Long-sequence deep learning frameworks for multivariate forecasting of tropospheric parameters Mert Bezcioglu GPS Solutions, 2026 This contribution presents the first multivariate deep learning frameworks for jointly forecasting four tropospheric parameters using nine long-sequence architectures representing Transformer, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Multi-Layer Perceptron (MLP) model families. Models were trained on seven years of hourly observations from 501 globally distributed stations with a 96-hour input and a 24-hour forecast horizon. The results demonstrate that the multivariate formulation consistently outperforms univariate forecasting, reducing 24-hour Root Mean Square Error (RMSE) by approximately 7% for both Zenith Tropospheric Delay (ZTD) and Zenith Wet Delay (ZWD) and providing typical full-horizon errors of 19.3 to 20.9 mm for ZTD and ZWD, 4.1 to 5.6 mm for Zenith Hydrostatic Delay (ZHD), and 3.1 to 3.3 mm for Precipitable Water Vapor (PWV). Moreover, the outcomes highlight that ZTD and ZWD can be forecasted at the sub-cm level in the 1–3 h time range. A central contribution of the study is the evaluation of physical consistency, which shows that the forecasted parameters preserve core atmospheric relationships, including the PWV/ZWD ratio and the short-term coupling between ZTD and ZWD, with violation rates below 0.01%. Although a few Transformer-based models show minor inconsistencies in the ZTD, ZHD, and ZWD closure, some architectures sustain high forecast accuracy with closure deviations constrained to 0.2 mm. These findings demonstrate the substantial benefit of multivariate deep learning for forecasting tropospheric parameters and highlight the need for future approaches that integrate explicit physical constraints to further enhance numerical stability and physical realism.
Evaluation of real-time PPP techniques for seismic and structural monitoring Mert Bezcioglu, Emre Bozdogan, Ahmet Anil Dindar, Cemal Ozer Yigit Survey Review, 2026 This contribution evaluates ability of Real-Time Precise Point Positioning (RT-PPP), Near RT-PPP, and Post-Processed (PP) RT-PPP to capture dynamic motions. Harmonic motions with different frequencies and amplitudes were generated on a shake table, along with ground-motion simulations of two major earthquakes and step functions. In both frequency and time domains, RT-PPP results from 20 Hz GPS observations were compared with Linear Variable Differential Transformer (LVDT) and PPP-Final references. Findings show that RT-PPP can successfully identify structural natural frequencies, seismic waveforms, and static or quasi-static displacements at the moment an event occurs in the field, immediately afterward, or one day later.
Evaluation of the single-frequency variometric approach based on low-cost GNSS observations and different satellite combinations for detecting short-term dynamic behaviors Berkay Bahadur, Mert Bezcioglu, Cemal Ozer Yigit Measurement Science and Technology, 2024 This study presents the capability of the single-frequency (SF) variometric approach (VA) technique with low-cost GNSS observations to detect short-term dynamic behaviors. Harmonic oscillations with amplitudes between 5 and 20 mm and frequencies between 0.3 and 5.0 Hz were generated employing a single-axis shake table to investigate the performance of the SF-VA technique in the structural health monitoring (SHM) system. Besides, a Mw 6.9 Kobe, 1995 earthquake simulation was generated using the shake table to analyze the SF-VA performance for the earthquake early warning (EEW) system. A low-cost u-blox ZED-F9P GNSS receiver and ANN-MB-00 patch antenna were used to collect GNSS observations at a 20 Hz sampling rate during the experiments. The observations were processed using the MATLAB-based open-source PPPH-VA software in real-time (RT) mode, considering eight different satellite combinations. The capability of the SF-VA technique to detect horizontal dynamic behaviors in RT mode was investigated in the frequency and time domains, accepting the displacements from the linear variable differential transformer sensor as a reference. The results in the frequency domain demonstrate that the SF-VA technique with low-cost GNSS observations can successfully detect the peak frequency value of short-term harmonic oscillations up to 5 Hz. Moreover, time domain findings emphasize that the short-time dynamic oscillations can be determined with the SF-VA technique with an accuracy ranging from 0.8 to 6.4 mm. Earthquake simulation experiment results demonstrate that the strong ground motions caused by mega earthquakes can be determined at mm-level by the SF-VA method. The results of both experiments show that multi-GNSS observations contribute to the SF-VA technique considerably. Overall, the findings reveal that the SHM and EEW systems can be operated with low-cost GNSS receivers, and the natural frequency of the man-made structures and accurate displacement values of seismic waveforms can be determined in RT with the SF-VA technique.
High-rate Single-Frequency Precise Point Positioning (SF-PPP) in the detection of structural displacements and ground motions Bezcioglu, Mert, Yigit, Cemal, Dindar, Ahmet, El-Mowafy, Ahmed, Wang, Kan Structural Engineering and Mechanics, 2024 This study presents the usability of the high-rate single-frequency Precise Point Positioning (SF-PPP) technique based on 20 Hz Global Positioning Systems (GPS)-only observations in detecting dynamic motions. SF-PPP solutions were obtained from post-mission and real-time GNSS corrections. These include the International GNSS Service (IGS)-Final, IGS real-time (RT), real-time MADOCA (Multi-GNSS Advanced Demonstration tool for Orbit and Clock Analysis), and real-time products from the Australian/New Zealand satellite-based augmentation systems (SBAS, known as SouthPAN). SF-PPP results were compared with LVDT (Linear Variable Differential Transformer) sensor and single-frequency relative positioning (SF-RP) solutions. The findings show that the SF-PPP technique successfully detects the harmonic motions, and the real-time products-based PPP solutions were as accurate as the final post-mission products. In the frequency domain, all GNSS-based methods evaluated in this contribution correctly detect the dominant frequency of short-term harmonic oscillations, while the differences in the amplitude values corresponding to the peak frequency do not exceed 1.1 mm. However, evaluations in the time domain show that SF-PPP needs high-pass filtering to detect accurate displacement since SF-PPP solutions include trends and low-frequency fluctuations, mainly due to atmospheric effects. Findings obtained in the time domain indicate that final, real-time, and MADOCA-based PPP results capture short-term dynamic behaviors with an accuracy ranging from 3.4 mm to 8.5 mm, and SBAS-based PPP solutions have several times higher RMSE values compared to other methods. However, after high-pass filtering, the accuracies obtained from PPP methods decreased to a few mm. The outcomes demonstrate the potential of the high-rate SF-PPP method to reliably monitor structural and earthquake-induced ground motions and vibration frequencies of structures.
