Roles of Physiological and Nonphysiological Information in Sun-Induced Chlorophyll Fluorescence Variations for Detecting Cotton Verticillium Wilt

Sun-induced chlorophyll fluorescence (SIF) has been a promising indicator of plants’ physiological status, but its response to physiological changes under the verticillium wilt (VW) stress of cotton plants is complicated by concurrent nonphysiological changes. The relative contributions of the above two components to SIF variations are unclear at different VW stress severities, hindering the accurate diagnosis of VW levels. Therefore, the objective of this study is to investigate the dynamic responses of SIF, physiological and nonphysiological factors, and to evaluate the contributions of the two factors to SIF variations under different VW stress degrees of cotton. We continuously observed the diurnal variation of the top-of-canopy reflectance and SIF on healthy and VW-infected cotton during the peak incidence period of VW disease. To accurately quantify the relative contribution of each component to SIF, we proposed a practical strategy to estimate unmeasurable parameter when using Lindeman, Merenda, and Gold method. The results demonstrated the dominant role of physiological factors in SIF with an arch diurnal change pattern at the early stages of VW development. As the VW severity increased, the contribution of physiological components declined, with the maximum contribution decreasing by 47.7%, and the diurnal pattern was disrupted, with the diurnal variation amplitude of the parameter declining by 63.4%, followed by a shift in the regulatory role toward nonphysiological factors. This study contributes to a further understanding of the roles of physiological and nonphysiological components in SIF variations of cotton under VW stress, thus advancing the accurate monitoring of VW stress severity.

SECTION I. Introduction

Cotton is an important cash crop and natural fiber supply source in the world. Verticillium Wilt (VW) is the most common and devastating disease occurring in the growth and maturation of cotton [1], [2]. The Verticillium dahliae that causes VW first invades the root surface of cotton in the soil [3], and then the vascular tissue, finally accumulates and expands in the cotton plant blocking the transport of water and nutrition. This obstruction caused a decrease in photosynthesis, the browning of vascular bundle, and the yellow wilting of cotton leaves.

Under VW stress, the morphology and physiological state of cotton undergoes changes, resulting in alterations in some characteristic bands of the spectrum compared with healthy plants. Previous studies have used remote sensing technology to extract spectral characteristics at both leaf [4], [5], [6] and canopy [7], [8], [9], [10], [11] scales, aiming to monitor the impact of VW on cotton plants and to determine the degree of VW stress. Disease monitoring based on reflectance spectral characteristics relies on spectral responses to variations in physiological, biochemical, and canopy structure indicators, such as vegetation chlorophyll content, leaf water content, and leaf area index [12]. However, the above indicators do not change significantly until the plant has been stressed for a period, which cannot reflect the stressed state of vegetation timely at an early stage [13]. Compared with reflectance spectra, chlorophyll fluorescence spectra can reflect physiological changes earlier and more sensitively [14], [15], [16], [17].

Sun-induced chlorophyll fluorescence (SIF) is a signal emitted by plant chlorophyll molecules in the range of 650–800 nm under natural sunlight. SIF is closely related to plant photosynthesis, and previous studies have demonstrated its potential to detect the state of vegetation under environmental stress quickly [18]. Some researchers have used SIF to monitor the response of vegetation under stress, such as disease stress [7], [8], [19] and water stress [20], [21], [22]. As a byproduct of electron re-excitation during the process of photosynthesis, SIF is first transmitted from the photosystem to the leaf surface, then interacts with the canopy leaves (reflection, scattering, and absorption) before escaping the canopy, and finally reaches the sensor. The complex process of radiation transmission makes the canopy SIF signal not only contain the information on plant photosynthetic physiology but also is affected by changes in canopy structure [23], [24], [25]. According to the light-use-efficient concept model of gross primary production (GPP) [26], the SIF conceptualization formula [18] is shown as follows:

$$\begin{equation*} {\rm{SIF\ = \ PAR \times fPAR \times \Phi }}F{\rm{ \times }}{{f}_{\text{esc}}}. \tag{1} \end{equation*}$$

The canopy SIF observed by the sensor consists of four factors, the photosynthetically active radiation (PAR), the fraction of the PAR absorbed by the vegetation (fPAR), the fraction of all emitted SIF photons escapes the canopy (f_esc), and the physiological SIF emission efficiency at the canopy level (ΦF). Among them, PAR, fPAR, and f_esc are usually used to characterize nonphysiological information, and ΦF is used to characterize physiological state of leaves [25]. When the sensor detects the change in SIF, it is actually the composite result of multiple factors.

