Rapid and nondestructive identification of low-temperature stress severity in Juncao seedlings: Application of chlorophyll a fluorescence combined with visible-near infrared spectroscopy and machine learning.

Abstract

Low-temperature stress severely restricts the geographical distribution and growth of Juncao (Cenchrus fungigraminus), causing growth retardation and yield reduction. Thus, rapid nondestructive monitoring of low-temperature stress is crucial for accurate severity assessment and timely agronomic intervention. The traditional temperature threshold method often misjudges due to individual plant differences. To address this, this study developed a nondestructive method integrating chlorophyll a fluorescence (ChlF), visible-near infrared (Vis-NIR) spectroscopy, and machine learning (ML) algorithms for precise identification of stress levels. Firstly, under gradient temperature treatments, ChlF parameters and Vis-NIR spectral data were synchronously collected from Juncao leaves. Then, a stress classification criterion was established via ChlF parameters and an unsupervised learning algorithm to calibrate the Vis-NIR dataset. Finally, identification models were constructed based on Vis-NIR data and ML algorithms. Results showed that most ChlF parameters were closely correlated with Juncao's physiological and biochemical indicators. All samples were classified into three categories-no stress, mild stress, and severe stress-using ChlF parameters combined with the K-means clustering algorithm. SHapley Additive exPlanations (SHAP) analysis revealed the maximum photochemical efficiency of PSII as the top contributing classification indicator. Clustering reliability was validated by significant intergroup differences (P < 0.05) in ChlF transients, antioxidant enzyme activities, malondialdehyde (MDA) content, photosynthetic pigments, and SPAD values. Specifically, with increasing stress intensity, ChlF induction kinetic curves, Vis-NIR reflectance curves, chlorophyll a, chlorophyll b, total chlorophyll, and SPAD values decreased, while superoxide dismutase (SOD), peroxidase (POD), and MDA content generally increased. Among all combinations, Savitzky-Golay (SG) smoothing of Vis-NIR data combined with a one-dimensional convolutional neural network (1D-CNN) exhibited the optimal and robust performance, with a test set accuracy of 90.00 ± 0.73 %. This study confirms that integrating ChlF, Vis-NIR spectroscopy, and ML enables rapid nondestructive identification of low-temperature stress severity in Juncao seedlings, providing an efficient technical tool for monitoring physiological status and chilling injury early warning of Juncao.

Plant phenotyping relevance

ChlF・Vis-NIR・機械学習を統合し、植物の低温ストレス重症度を非破壊推定する手法の開発と検証が中心である。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.