Application of LSTM-DT Fusion Algorithm in Predicting Ocean Temperature and Salinity Fields

With the swift progress in ocean monitoring technology, the ease of observing sea surface information contrasts with the paucity of subsurface data, which constrains the profound exploration of ocean dynamics and impedes the comprehensive use of oceanic information. To address this issue, this paper introduces a new method that combines satellite remote sensing, reanalysis, and actual measurement data to facilitate the three-dimensional reconstruction of ocean subsurface hydrological profiles. The study utilized 36 independent variables such as sea surface temperature, sea surface salinity, and evaporation to conduct predictive analysis of the temperature and salinity fields beneath the surface of certain areas of the northwest Pacific Ocean. A prediction using the LSTM (Long Short-Term Memory) model for temperature and salinity was proposed. Given that the model's performance is sensitive to hyperparameters, it was enhanced with the Crested Porcupine Optimizer to improve the forecasting of temperature and salinity, significantly reducing the RMSE (Root-Mean-Square Errors) . Following this, an innovative method, the LDFA (LSTM-DT Fusion Algorithm) model was introduced, which, compared to the LSTM and DT (Decision Tree) algorithms, markedly increased prediction accuracy, bringing the mean RMSE for predicted temperature and salinity down to 1.193 and 0.123, respectively.