,000 m, with GTrop and CTrop showing marked accuracy improvements at higher elevations. Temporal analysis further highlighted CTrop’s robustness, exhibiting minimal errors across seasonal and diurnal variations. However, all models were constrained by limited Tm value ranges. Collectively, these results establish CTrop as the most accurate and stable Tm model for PWV retrieval in China.