RFAGB model: A new machine learning model for microplastic inversion based on remotely sensed data in Bohai Sea.

Area of Science:

• Environmental Science
• Remote Sensing Technology
• Marine Pollution Monitoring

Background:

• Microplastic pollution is a significant global environmental issue with potential risks to ecosystems and human health.
• Traditional spectral methods for microplastic monitoring are costly and time-consuming.
• Remote sensing offers potential for cost-effective, large-scale microplastic detection but requires improved accuracy.

Purpose of the Study:

• To develop an accurate and efficient method for microplastic abundance inversion using remote sensing data.
• To investigate the spatial distribution of microplastics in the Bohai Sea.
• To identify key factors influencing microplastic variations.

Main Methods:

• Development and application of a novel Random Forest-absorbed-gradient boosting model (RFAGB).
• Utilizing remote sensing data for microplastic abundance inversion.
• Analyzing microplastic distribution and identifying high-value zones in the Bohai Sea.

Main Results:

• The RFAGB model demonstrated significant improvements in accuracy (23% R-squared, 67% RMSE reduction) compared to single machine learning models.
• Laizhou Bay exhibited the highest average microplastic abundance (1.06 ± 0.48 items m⁻³).
• Two distinct microplastic hotspots were identified in the central Bohai Sea area.

Conclusions:

• The RFAGB model offers enhanced robustness and accuracy for remote sensing-based microplastic monitoring.
• Satellite remote sensing technology holds substantial promise for regular and large-scale marine microplastic surveillance.
• Pollution source input and hydrodynamic factors are likely drivers of microplastic spatial and temporal variations.