Analysis of Tropospheric Delay Bias in Real-Time PPP on Marine Buoy Platform and LSTM Modeling Prediction

Abstract Tropospheric delay is a key factor affecting the accuracy of Global Navigation Satellite System (GNSS) positioning and meteorological applications. However, in marine environments, the tropospheric delay derived from real-time Precise Point Positioning (PPP) often exhibits non-negligible biases. This study aims to improve the solution accuracy of tropospheric delay in marine environments and proposes a method for predicting Zenith Wet Delay (ZWD) based on the Long Short-Term Memory (LSTM) network. Using experimental data collected from buoy platform deployed in the South China Sea, the errors of positioning solution, satellite orbit and clock, as well as tropospheric delay were analysed. Long-term ERA5 reanalysis data were employed to construct and train the LSTM prediction model, incorporating temperature, vapor pressure, and atmospheric pressure profiles as inputs. Validation results demonstrate that the proposed model achieves ZWD prediction RMS of up to 0.010m, outperforming the real-time PPP solution with an RMS of 0.034m (RMS). The ZWD predicted by LSTM is suitable for integration into real-time PPP processing in marine environments to improve the accuracy of real-time ZWD estimation.