Accuracy Test and Analysis for GPT2w Model in China

GPT2w model is commonly used to calculate the meteorological parameters at certain location, such as temperature, weighted mean temperature, pressure and vapor pressure. It is also the public empirical model for tropospheric delay with the best nominal accuracy. In this paper, meteorological sounding data from 2013-2015 of 86 stations in China is used, which have participated the global meteorological exchange. Precisions of meteorological parameters from GPT2w are examined and analyzed. It turns out that the average bias (Bias) and root mean square error (RMS) of temperature are 1.31℃ and 3.62℃, respectively. For weighted mean temperature, the Bias is -1.58 K and the RMS is 4.07 K. For pressure and vapor pressure, the absolute values of Bias are smaller than 1 hPa, and the RMS are 6.98 hPa and 3.04 hPa, respectively. Using the data from 2006-2015, periodic characterization of the accuracy of different latitude models are analyzed. It turns out that the RMS of temperature, weighted mean temperature, pressure and vapor pressure shows certain periodic patterns, and differs with different latitude regions. In general, GPT2w model exhibits high precision and stability within the area of China.