Sensing and classification of rice (Oryza sativa L.) drought stress levels based on chlorophyll fluorescence

Abstract

Sensing and classification of drought stress levels are very important to agricultural production. In this work, rice drought stress levels were classified based on the commonly used chlorophyll a fluorescence (ChlF) parameter (F_v/F_m), feature data (induction features), and the whole OJIP induction (induction curve) by using a Support Vector Machine (SVM). The classification accuracies were compared with those obtained by the K-Nearest Neighbors (KNN) and the Ensemble model (Ensemble) correspondingly. The results show that the SVM can be used to classify drought stress levels of rice more accurately compared to the KNN and the Ensemble and the classification accuracy (86.7%) for the induction curve as input is higher than the accuracy (43.9%) with F_v/F_m as input and the accuracy (72.7%) with induction features as input. The results imply that the induction curve carries important information on plant physiology. This work provides a method of determining rice drought stress levels based on ChlF.

Keywords: Ensemble model; K-Nearest Neighbors; Support vector machine; chlorophyll a fluorescence; drought stress.

Fig. 1. Mean ChlF induction curves for three different drought levels (no drought, drought for 1 h, and drought for 4 h). The error bars are the standard deviations of O, J, I, and P points of the rice samples under the same drought stress level.

Fig. 2. One-way ANOVA of ChlF induction features under different drought levels (different lowercase letters indicate significant differences at p<0.05). F_o – minimal chlorophyll a fluorescence intensity in the dark; F_j – chlorophyll a fluorescence intensity at the J step; F_i – chlorophyll a fluorescence intensity at the I step; F_m – maximal chlorophyll a fluorescence intensity; F_v – variable fluorescence intensity in the dark; V_j – relative variable fluorescence intensity at the J step; V_i – relative variable fluorescence intensity at the I step; F_m/F_o – electron transport through PSII; F_v/F_o – quantum efficiency of PSII; F_v/F_m – maximum photochemical quantum yield of PSII in the dark; M₀ – approximated initial slope (in ms^–1) of the fluorescence transient; Area – area between fluorescence curve and F_m (background subtracted); Fix Area – area below the fluorescence curve between F_40μs and F_1s (background subtracted); S_m – normalized area between ChlF induction curve and the line F = F_m, (multiple turnover);S_s – the smallest S_m turnover (single turnover); N – reduction times of Q_A from F_o to F_m; φP₀ – maximum quantum yield of PSII; ψ₀ – probability that a trapped exciton moves an electron further than Q_A^–; φE₀ – quantum yield of electron transport; φD₀ – quantum yield of energy dissipation; φPav – average (from time 0 to t_Fm) quantum yield for primary photochemistry; PI_ABS – performance index for energy conservation from exciton to the reduction of intersystem electron acceptors; ABS/RC – absorption per reaction center; TR₀/RC – trapped energy flux per reaction center; ET₀/RC – electron transport per reaction center; DI₀/RC – dissipation per reaction center (at t = 0).

Fig. 3. Confusion matrix of Support Vector Machine (SVM) classification of rice drought levels for the training dataset. (A) Based on the maximum photochemical quantum yield of PSII (F_v/F_m), (B) based on induction features, (C) based on the induction curve. D0, D1, and D4 indicate no drought treatment, drought treatment for 1 h, and drought treatment for 4 h, respectively.