Investigating the Efficiency of Deep Learning Methods in Estimating GPS Geodetic Velocity

Abstract

Observations and processing of Iran's national geodynamic network, global positioning systems (GPSs) track and record signals received from GPS satellites 24 hr a day. The information collected at each station is received in the form of observation and navigation files, meteorological data and tilt meters in the main processing centers, using the telephone line and modem. These centers are located in Tehran and the surveying managements of the surveying organization in the cities of Mashhad, Tabriz, Hamedan and Ahvaz, which receive this information daily. In these data centers, after performing quality control and format conversion, the received data are ready to enter the GPS processing software. Processing in these centers is done on a daily and final basis, the results of which are presented in the form of stations exact coordinates in the international terrestrial reference frame (ITRF) coordinate system, coordinate time series and velocity vectors along with the error ellipses of each station. The results of processing global geodynamic networks are in the form of each station time series, displacement and velocity vectors, and exact ITRF coordinate system and their conversion in the world geodetic system 1984 (WGS84) coordinate system along with their accuracy. These results, as well as observation and navigation files, are stored in data centers (Djamour et al., 2010).Artificial neural network (ANN) has been successfully used in various fields of geodesy. Coordinate transformation with ANN was found to be a practical tool, effective and feasible (Cakir and Konakoglu. 2019). Geoid undulation determination with ANN has root mean square error of 10.81 cm (Konakoglu et al., 2022). Ionospheric variations with ANN has an R value greater than 0.74 (Inyurt and Sekertekin. 2019).Estimation of geodetic velocity is of special importance for geodynamic studies especially fault displacement. GV is determined based on GPS and global navigation satellite system (GNSS) observations at different time intervals and can be used to measure tectonic movements (Sorkhabi, Alizadeh, et al., 2022). GNSS stations, due to their high cost and difficulty in measuring and maintaining them, need methods that can accurately estimate GV in gap areas. Memarin Sorkhabi and Djamour (2015) in northwestern Iran estimated geodetic velocity (GV) with an ANN reaching root mean square error (RMSE) less than 3.5 mm. In 2021, Konakoglu estimated the GV with the multi-layer perceptron neural network method, which was more accurate than classical mathematical methods. MariaMrówczyńska et al. (2020) used ANN and principal component analysis (PCA) to estimate GV, which performs both compression and reconstruction methods with high accuracy. Back propagation artificial neural network with 42 GPS stations has an RMSE of less than 2 mm in determining the GV of northwestern Iran (Sorkhabi, Alizadeh, et al., 2022).Various deep learning methods have been developed that can be used according to the type of problem. The comparison of this research with the pr...

Pca-Som of Grace-Fo Total Water Storage for Global Climate Decisions

“…Due to a large amount of data and the complexity of the results, it is necessary to use a tool such as PCA-SOM to simplify the output to make the correct interpretation. SOM is a kind of artificial neural network that has found many applications in recent years in the fields of engineering to medicine, biology and economics (Acevedo-Acosta et al, 2021;Sorkhabi et al, 2022b;Sorkhabi & Milani, UDC 528.8…”