Evaluation of machine learning methods in estimating precipitable water vapor Case Study: Hashtgerd GNSS station

This paper studies the application of three machine learning methods to estimate precipitable water vapor (PWV) using the observations of five GNSS stations of the National Geographical Organization (NGO) from November ۱۰ to December ۱۰, ۲۰۲۱. Since PWV is a key parameter in meteorological studies and forecasting atmospheric events such as rain and floods, estimating this parameter with high accuracy is very important. In the first step, the zenith tropospheric delay (ZTD) and zenith hydrostatic delay (ZHD) is calculated with the Gamit software and Saastamoinen model, respectively. By subtracting the ZHD from the ZTD, the zenith wet delay (ZWD) is obtained and then, the values of ZWD are converted to PWV. The obtained PWV values from this step are considered as the optimal output of all three models generalized regression neural networks (GRNN), support vector regression (SVR) and random forest (RF). Also, the input observations of all three models will be the latitude and longitude values of each GNSS station, day of the year (DOY) and time. After the training and achieving the minimum cost function value for all three models, the PWV value is estimated by the trained models and compared at the location of the test station. Hashtgerd GNSS station is considered as a test station. The average RMSE of the three models was ۱.۱۴, ۲.۱۵ and۱.۲۷ mm, respectively. The average correlation coefficient of GRNN, SVR and RF models was ۰.۹۵, ۰.۹۲ and ۰.۹۴, respectively.

کلیدواژه ها:

Hashtgerd GNSS station ، Precipitable Water Vapor ، GRNN ، SVR ، RF.