We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Improving monitoring of microplastics in soils and biosolids using near-infrared spectroscopy: the effect of organic matter
Summary
Scientists tested whether a light-based scanning tool (near-infrared spectroscopy) can detect tiny plastic bits in soil and sewage sludge that's often used as farm fertilizer, even though the organic matter in these materials can make plastics harder to "see." Using smart computer models, they found this method accurately measured plastic contamination in both clean soil and nutrient-rich sludge, offering a faster, cheaper way to track microplastics before they end up in our food and water supply.
Microplastics (MPs) are emerging contaminants ubiquitously present in the environment, yet their rapid quantification in complex matrices such as soils and biosolids remains challenging because of spectral interference from organic matter (OM). This study investigates, for the first time, how high OM contents influence the spectral detectability and machine-learning-based quantification of high-density polyethylene (HDPE) and polystyrene (PS) polymers (at 0-8% v/v) in soils and biosolids. Pre-processed NIR spectra were analysed using partial least squares-discriminant analysis (PLS-DA) to identify MP- and OM-related wavelength regions via variable importance in projection (VIP), which were subsequently used as inputs for four regression algorithms (PLS, support vector machines, random forest and custom neural network). In soils with low OM (0.78-1.23%), HDPE and PS produced polymer-specific absorption features and were predicted with excellent accuracy (up to R 2 = 0.96 and residual prediction deviation, RPD = 5.19), using a relatively small set of VIP-selected wavelengths. In biosolids with very high OM (64-72%), polymer bands were partially masked, and VIP scores shifted towards OM-dominated regions. However, prediction performance remained robust, especially for HDPE with neural network model (R 2 = 0.98, RPD = 6.41). Overall, although the OM strongly influenced the spectral regions driving model predictions and partially reduced discrimination among MP concentration levels, it does not prevent accurate quantification of MPs when appropriate variable selection and machine-learning strategies were applied. These findings demonstrate that NIR spectroscopy coupled with targeted variable selection and advanced regression can provide robust, non-destructive estimation of MP contamination in OM-poor soils and OM-rich biosolids, highlighting key spectral features to prioritise in future studies targeting natural organic-rich samples with low-MP concentration.