The Hong Kong-Zhuhai-Macao Bridge (HZMB), spanning 55 kilometers, stands as a mega-project situated on the southeastern coast of China in the Pearl River Estuary. It plays a pivotal role in connecting three major cities in the Greater Bay Area of China: Hong Kong, Zhuhai, and Macao. As the world's longest sea-crossing bridge, it is crucially important to monitor the deformation and displacement of bridge in all-weather and all-time mode for safety and longevity. GNSS technique with real-time monitoring capability becomes an ideal tool for the deformation monitoring of the HZMB. The present study contributes to this research trend by constructing a suitable GNSS positioning model for long-span sea-crossing bridge deformation monitoring, and by further improving the GNSS positioning performance by multipath mitigation. An ambiguity-fixed ionosphere-weighted PPP-RTK model is constructed to provide real-time high-accuracy positioning service for long-span bridge deformation monitoring. In additional, a multipath model is developed for GPS and BDS in the spatial domain to mitigate multipath effects, addressing challenges posed by large reflection sources such as the sea surface and bridge tower structures. That is the elevation and azimuth of each satellite is recorded so that when the satellite appears in the same location, multipath can be significantly reduced by the multipath correction. On this basis, a continuously operating reference stations (CORS) network and three monitoring stations on the HZMB are used to collect GNSS data for 12 days (day of year (DOY) in 2022: 175-186) for conducting a GNSS deformation monitoring test. The results show that PPP-RTK can be well used for HZMB deformation monitoring. The multi-constellation multipath model can significantly improve performance. With the multipath mitigation method, PPP-RTK has obviously gained in the convergence time to reliably fix the ambiguities, as well as in the positioning accuracy and normal distribution characteristics of the observation residuals. Our study presents five consecutive days of bridge deformation monitoring results for the HZMB, demonstrating the high positioning accuracy and fast convergence time achieved with PPP-RTK and multipath mitigation. Moreover, our model is capable of identifying the high-frequency vibration and low-frequency displacement of the bridge. It is worth noting that during the experiment, Typhoon Chaba made landfall in the experimental area (during DOY: 182-184). The natural frequency of 0.373 Hz in the vertical direction of the bridge under the typhoon Chaba excitation has been successfully detected. This research offers valuable insights into long-span bridge deformation monitoring and demonstrates the effectiveness of GNSS technology for real-time, high-precision positioning in infrastructure health monitoring.