Predicting microplastic impacts on carbon and nitrogen cycling in agroecosystems using Gradient Boost Regression (GBR) modeling

Microplastics can seriously disrupt soil carbon (C) and nitrogen (N) biogeochemical cycling in agroecosystems, strongly influencing crop growth and quality. However, our mechanistic understanding of how microplastics affect these cycles remains poorly understood. Specifically, direct in situ measurements in greenhouse and field settings are often impractical. However, lab-based and ex-situ measurements can train machine learning models to develop mechanistic understanding of microplastic behaviour. Thus, we aim to train and test the gradient boost regression (GBR) model to estimate the effects of key microplastic properties on C and N cycling and subsequently on plant biomass. The role of soil type in controlling the microplastics effects was also estimated. During prediction, datasets from published experiments (n = 52) were divided into a ratio of 80:20 for training and testing the model. GBR prediction showed R2 values ranged between 57% to 99% and MSE values ranged between 0 to 0.09 for the contents of dissolved organic carbon (DOC), soil organic carbon (SOC), soil organic matter (SOM), ammonium (NH4+), and nitrate (NO3-), emissions of CO2 and N2O. Overall, there was distinct effects of microplastic properties on soil C pools. Microplastic size contributed 34% in altering DOC while the maximum CO2 emissions (39%) were altered by microplastics incubation period in soil. However, microplastic shape contributed 47 to 60% to SOC and SOM. Microplastic size strongly altered NH4+ and NO3- by 36 to 51% in microplastic polluted soils resulting in the highest N2O emission. Plant biomass was strongly (76%) affected by microplastic types. Our results conclude that GBR model appeared a powerful machine learning tool for predicting the impacts on C and N cycling as well as plant performance following microplastic pollution. As microplastic pollution is increasing in soils globally, there is an urgent need to implement and strengthen such tools in modern research to tailor sustainable solutions.