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

Microplastic pollution has become a global environmental problem, posing a potential threat to ecosystems and human health. Traditionally, microplastic monitoring has relied on spectral methods, which have significant limitations in terms of cost and time efficiency. To achieve low cost, rapid and large-scale detection, remote sensing technology has been applied to microplastic monitoring, but its accuracy needs to be improved. To address such issue, the Random forest-absorbed-gradient boosting model (RFAGB) is proposed to invert the microplastic abundance in the Bohai Sea based on remote sensing data. The proposed model has 23 % improvement in R and 67 % improvement in mean squared error (RMSE) compared with the single machine learning model. Our results shows that RFAGB has better robustness and accuracy. Armed with RFAGB, the microplastic distribution in the Bohai Sea was investigated. In terms of the three major bays of Bohai Sea, the average abundance in Laizhou Bay was the highest with the concentration of 1.06 ± 0.48 items m. Furthermore, there were two microplastic high-value zones within the central sea area. The main factors contributing to the spatial and temporal variations of microplastics may be the input of pollution sources and the role of hydrodynamics. This study reveals the great potential of satellite remote sensing technology in monitoring marine microplastics, which can be an important tool for regular monitoring of marine microplastics in the future.