Rapid detection and quantification of atmospheric heavy metal deposition on plant leaves using machine learning-enhanced NIR spectroscopy

Abstract

Plant leaves serve as effective bioindicators for monitoring atmospheric heavy metal pollution because of their efficient adsorption and accumulation of airborne contaminants. This study presents an enhanced integration of near-infrared (NIR) spectroscopy with advanced machine learning algorithms for rapid, nondestructive detection of atmospheric heavy metals in Ligustrum japonicum ’Howardii’ leaves. We developed a comprehensive analytical framework incorporating 15 preprocessing methods, six advanced feature selection algorithms, and state-of-the-art machine learning models. The NIR spectra (780–2,500 nm) revealed characteristic absorption features in the 900–1,000, 1,400–1,500, and 1,900–2,000 nm regions corresponding to heavy metal-induced physiological changes in chlorophyll, water-protein structures, and metabolites, respectively. The proposed framework employed systematic optimization strategies to test 450 unique preprocessing–feature selection model combinations through rigorous validation protocols and systematic noise robustness testing. For qualitative classification, the optimal combination of baseline correction preprocessing, principal component analysis-based feature selection, and logistic regression achieved perfect discrimination accuracy (100.0 % Leave-One-Out Cross-Validation) across all three heavy metal types (lead, chromium, and nickel), maintaining exceptional noise tolerance with >94 % accuracy under 20 % noise conditions. For quantitative analysis, metal-specific optimization strategies yielded superior performance, achieving R² values exceeding 0.83 across all three heavy metals under study, with chromium and nickel surpassing 0.88. The proposed enhanced methodology demonstrated substantial improvements over traditional single-method approaches while providing mechanistic insights into heavy metal-plant interactions suitable for regulatory applications and automated environmental monitoring systems.