Machine learning-driven analysis of soil microplastic distribution in the Bang Pakong Watershed, Thailand

Microplastics (MPs) have emerged as a pervasive environmental pollutant due to their persistence and global distribution. However, MPs relationships with covariables remain largely unexplored. This study investigates factors influencing MPs occurrence and distribution in the Bang Pakong Watershed, using 40 soil samples across various land-use types and assess machine learning for their spatial distribution. Samples were sorted into three sizes: 1.2 μm–500 μm, 500 μm–1 mm, and 1–2 mm and analyzed using zinc chloride (ZnCl₂) density separation, hydrogen peroxide (H₂O₂) digestion, and Fourier transform infrared spectroscopy (FTIR) for polymer identification. Results reveal a significant MPs presence, averaging 1121 ± 2465.6 items/kg dry soil, with particles <0.5 mm (49 %), fragments (74.2 %), transparent (49 %), and polypropylene (PP) (52 %) predominating. Urban soils contained highest concentrations (67.6 %) at 2331 ± 4114 items/kg, followed by irrigation (555 ± 571), agricultural (552 ± 432), and forest soils (417 ± 365). Predictive modeling incorporated 14 variables, including soil properties and environmental factors. The Random Forest model (RF), optimized for complex non-linear relationships and high data variability, shows higher predictive accuracy (R² = 0.82), with silt content and distance-to-river as key variables. Spatial distribution analysis, developed on model predictions and inverse distance weighting (IDW), demonstrates a concentration gradient increasing southwestward toward the Bang Pakong River. Flood susceptibility and drainage density analysis correlate with interpolation results, suggesting that these factors influence MPs transport and deposition processes. These results refine MPs management, emphasizing urbanization and hydrological factors as drivers for distribution, necessitating targeted mitigation in high-risk areas.