Sources of variance in handheld NIR analysis of soil-embedded microplastics

Soil-embedded microplastics are challenging for diffuse-reflectance NIR spectroscopy because spectra are dominated by a heterogeneous soil matrix, while polymer contributions appear as weak spectral perturbations that can be masked unless sample presentation is well controlled. Here, we quantified the main sources of spectral variability in standardized agricultural soil spiked with four common polymers (PA66, PMMA, PS, and PC) at three mass loadings (0.5-4.0 wt%) and multiple particle-size fractions (25-50 μm to >425 μm). Benchtop FT-NIR spectra (Büchi NIRFlex N-500) were evaluated using principal component analysis, and separations between sample groups were quantified as distances between group-average spectral signatures in a common principal component space. Within the controlled design applied here, polymer identity provided the strongest chemically meaningful separation, while increasing polymer loading produced smooth, systematic shifts. In contrast, particle-size fraction did not yield a reproducible ordering after scatter correction. Size-related shifts were small and non-systematic after scatter control, indicating that particle-size fraction is not a dominant confounder for polymer discrimination and loading trends under these soil-dominated conditions, with the clearest residual effect confined to the coarsest fraction. Additionally, spectra acquired on two handheld FT-NIR NeoSpectra units revealed a pronounced serial-number effect, i.e., a unit-specific spectral structure that can dominate apparent specificity of microplastics properties. Removing this unit-dependent component with a low-rank correction aligned the two devices while retaining chemically relevant differences in the present setting. Overall, the study explores the variance hierarchy that governs diffuse-reflectance NIR analysis of soil-embedded microplastics under controlled dry-soil conditions. These results demonstrate that, for on-site portable NIR analysis of soil microplastics, the most robust recoverable information is polymer-related differentiation together with broad mass-loading trends, while particle-size-related effects remain secondary and can be masked or distorted when sample presentation is not sufficiently controlled.