Previous studies have employed active chlorophyll fluorescence to monitor the effect of VW at the leaf scale [5], [27]. Chlorophyll fluorescence at the leaf level characterizes the physiological status of individual leaves, and the physiological status of leaves at different positions on the same plant varies. Therefore, solely relying on leaf-level fluorescence is challenging to accurately reflect the degree of stress experienced by the entire vegetation. Fluorescence at the canopy level is needed to reflect the stress status of the vegetation. However, the observed SIF at the canopy scale is also influenced by the canopy structure. The presence of VW in cotton plants induces changes in both photosynthetic physiology and canopy structure. Both of these alterations may lead to a reduction in SIF at the canopy scale. Identifying the primary factor driving this reduction would facilitate the differentiation of cotton plants under VW stress from healthy ones and contribute to the accurate monitoring of VW severity. Therefore, it is essential to understand the relative importance of each factor in influencing the change of SIF under the stress of VW at different levels in order to further understand the impact of VW stress on cotton.

The relative importance analysis method proposed by Lindeman, Merenda, and Gold (LMG) is employed to explore the contribution of a single regression to the change of the dependent variable [28]. This method effectively mitigates the influence of variable order in regression and accurately computes the relative importance of each independent variable concerning the dependent variable. Some scholars have used it to investigate the relative importance of factors in (1) to the change of SIF [29], [30]. The parameter ΦF in the SIF conceptual model, which cannot be measured directly, is usually derived through calculations involving SIF and other parameters in current research, as shown in (2). It is difficult to represent f_esc and fPAR individually; thus, former studies always using a spectral index to represent fPAR × f_esc, such as near-infrared reflectance of vegetation (NIRv). When calculating the relative importance of PAR, ΦF, and NIRv to SIF variations, using the same index to represent fPAR × f_esc and calculate ΦF results in ΦF being expressed linearly in terms of other parameters (SIF, PAR, and NIRv). This violates the model's requirement for the independent variables to be linearly independent. We found that introducing random errors could mask this conflict, creating the illusion of data that satisfies the requirements, thereby enabling the model to produce a result. This will introduce uncertainty into the results, as random errors may alter the relative importance of each independent variable. It is essential to propose a practical strategy for estimating ΦF that meets the model's requirements and assess the influence of random errors of the original estimating method

$$\begin{equation*} {\rm{\Phi }}F = \frac{{\text{SIF}}}{{{\rm{PAR \times fPAR \times }}{{f}_{\text{esc}}}}}.\ \tag{2} \end{equation*}$$

This study aims to explore the effects of physiological and nonphysiological factors on the SIF of cotton under VW stress. We conducted continuous observation on seven cotton canopies and performed the following work:

1. explore the responses of nonphysiological factors, physiological factors, and SIF under different levels of VW stress;

2. quantify the contributions of physiological and nonphysiological information to SIF at different degrees of VW severity;

3. propose a practical strategy to estimate ΦF and quantify the impact of random errors on relative importance results.

SECTION II. Materials and Methods

A. Study Site

The study area is located in the VW disease field of the Shihezi Institute of Agricultural Sciences (44.33°N, 86.05°E) and the healthy field of the Agriculture College of Shihezi University (44.33°N, 86.06°E). In contrast to the healthy field, where the soil is free of Verticillium dahliae, the soil of the VW disease field is inoculated with the pathogen of VW before planting cotton each year. The two experimental fields have close distance and similar management measures, so they are good control subjects. Shihezi is located at the northern foot of Tianshan Mountain and the southern edge of Junger Basin. It has a temperate continental climate, with an average annual precipitation of 125–208 mm and an average annual temperature of 6.5–7.2 °C [10].

B. Field Measurements

Canopy scale spectrum and SIF diurnal variation data were obtained using an automated fluorescence observation system, including a QE65 Pro spectrometer (Ocean Optics Inc., Dunedin, FL, USA). The QE65 Pro spectrometer has a band range of 640–800 nm, a spectral resolution of 0.39 nm, a spectral sampling interval of 0.15 nm, and a signal-to-noise ratio of 1000:1. The optical fiber used to measure the incident downwelling irradiance (E) is fitted with a cosine corrector, and the upwelling radiance (L) is measured through a bare optical fiber with a field-of-view of 25°.

From 26 July 2023 to 26 August 2023, long-term diurnal cotton canopy monitoring was conducted at six typical sites in the VW disease-affected fields (denoted as VW1–VW6) and at one typical site in the healthy field (denoted as HL1)(see Fig. 1). The probe for measuring E was installed on a straight rod and kept vertically upward at all times. The bare optical fiber for measuring L was installed on the instrument bracket to keep the probe always vertically downward to the canopy. Considering the width of the cotton plants in the field, the probe was positioned approximately 1.1 m above the top of the canopy. During the observation period, the daily observation time was from 10:00 to 18:00 (Beijing time) so as to prevent large errors in the measurement data when the solar altitude angle was too small.

