Simulating carbon and nitrogen cycling in microplastic contaminated agroecosystems using gradient boost regression model

Microplastics (MPs) pollution poses significant threats to carbon (C) and nitrogen (N) cycling, affecting plant biomass. These threats are often difficult to measure due to the multifaceted effects of MPs. Hence, machine learning models offer a promising approach to effectively estimate nutrient dynamics. Here, we employed a Gradient Boosting Regression (GBR) model to estimate soil C and N contents, greenhouse-gas emissions, and plant biomass in MPs-contaminated soils. We also evaluated the mediation of soil type and incubation duration. To train and test the GBR model, we used data compiled from 55 peer-reviewed publications. The model results showed strong agreement between observed and predicted values for dissolved organic carbon (DOC). Training and testing R2 values differed by 3% for NH4+ and 2% for NO3−. For CO2 and N2O emissions, low mean squared error values indicated strong model performance in predicting emissions under MPs pollution. The model demonstrated strong predictive performance for biomass, achieving high R2 values for both the training and testing datasets. Across properties, MPs size caused the greatest changes in DOC and CO2 emissions. However, the MP shape had the greatest impact on the SOM content (60%). The greatest changes in soil NH4+ (36%) and NO3− (51%) content and plant biomass (77%) were caused by MP type. For N2O emissions, the size, type, dose, and incubation period of MPs had substantial effects. Our results conclude that the GBR model is a powerful tool to estimate the effects of MPs pollution on nutrient cycling and plant biomass.