Document Type : Research Article
Authors
1
Department of Geomatics Engineering, Faculty of Geodesy & Geomatics Engineering, K. N. Toosi University of Technology, Tehran, Iran
2
khaje nasir toosi university of technology
3
Department of Geoscience Engineering, Arak University of Technology, Arak, Iran
Abstract
Precipitable water vapor (PWV) is one of the most important parameters in meteorological studies, which can be calculated using ground measurements such as radiosondes and radiometers, as well as indirect methods such as GPS observations. Due to the spatial limitations of meteorological stations, as well as observational discontinuities in the time domain, PWV modeling with new methods is of great importance. Various models have been developed to estimate the amount of water vapor (WV) and tropospheric refraction component. These models can be divided into two categories: empirical and analytical models. Among the analytical models, we can mention to the tomography method. Considering the spatial limitation of GPS and meteorological stations, as well as the discontinuity of observations in the time domain, modeling of PWV is very important. In order to overcome these limitations in PWV forecasting, the idea of using machine learning (ML) models has been proposed. In the first step, the zenith tropospheric delay (ZTD) is calculated with Gamit software. Then using the Saastamoinen empirical model as revised by Davis the value of zenith hydrostatic delay (ZHD) is calculated. By subtracting the amount of ZHD from ZTD, zenith wet delay (ZWD) is obtained at each GPS stations. Using the formula by Bavis, ZWD cab be converted to PWV for any time. In this paper for the first time in Iran, the amount of PWV has been modeled and forecasted using the long short-term memory (LSTM) method. LSTM is one of the recurrent neural networks which is mostly used for time series purposes. The observations of Tehran GPS network in the 2021 and 2022 for the periods of 312 to 347 and 33 to 78 DOY have been used to evaluate the performance of the LSTM. It should be noted that the parameters of longitude, latitude and altitude, as well as day of year (DOY) and time (hour) are considered as inputs to LSTM model to estimate PWV. The results of this model are compared with the results of GRNN which is type of radial basis function (RBF), GPT3 and Saastamoinen empirical models. Statistical indicators of root mean square error (RMSE) and correlation coefficient are used to check the accuracy of the models in control station. The RMSE for the control station of LSTM, GRNN, GPT3 and Saastamoinen models in 2021 is 0.5, 1.34, 7.12 and 7.65 mm, respectively. Also, the RMSE of three models for the control station in 2022 is 0.9, 1.19, 3.32 and 3.28 mm, respectively. The results show that the LSTM model has high accuracy to forecast the amount of PWV compared to GRNN and empirical models. As a result, the new model presented in this paper can be used as an alternative to empirical models in forecasting precipitable water vapor.
Keywords
Main Subjects
)