In this study, the 3FLD (three bands FLD, 3FLD) algorithm proposed by Maier et al. [31] was used for SIF inversion, where ${{\lambda }_{{\rm{in\ }}}} = \text{760.52 nm}$, ${{\lambda }_{{\rm{left}}}} = \text{758.05 nm}$, and ${{\lambda }_{{\rm{right }}}} = \text{768.03 nm}$.

C. Parameter Calculation

We used near-infrared (R_NIR) and red (R_red) reflectance of plant canopy to calculate the normalized difference vegetation index (NDVI) and then the NIRv, which is widely used to approximate fPAR × f_esc [32]. We calculated NIRv using the canopy reflectance at 650–670 nm (red band, ${{R}_{\text{red}}}$) and 760–800 nm (near-infrared band, ${{R}_{\text{NIR}}}$). NIRv is expressed as follows [33]:

$$\begin{equation*} {\rm{NIRv\ = \ NDVI \times \ }}{{R}_{\text{NIR}}} = \frac{{{{R}_{\text{NIR}}} - {{R}_{\text{red}}}}}{{{{R}_{\text{NIR}}}{\rm{ + }}{{R}_{\text{red}}}}}{\rm{\ \times }}{{R}_{\text{NIR}}}. \tag{3} \end{equation*}$$

The fluorescence correction vegetation index (FCVI) is an index developed for disentangling physiological and nonphysiological information infrared SIF [34]. In addition, the combined impacts of fPAR and f_esc can also be effectively approximated using FCVI. It was originally defined as the difference between near-infrared reflectance and broadband visible reflectance ranging from 400 to 700 nm (${{R}_{\overline {\text{vis}} }}$)

$$\begin{equation*} {\rm{FCVI\ = }}{{R}_{\text{NIR}}} - {{R}_{\overline {\text{vis}} }}. \tag{4} \end{equation*}$$

Due to the limitation of the sensor band range, the reflectance of 650–700 nm is used to approximate the wideband visible light reflectance according to Yang et al. [35].

Canopy scale chlorophyll content was characterized by red edge NDVI (RENDVI) [36]

$$\begin{equation*} {\rm{RENDVI\ = }}\frac{{{{R}_{\text{NIR}}} - {{R}_{{\rm{red\ edge\ }}}}}}{{{{R}_{{\rm{NIR\ }}}}{\rm{ + }}{{R}_{{\rm{red\ edge\ }}}}}}.\ \tag{5} \end{equation*}$$

In this study, no equipment could continuously observe PAR throughout the day, so the integration of incident radiation in the spectrum of 660–800 nm was used to approximate PAR, and the feasibility of this characterization method was also tested using actual measured PAR values. We denoted the integration of incident radiation in the spectrum of 660–800 nm as PAR_proxy and defined SIF_PAR = SIF/PAR_proxy.

It can be found from (1) that ΦF is a parameter that characterizes the physiological state of leaves, and PAR, fPAR, and f_esc represent the nonphysiological information, where PAR is photosynthetically active radiation, which is independent with the growth state of vegetation. Therefore, we separated PAR as an independent factor and combined fPAR and f_esc to represent the nonphysiological state of plant. In this study, we used NIRv and FCVI to approximate the characterization of fPAR × f_esc [32], [33], [34], then (1) convert into the following equation:

$$\begin{align*} \text{SIF} \approx& {\rm{PAR \times \Phi F \times NIRv\ }}\\ \approx& {\rm{PAR \times \Phi F \times FCVI}}. \tag{6} \end{align*}$$

According to (6), ${\rm{\Phi }}F$ is calculated as follows:

$$\begin{equation*} {\rm{\Phi }}F = \frac{{\text{SIF}}}{{{\rm{PAR \times fPAR \times }}{{f}_{\text{esc}}}}}\ \approx \frac{{\text{SIF}}}{{{\rm{PAR \times FCVI}}}}. \tag{7} \end{equation*}$$

D. Relative Contributions of Physiological and Nonphysiological to SIF

This study used the LMG model [28] to analyze the relative importance of illumination (PAR), physiological (ΦF), and nonphysiological (fPAR × f_esc) contributions to SIF changes. For variables with a linear relationship between the dependent and independent variables, the model can quantify the contribution of an independent variable to the change of dependent variable while considering the varying independent contributions of independent variables to dependent variables, as well as the different correlations between independent variables. The core idea of the LMG model is that R² represents the proportion of variance explained by a set of independent variables, different R² values are obtained by different combinations of independent variables, and finally the individual contribution of each independent variable to the change in the dependent variable is calculated. In this study, we first took the natural logarithm of (6) to make it into linear form, thus obtaining (8), and then used R language package relaimpo [37] to calculate the contribution of each independent variable to the dependent variable

$$\begin{equation*} {\rm{ln(SIF)\ = \ ln(PAR) + ln(NIRv) + ln(\Phi }}F{\rm{)}}{\rm{.}} \tag{8} \end{equation*}$$

In relative importance analysis, we used NIRv to denote nonphysiological information, and when estimating ΦF, we employed FCVI to represent nonphysiological information according to (7).

