Abstract:Based on the iterative least square method, we develop two empirical Tm models for Hong Kong region: Tm_hk1 and Tm_hk2, with vertical adjustment using the radiosonde data of Kings Park station during the 2012-2017 period. The precision and reliability of the developed models, Bevis and global pressure temperature 2 wet(GPT2w), over Hong Kong region are evaluated using the sounding profiles throughout 2018. Results show that Tm_hk1, which requires the surface temperature at the station, can achieve a high precision with annual mean bias better than 0.3 K and the root mean square error(RMSE) within 1.8 K. Compared with the Bevis formula and GPT2w model, the accuracy of Tm_hk1 model increased by 35.4% and 29.7%, respectively. The Tm_hk2 model without the requirement of the meteorological parameter can achieve the same accuracy as the GPT2w model, and annual mean RMS error of both models are better than 2.5 K. Bevis formula has the worst accuracy(RMSE=2.7 K) and a large negative bias of -1.8 K. From the analysis, it can be found that the precision of Bevis, Tm_hk2, and GPT2w models show an obvious seasonal variation. The overall precision of the models during summer is higher(RMSE=1.3-2.2 K) than that during winter(RMSE=3.0-4.4 K). Furthermore, Tm_hk1 model performs the highest precision and applicability during all seasons, with the RMSE ranging from 1.4-2.4 K.