Application of machine learning in assessing spatial distribution patterns of soil microplastics: a case study of the Bang Pakong Watershed, Thailand

Microplastics (MPs) are persistent environmental pollutants introduced through various factors. Understanding the spatial distribution characteristics of MPs is crucial for developing effective mitigation strategies. However, the relationship between MPs and their covariables remains largely unexplored. In this study, we investigated the abundance of MPs in 40 soil samples from different land-use types in the Bang Pakong Watershed and assessed the applicability of machine learning models for predicting the spatial distribution patterns of soil MPs. The samples were divided into three categories: 1.2 µm-500 µm, 500 µm-1 mm, and 1-2 mm. They were analyzed using ZnCl2 density separation, H2O2 digestion process, and FTIR for polymer identification. Results showed a high presence of MPs, averaging 1121 ± 2465.6 items/kg of dry soil. The majority were MPs smaller than 0.5 mm (49%), fragments (74.2%), transparent-coloured particles (49%), and polypropylene (52%). According to the land-use, the highest MPs abundance was found in urban soil (67.6%) at 2331 ± 4114 items/kg, followed by irrigation soil at 555 ± 571 items/kg, agricultural soil at 552 ± 432 items/kg, and forest soil at 417 ± 365 items/kg. In model development, 14 variables were applied, including soil properties, physicochemical, and environmental variables. The Random Forest (RF) model performed better with higher prediction accuracy (R² = 0.85) and identified silt content and distance-to-river as significant influencing variables. The spatial distribution pattern of soil MPs was developed using the predicted MPs and the inverse distance weight (IDW) interpolation method, revealing a trend of increasing concentration south-westward towards the Bang Pakong River. Susceptibility analysis of flood and drainage density correlated with the interpolation result, suggesting these factors could influence the transportation and deposition pathways of MPs. This study provides insights into the abundance and distribution of MPs, their relationship with covariables, and the potential of machine learning in pollution management.