Sea Surface Temperature Prediction Enhanced by Exploring Spatiotemporal Correlation Based on LSTM and Gaussian...

Area of Science:

• Oceanography and Climate Science
• Artificial Intelligence in Environmental Modeling

Background:

• Accurate sea surface temperature (SST) prediction is critical for marine studies, climate dynamics, and environmental forecasting.
• General SST prediction models face challenges due to regional variations and complex climate phenomena.

Purpose of the Study:

• To develop an improved SST prediction model addressing spatial and temporal dependencies.
• To enhance the accuracy and reliability of sea surface temperature forecasts.

Main Methods:

• Proposed a hybrid framework combining Long Short-Term Memory (LSTM) networks with Gaussian processes.
• LSTM module captures temporal trends (long and short-term) in SST data.
• Gaussian process integrates spatial dependencies from neighboring data to refine predictions and estimate uncertainty.

Main Results:

• The hybrid framework demonstrated superior performance compared to state-of-the-art methods in SST prediction.
• Experiments conducted on the OISST dataset, focusing on the Bohai Sea and South China Sea, validated the model's effectiveness.
• The model successfully estimated prediction uncertainty, adding practical value.

Conclusions:

• The Gaussian process-enhanced LSTM network offers a robust solution for accurate SST prediction.
• The framework effectively models complex spatial and temporal dynamics in SST data.
• This approach provides reliable SST forecasts with valuable uncertainty quantification for marine and climate applications.