Trop_XS: an intelligent architecture for real-time wide-area ZTD corrections modeling driven by heterogeneous observations

Modeling real-time zenith total delay (ZTD) remains a challenge owing to the rapidly changing water vapor conditions influenced by geography and climate. In this contribution, we propose Trop_XS, an innovative architecture that integrates the extreme gradient boosting (XGBoost) and spherical cap harmonic analysis (SCHA) algorithms with multi-source heterogeneous data, to support real-time wide-area tropospheric augmentation services. XGBoost enables rapid fusion of epoch-wise features, generating ZTD corrections at predefined grid points with a mean RMS below 11 mm. Building on the empirical ZTD model, the enhanced SCHA-E solution is employed to model the ZTD corrections from reference sites and regular grids, achieving a favorable trade-off between accuracy and communication burden. Validated against 6-month post-processing ZTD from 31 external test stations, Trop_XS-derived ZTD achieves a mean RMS of 13.4 mm, outperforming MOFC, VMF3_SCHA, and GNSS_interp solutions by 25.3%, 10.5%, and 11.4%, respectively. Furthermore, the proposed model demonstrates superior spatiotemporal applicability in areas characterized by complex terrain and limited station density, as well as during periods with active water vapor. Statistical results of the real-time precise point positioning (PPP) demonstrate that incorporating Trop_XS ZTD constraints effectively weakens the vertical error and accelerates initialization, shortening vertical convergence times by 36.4% in winter and 32.3% in summer, respectively. Given its scalability, the Trop_XS solution holds potential in other challenging regions, enhancing GNSS satellite-based PPP services and numerical weather forecasting capabilities.