Short-Term Forecast of Tropospheric Zenith Wet Delay Based on TimesNet

Area of Science:

• Geodesy and Atmospheric Science
• Meteorological Forecasting
• Machine Learning Applications in Geosciences

Background:

• Tropospheric zenith wet delay (ZWD) is vital for atmospheric water vapor inversion and short-term rainfall prediction.
• Traditional ZWD prediction models face limitations in handling non-stationarity and local feature extraction.
• Accurate ZWD prediction is essential for enhancing meteorological forecasting accuracy.

Purpose of the Study:

• To develop and evaluate a novel ZWD prediction method using TimesNet's dynamic temporal decomposition module.
• To improve the accuracy and adaptability of ZWD predictions by reconstructing time series data into tensors.
• To assess the performance of TimesNet against conventional models (CNN-ATT, Informer) considering various environmental factors.

Main Methods:

• Reconstruction of 1D ZWD time series into 2D tensors using TimesNet.
• Integration of topographical, climatic, and seasonal factors into the prediction model.
• Comparative analysis of TimesNet with CNN-ATT and Informer using 30-second ZWD data from 20 IGS stations across different seasons.

Main Results:

• TimesNet achieved an average seasonal Root Mean Square Error (RMSE) of 5.73 mm, outperforming Informer (7.89 mm) and CNN-ATT (10.02 mm).
• The model demonstrated superior seasonal adaptability and topographical robustness, maintaining high accuracy in challenging environments.
• Sub-5 mm precision was achieved in stable meteorological conditions, highlighting TimesNet's reliability.

Conclusions:

• TimesNet offers a reliable algorithmic foundation for real-time meteorological applications, particularly short-term precipitation forecasting.
• The dynamic temporal decomposition approach effectively addresses the non-stationarity of ZWD data.
• This method enhances the potential of Global Navigation Satellite System (GNSS) meteorology for improved weather prediction.