A Machine Learning-Based Tropospheric Prediction Approach for High-Precision Real-Time GNSS Positioning

Area of Science:

• Geodesy and Geomatics
• Atmospheric Science
• Machine Learning Applications

Background:

• Tropospheric errors significantly impact high-precision Global Navigation Satellite System (GNSS) positioning, reaching meters.
• Current real-time troposphere correction services have limited regional availability.
• Existing post-mission data from the International GNSS Service (IGS) has a latency of 1-2 weeks, unsuitable for real-time use.

Purpose of the Study:

• To develop a real-time troposphere prediction method for GNSS applications.
• To overcome the latency limitations of current tropospheric correction data.
• To enable high-precision positioning by mitigating real-time tropospheric errors.

Main Methods:

• Developed a real-time troposphere prediction model utilizing machine learning techniques.
• Leveraged International GNSS Service (IGS) post-processing products as input data.
• Validated the prediction method using a year-long dataset.

Main Results:

• Achieved a root mean square error (RMSE) of 2 cm for tropospheric predictions.
• Demonstrated the suitability of the developed method for real-time applications.
• Successfully eliminated the long latency associated with traditional post-mission data.

Conclusions:

• The proposed machine learning-based method provides accurate, real-time troposphere predictions for GNSS.
• This advancement supports high-precision positioning applications by addressing a key error source.
• The method offers a viable alternative to existing services with limited coverage or high latency.