Enhancing tomato leaf nitrogen analysis through portable NIR spectrometers combined with machine learning and chemometrics

Plant-Nitrogen is a vital element that significantly influences plant growth, fruit quality, and yield. However, excessive Nitrogen (N) fertilizer application can have adverse effects on both plant health and environment. Traditional methods of plant-N analysis face numerous challenges like time-consuming procedures, hazardous chemicals, low precision, and limited throughput, hindering efficient and accurate analysis in laboratories. In recent years, Near-Infrared Spectroscopy (NIRS) has emerged as an innovative and rapid method for plant-N analysis, offering non-destructive, cost-effective, and real-time measurements with high accuracy and throughput. This research aimed to explore the potential of utilizing miniaturized handheld Near-Infrared (NIR) sensors in conjunction with machine learning (ML) and chemometrics to assess the N status of tomato plants leaves in the field. Through this study, A comprehensive analysis involving 18 models utilizing different regression and spectral preprocessing methods was conducted. The findings indicate that combining the first- or second-order SG derivatives with the MSC is the optimal spectral data preprocessing method, and that incorporating wavelengths selection techniques, such as GA-PLS and GA-SVM, improves the predictive accuracy of the models by reducing the number of wavelengths considered. The best model's performance was assessed using R2, RMSE, and RPD, yielding values of 81.0–82.8, 7.4–7.8, and 2.3–2.4, respectively. Overall, this study demonstrates that combining NIRS with ML and chemometrics enables effective evaluation of plant-N status using leaves total N content as a proxy variable and emphasizes the significance of integrating spectral preprocessing, modeling, and wavelengths selection techniques to enhance data quality and mitigate the influence of environmental factors that may impact the accuracy of NIRS-based evaluations.