Identifying engineered nanomaterials (ENMs) made from earth-abundant elements in soils is difficult because soil also contains natural nanomaterials (NNMs) containing similar elements. Here, machine learning models using elemental fingerprints and mass distributions of three TiO2 ENMs and Ti-based NNMs recovered from three natural soils measured by single-particle inductively coupled plasma time-of-flight mass spectrometry (spICP-TOFMS) was used to identify TiO2 ENMs in soil. Synthesized TiO2 ENMs were unassociated with other elements (>98%), while 40% of Ti-based ENM particles recovered from wastewater sludge had distinguishable elemental associations. All Ti-based NNMs extracted from soil had a similar chemical fingerprint despite the soils being from different regions, and >60% of Ti-containing NNMs had no measurable associated elements. A machine learning model best distinguished NNMs and ENMs when differences in Ti-mass distribution existed between them. A trained LR model could classify 100 nm TiO2 ENMs at concentrations of 150 mg kg-1 or greater. The presence of TiO2 ENMs in soil could be confirmed using this approach for most ENM-soil combinations, but the absence of a unique chemical fingerprint in a large fraction of both TiO2 ENMs and Ti-NNMs increases model uncertainty and hinders accurate quantification.
Keywords: TiO2; classification; engineered nanomaterials; machine learning; sewage sludge; soil; spICP-TOFMS.