Researchers at the Xinjiang Astronomical Observatory of the Chinese Academy of Sciences have developed a hybrid deep learning model that significantly enhances the accuracy of atmospheric delay prediction-a key factor affecting both astronomical observations and geodetic measurements.
Electromagnetic waves slow as they pass through the Earth's atmosphere due to variations in air density an by Riko Seibo
Tokyo, Japan (SPX) Oct 21, 2025
Researchers at the Xinjiang Astronomical Observatory of the Chinese Academy of Sciences have developed a hybrid deep learning model that significantly enhances the accuracy of atmospheric delay prediction-a key factor affecting both astronomical observations and geodetic measurements.
Electromagnetic waves slow as they pass through the Earth's atmosphere due to variations in air density and water vapor, producing what scientists call "tropospheric delay." This phenomenon introduces measurement errors in Very Long Baseline Interferometry (VLBI) and Global Navigation Satellite System (GNSS) positioning by bending and delaying signals as they travel through the atmosphere.
Using multi-year GNSS and meteorological data collected from the NanShan 26-meter Radio Telescope, the research team led by LI Mingshuai designed a hybrid model combining Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) neural networks. This deep learning approach enables automatic extraction of delay variation patterns from large datasets and delivers high-precision short-term forecasts of the Zenith Tropospheric Delay (ZTD).
Spectral analysis of long-term GNSS data revealed clear annual and semi-annual cycles in ZTD, with higher delays during summer and lower in winter-closely linked to temperature and humidity variations. By pairing GRU's ability to capture short-term dynamics with LSTM's capacity for modeling long-term trends, the hybrid network effectively simulates both transient fluctuations and seasonal patterns.
Test results show that the model achieves an average prediction error of just 8 millimeters and a correlation coefficient of 96 percent, surpassing the performance of traditional empirical or single-network methods.
These advances in atmospheric delay prediction promise substantial benefits for high-precision VLBI calibration, improved baseline and radio source positioning, and enhanced meteorological support for millimeter-wave astronomy. The approach also offers potential applications in water vapor retrieval and weather forecasting.
According to the observatory, this development lays critical groundwork for advanced atmospheric calibration techniques essential to the upcoming Qitai 110-meter Telescope and future multi-station interferometric projects.
Research Report:Enhanced Zenith Tropospheric Delay Forecasting Using a Hybrid GRU-LSTM Deep Learning Model
Related Links
Xinjiang Astronomical Observatory
Earth Observation News - Suppiliers, Technology and Application