Objectives The global navigation satellite system‑acoustic (GNSS‑A) ranging combined technology extends the geodetic observation network to the ocean, becoming an important technical means for monitoring submarine geophysical phenomena. Currently, high-precision seafloor geodetic positioning typically relies on a reference sound speed profile (SSP), combined with acoustic ray-tracing methods to correct refraction errors, and direct inversion or parameterized compensation to mitigate the effects of spatiotemporal variations of sound speed. However, in-field SSP measurements not only increases the cost of seafloor geodetic observations, but also limits the timeliness of various seafloor geodetic monitoring efforts. Inverting the SSP based on GNSS‑A observation data is an effective approach to replace in-field SSP measurements.

Methods Based on the long-term GNSS‑A observation data from the MYGI station publicly released by Japan, we systematically compare the performance of the Munk model, bilinear model, and self-constructed empirical model in inverting SSP under different constraint conditions. The positioning solutions are then calculated using the ray-tracing model M1, which considers only the vertical variation of sound speed, and the sonar zenith delay model M2, which further incorporates parameters to compensate for the spatiotemporal variations of sound speed, based on the inverted SSP. Finally, using the in-field SSP and its corresponding positioning results as references, a comprehensive evaluation is conducted to assess the impact of different SSP models and constraint schemes on the accuracy of seafloor geodetic positioning.

Results The experimental results show that inverted SSP can effectively represent the overall variation trend of SSP, with more accurate characterization in deeper layers and relatively poorer characterization in shallower layers. The self-constructed empirical model under appropriate sea surface constraints and relaxed seafloor gradient constraints generally exhibits relatively higher inversion accuracy, with the root mean square (RMS) of differences compared to in-field measured SSP being approximately 5 m/s in shallow waters (in the depth range of 10-1 000 m) and about 1 m/s in deep waters (in the depth range of 1 000-1 727.8 m). When the positioning model only considers the vertical variation of sound speed, the self-constructed empirical model under different constraint conditions achieves the highest or comparable positioning accuracy compared to other SSP models, but high-precision positioning cannot yet be achieved. When the positioning model further accounts for the spatiotemporal variation of sound speed, the three-dimensional positioning accuracy is significantly improved. Except for the bilinear model under tight seafloor gradient constraints, the three SSP models under the remaining constraints achieve high-precision positioning. The self-constructed empirical model under appropriate sea surface constraints and relaxed seafloor gradient constraints achieves the relatively highest positioning accuracy, with RMS of differences compared to in-field SSP positioning results of 0.002 m, 0.002 m, and 0.022 m in the east, north, and up directions, respectively.

Conclusions Experimental results demonstrate that, to achieve high-precision positioning with inverted SSP, the influence of spatiotemporal variation of sound speed must be considered in the positioning model. It is recommended to adopt the self-constructed empirical model under appropriate sea surface constraints and relaxed seafloor gradient constraints, along with a two-level optimization method, to achieve seafloor geodetic positioning as an alternative to in-field SSP measurements. It should be noted that the proposed method still has certain limitations. On the one hand, the difficulty in obtaining prior information under special conditions, such as in shallow seas or polar regions, constrains inversion accuracy. On the other hand, the current algorithm relies on relatively accurate historical SSP initial values to ensure iterative convergence. Future research should explore how to integrate multi‑source ocean data to establish a more robust prior constraint system.