Towards a fast and generalized microplastic quantification method in soil using terahertz spectroscopy

Abstract

Extensive investigation of microplastic abundance in soil environment calls for rapid, accurate, efficient and harmonized quantification methods. Development of rapid quantification method requires made-to-measure soil samples with additions of standard polymers. Existing rapid quantification methods ignore the gap between standard polymers in laboratory and household microplastics in soil environment. Here, terahertz (0.6–1.67 thz) and NIR (950–1660 nm) spectroscopy were compared to explore a fast, accurate and potentially generalizable microplastic quantification method in soil. Soil sample was spiked with two standard polymers (polyvinyl chloride (PVC) and polystyrene (PS)) and their additive-containing household microplastics. Two standard sample sets and two household sample sets were prepared in concentrations ranging from 0.5 to 10%. Nine commonly used preprocessing methods and three machine learning algorithms were coupled to develop methods. Models were constructed by training sets from standard sample sets. When models transferred to household samples, prediction error (RMSE) of proposed terahertz method (Wdenosie_PLSR) only increased by 0.4% for PVC and 0.19% for PS, yet that of the NIR method increased by 1.49% and 1.16% respectively. The proposed terahertz method presented a detection limit around 1.12% and the NIR method showed a detection limit around 3.24%. Overall, our results suggest that compared with NIR method, the proposed terahertz method is not only more accurate but also demonstrate stronger generalizability to bridge the gaps between standard PVC/PS polymers and household PVC/PS microplastics. We also propose MMD heatmap for diagnosing spectral preprocessing methods to further improve method efficiency.