Recently, the Nearshore Environmental Safety Laboratory at Tianjin University’s School of Marine Science and Technology has made a series of advancements in the field of intelligent monitoring of microplastics. These findings have been published in top international environmental journals, including *SCIENCE OF THE TOTAL ENVIRONMENT* (JCR Q1, IF2025 = 8.0) and *WATER RESEARCH* (JCR Q1, IF2025 = 12.4).

Microplastics (MPs), as a new pollutant, pose significant threats to marine ecosystems and human health. The discontinuity, complexity, and high costs associated with marine field sampling data make it difficult to meet the needs of microplastics research and pollution control. Therefore, there is an urgent need to develop rapid monitoring technologies for efficient data collection on MPs.

Remote sensing technology has great potential for monitoring microplastics due to its advantages, such as the ability to conduct large-scale monitoring, continuous observations at the same location, extensive coverage through multispectral data, and no geographical limitations. To this end, a multi-regression inversion model was established by integrating remote sensing technology and ground-truth data to monitor the microplastic pollution status in the Bohai Sea.

This study proposed three variable selection methods: Sequential Projection Algorithm (SPA), Band Combination Method, and Remote Sensing Index Method. By comparing accuracy evaluation indicators, the SPA-based method was selected to analyze the abundance and spatiotemporal distribution of microplastics in the Bohai Sea for the year 2022. The coefficient of determination for the SPA model was found to be 0.75, and the root mean square error was 0.38 pieces/m³, indicating that the model error is within an acceptable range. The research demonstrates the significant potential of satellite remote sensing technology in monitoring multi-point source pollution in marine environments.

The statistical regression model developed in the above research exhibited shortcomings in precision and significant errors when inferring the presence of microplastics. To address this issue, we further developed a new inversion model based on satellite remote sensing technology and ensemble learning algorithms—namely, the Random Forest Absorption Gradient Boosting Model (RFAGB).

This model effectively captures complex nonlinear relationships in remote sensing data by integrating the strengths of two heterogeneous base learners: Random Forest (RF) and Gradient Boosting (GB), while also incorporating feature engineering strategies such as interaction term construction and binning. Through five-fold cross-validation, meta-features were generated, which were then adaptively fused through a meta-learner, significantly enhancing the model performance. Compared to the single Random Forest model, the RFAGB model improved the coefficient of determination (R²) on the test set by 23% (reaching 0.8667) and reduced the root mean square error (RMSE) by 67% (down to 0.3457), demonstrating excellent robustness.

Using this model, we inverted the surface microplastic abundance in the Bohai Sea from 2014 to 2023. The results indicated significant spatiotemporal heterogeneity in the abundance of microplastics: higher concentrations were observed in coastal and estuarine areas, with the highest average abundance recorded in Laizhou Bay (1.06 ± 0.48 pieces/m³) and two distinct high-value zones present in the central sea area. Seasonal variations in microplastic abundance were also evident, with higher levels observed in summer (1.65 ± 0.43 pieces/m³) and lower levels in winter (0.68 ± 0.32 pieces/m³). Further analysis revealed that pollution sources, particularly the runoff from the Yellow River and Liao River, along with marine dynamic processes, are key factors influencing the distribution of microplastics in the Bohai Sea.

This study confirms the immense potential of combining satellite remote sensing with ensemble learning models for monitoring marine microplastics. The proposed RFAGB model offers reliable technical support for large-scale, routine monitoring and assessment of microplastic pollution in marine environments.

Reference:

1. Pingping Hong, Jingen Xiao, Hongtao Liu, Zhiguang Niu, Yini Ma, Qing Wang, Dianjun Zhang*, Yongzheng Ma*. (2024). An inversion model of microplastics abundance based on satellite remote sensing: a case study in the Bohai Sea. Science of The Total Environment. 2024, 909, 168537.

//doi.org/10.1016/j.scitotenv.2023.168537

2. Ao Shen, Yongzheng Ma*, Yuan Li, Pingping Hong, Zhiguang Niu, Ying Zhang, Jingen Xiao, Dianjun Zhang*. RFAGB model: A new machine learning model for microplastic inversion based on remotely sensed data in Bohai Sea, Water Research, 2025, 043, 124490

//doi.org/10.1016/j.watres.2025.124490