Optimal Window Size for Near Real-Time ZTD Estimation Using PPP in Thailand

This study investigates optimal processing parameters for near real-time Zenith Total Delay (ZTD) estimation using Ultra-Rapid products and a sliding window approach. GNSS data from 20 CORS stations in Thailand were processed with GipsyX over 90 days representing three seasons. Window sizes from 30 to 540 minutes were evaluated using 15-minute steps, and overlapping estimates were combined by weighted least squares averaging. Results show that RMSE decreases rapidly for short windows (30–100 minutes) and stabilizes at approximately 4-8 mm for window sizes exceeding 250 minutes, remaining below the 13 mm accuracy threshold for NWP nowcasting. Most stations achieve operational accuracy with window sizes of about 120 minutes, while the network-wide average optimal window size is 79 ± 20 minutes after excluding outliers. Observation data quality, particularly data gaps and cycle slips, significantly influences ZTD stability and required window length. Seasonal analysis indicates that increased atmospheric variability leads to higher ZTD variability and longer optimal windows. Overall, Ultra-Rapid products enable reliable near real-time ZTD estimation within 1–3 hours, offering a practical balance between timeliness, accuracy, and computational efficiency for GNSS meteorology in tropical regions.