Rapid Screening of Pesticide Residues Using Vibrational Spectroscopy, Theoretical Calculations, and Machine Learning: a Case Study on Chlorothalonil

Abstract

Chlorothalonil, a fungicide commonly used in agriculture to combat fungal infections, is a major concern due to its persistent residues in the food supply, especially in countries like Kenya where vegetable consumption is high. This thesis develops a rapid, non-destructive methodology to detect chlorothalonil residues on vegetables, integrating vibrational spectroscopy (Raman and Near-Infrared Spectroscopy - NIR), computational spectroscopy (Density Functional Theory - DFT), and machine learning tools. The study focuses on four key vegetables: bell peppers (Capsicum annuum), kale (Brassica oleracea var. sabellica), tomatoes (Solanum lycopersicum), and pigweed (Amaranthus spp.) due to their widespread consumption and vulnerability to fungal diseases, for which chlorothalonil is widely applied. Through Raman and NIR spectroscopy, chlorothalonil residues were identified by exploiting their unique molecular vibrational fingerprints. These were further enhanced using DFT simulations, incorporating anharmonic corrections through Generalized Vibrational Perturbation Theory (GVPT2), along with the B3LYP and B2PLYP functionals paired with advanced basis sets like N07D and SNSD. This computational approach significantly improved the accuracy of vibrational spectra simulations, particularly in resolving overlapping spectral bands often encountered in complex agricultural matrices. Two-Dimensional Correlation Spectroscopy (2D-COS) was employed alongside chemometric techniques such as Principal Component Analysis (PCA) and Support Vector Machines (SVM) to improve the interpretability and classification of spectral data. The combined Raman and NIR methodologies demonstrated remarkable efficiency, with spectral acquisition completed in just 30 seconds per sample and the developed machine learning pipeline delivering results within 10 seconds. These rapid, non-destructive processes, coupled with high accuracy, underscore the potential of integrating advanced spectroscopy techniques with computational and machine learning methods for real-time pesticide residue detection in agricultural matrices.