Cross-sensor spectral fusion for coastal water physicochemical assessment using optimized ensemble machine learning...

Area of Science:

• Environmental Science
• Remote Sensing
• Machine Learning

Background:

• Coastal pollution necessitates advanced monitoring techniques.
• Existing remote sensing-machine learning models struggle with complex coastal water optics due to limited spectral features and conventional models.

Purpose of the Study:

• To enhance random forest (RF) modeling for coastal water quality index (WQI) assessment.
• To integrate teaching-learning-based optimization (TLBO) and student-psychology-based optimization (SPBO) with diverse spectral features for improved accuracy.

Main Methods:

• Calculated WQI using parameters like chemical oxygen demand and dissolved oxygen.
• Utilized Landsat 8 and Sentinel 2 satellite data as input features.
• Applied and compared RF-TLBO and RF-SPBO models with various spectral features.

Main Results:

• Over 44% of observed WQI indicated poor to severe pollution in key Chinese coastal areas.
• Multisource satellite data and comprehensive spectral features improved model sensitivity and accuracy over individual bands or indices.
• RF-TLBO slightly outperformed RF-SPBO in accuracy metrics (MAE, MSE, R², Willmott index) with low bias and uncertainty.

Conclusions:

• The enhanced RF-TLBO framework effectively improves coastal water quality monitoring and identification of high-risk pollution areas.
• Multisource satellite data integration is crucial for accurate coastal water quality assessment.
• Future research should explore multi-target regression for interdependency analysis of water quality parameters.