GPT2模型的精度检验与分析

Accuracy Assessment and Analysis for GPT2

GPT模型常被用于计算气温、气压等对流层延迟气象参数,针对其不足之处,Lagler提出了改进的全球经验模型GPT2,该模型不仅提高了GPT气温和气压模型的精度,而且可提供比湿、水汽压、映射函数等对流层参数。但是目前没有相关文献对GPT2的精度进行详尽的分析,本文利用ECWMF及NOAA提供的高精度气象数据,对GPT2气温、气压和水汽压模型进行精度检验及分析。结果表明,气温的Bias均值为-0.59°C,RMS均值为3.82°C左右;气压和水汽压的Bias均值绝对值在1mb以内,气压的RMS均值为7mb左右,水汽压则不超过3mb,不同纬度精度存在差异,三者均具有明显的季节性。总体而言,GPT2模型在全球范围内具有很高的精度和稳定性。

GPT（global pressure and temperature） is a global empirical model usually used to provide temperature and pressure for the determination of tropospheric delay, there are some weakness to GPT, these have been improved with a new empirical model named GPT2, which not only improves the accuracy of temperature and pressure, but also provides specific humidity, water vapor pressure, mapping function coefficients and other tropospheric parameters, and no accuracy analysis of GPT2 has been made until now. In this paper high-precision meteorological data from ECWMF and NOAA were used to test and analyze the accuracy of temperature, pressure and water vapor pressure expressed by GPT2, testing results show that the mean Bias of temperature is -0.59℃, average RMS is 3.82℃; absolute value of average Bias of pressure and water vapor pressure are less than 1 mb, GPT2 pressure has average RMS of 7 mb, and water vapor pressure no more than 3 mb, accuracy is different in different latitudes, all of them have obvious seasonality. In conclusion, GPT2 model has high accuracy and stability on global scale.