Multi-parameter prediction of seawater quality based on dynamic spatio-temporal relationship network.

Area of Science:

• Marine environmental monitoring
• Environmental data science
• Oceanography

Background:

• Seawater quality parameters are crucial indicators of marine ecosystem health.
• Traditional models struggle with multi-buoy spatio-temporal data, limiting prediction accuracy.
• Accurate prediction is vital for sustainable marine development.

Purpose of the Study:

• To develop an advanced prediction model for multi-parameter seawater quality.
• To effectively capture and utilize unstructured spatio-temporal relationships from multiple buoys.
• To enhance the accuracy and reliability of marine environmental monitoring.

Main Methods:

• Proposed a Dynamic Spatio-temporal Relationship Network (DSTRN) model.
• Utilized graph neural network concepts with an adaptive adjacency matrix.
• Implemented a multi-buoy collaborative prediction strategy.

Main Results:

• The DSTRN model demonstrated superior performance over single-buoy strategies.
• Achieved the best prediction accuracy compared to other advanced models.
• Exhibited robust performance across various prediction time horizons, with high R² values (e.g., 0.978 for buoy 14).

Conclusions:

• The DSTRN model effectively captures and utilizes spatio-temporal correlation information.
• The model shows strong spatio-temporal collaborative predictive capabilities for seawater quality.
• Confirms the effectiveness of multi-buoy collaborative prediction for marine health monitoring.