Abstract: Based on satellite remote sensing sea surface temperature and sea surface height data and Argo data, this paper uses the Extreme Gradient Boosting (XGBoost) algorithm to study the inversion model of three-dimensional thermohaline at different depths. The inverted temperature and salinity information are consistent with the spatial and stratified distribution characteristics of the Argo data. The mean absolute error and root mean square error of global temperature (salinity) inversions at 58 layers are 0.319 °C (0.050) and 0.497 °C (0.077), respectively. Also, the thermohaline profiles of the South China Sea and the western Pacific Ocean are also in good agreement with the Argo data. Thereafter, the thermohaline information obtained from the XGBoost inversion model was applied to the sound velocity simulation. It can be seen that the XGBoost Inversion of thermohaline in this paper significantly improves the inversion efficiency on the premise of ensuring the accuracy. It can fully reflect the stratified structure of the ocean, and be applied to the simulation of sound velocity to provide technical support for marine environmental information assurance.