Parallel Prediction of Ocean Three-dimensional Fine Thermohaline Structure based on Surface Satellite Remote Sensing Data
Abstract
In this study, using sea surface temperature and sea surface height data, the extreme gradient boosting (XGBoost) parallel model was selected through multi-model comparison to predict the three-dimensional temperature and salinity information. The 58 layers of global temperature and salinity information were forecasted within 1 minute, and the mean absolute error (MAE) was 0.319℃ and 0.05psu, respectively. In particular, the prediction accuracy of the thermocline is poor, about 0.65°C, and the mid-deep layer is higher, about 0.3°C, which fully reflects the sensitivity of the model to the stratified structure of the ocean.