As mentioned in Section I, adding random errors to input data enables the model to produce results. This study introduced 5 random error terms (see Table I) to 15 parallel samples, exploring their diverse impacts on the outcomes. To reduce the randomness of the experimental results, for the latter four types of Gaussian random noise in Table I, the experiment was repeated five times, and the average and variance of the relative changes were compared.

TABLE I Random Errors

E. Data Preprocessing

Since the fluorescence automatic observation system used in the experiment works in a sequential measurement mode, it allows only one channel signal to be measured at a time using a single spectrometer. For each observation point, the E was measured (E_M), and then the L, with a 15–60 s delay in the measurement time. Under unstable weather conditions, rapid changes in the illumination occurred during the measurement, leading to notable differences between the real E when measuring L (E_L) and the measured E_M. It will result in a significant error in the calculation of canopy reflectance SIF derived from E and L, introducing uncertainty into the subsequent data analysis [38]. Therefore, before data processing, we filtered and removed abnormal data caused by abrupt changes in light intensity.

Cotton field disease outbreaks are influenced by the climatic conditions [39], especially temperature and humidity. The optimum temperature for the onset of VW of cotton is 22–28 °C, higher than 30 °C is not conducive to the onset, and more than 35 °C does not develop symptoms [40], [41], [42]. Meanwhile, the occurrence of VW of cotton is also related to humidity, and it is easy to develop when the relative humidity is suitable. Cotton flowering and boiling stage is during the summer in Xinjiang; most of the time, the temperature is high and the relative humidity is low, and the development of the disease is inhibited. When there is a sudden drop in temperature accompanied by rainfall, offering suitable conditions for disease onset, it is easy to cause a large area outbreak of VW. Based on the characteristics of cotton VW, the whole experimental period was divided into four subperiods (T1–T4) according to two temperature drop events and one regular irrigation event in the field. As shown in Fig. 2, T1–T2 and T2–T3 are divided by temperature drop events. T3–T4 is divided by a regular irrigation event since large-scale irrigation in a field will substantially change soil temperature, humidity, and near-surface microclimate. It was believed that the stress degree of cotton plants was prone to become more severe after each event. In this study, we used T1|T2 to represent the time period between T1 and T2, and similarly, this applies to other time periods.

Fig. 1.

Canopy observation points (VW1–VW6 are the six observation points in the disease field, and HL1 is the healthy points in the healthy field).

Fig. 2.

Subperiod division during the experiment.

SECTION III. Results

Based on the development of VW stress in cotton plants throughout the entire experimental period, we categorized six VW observation points into three groups. VW1 represented a rapid intensification of VW stress during a certain subperiod, and VW2 and VW3 represented a gradual onset of disease throughout the entire experimental period, culminating in a severe stress level; VW4, VW5, and VW6 represented a gradual onset of disease throughout the entire experimental period, resulting in a relatively lighter stress level. In the rest of this article, we will analyze the results using VW1, VW3, and VW4 as representatives, and the results for the rest VWs are shown in Appendix (see Figs. 9 and 10).

A. Responses of Vegetation Parameters to VW Stress

The changes of NIRv and FCVI, which represent the canopy structure characteristics, are shown in Fig. 3. The HL1canopy remained stable through all periods, whereas the canopy values of all VW observation points showed a downward trend with the increasing of VW severity. Several observation points, including VW3 and VW4, exhibited a gradual decline from T1 to T4, while specific observation points, such as VW1 during the T2 period, experienced a sudden plummet. As shown in Fig. 4, the VW1 canopy during T2 was much sparser than that during T1 due to the outbreak of VW of cotton plants in the canopy of VW1 region with the abrupt temperature drop of T1|T2. Based on the daily on-site investigation of cotton plants, the survey results indicate that during the T1 period, VW1 already exhibited mild symptoms of verticillium wilt. As VW1 had already been affected by verticillium wilt prior to the T1|T2 event, it was comparatively more vulnerable than the cotton plants at other observation points. Therefore, experiencing the same temperature drop event during T1|T2, VW1 would be more significantly affected and subsequent changes would be more pronounced. The RENDVI value of HL1 was stable in T1–T3 period, with a slight decline in T4 period. In contrast, the canopy exposed to VW stress had decreased before T4 period. VW1 exhibited a decline in the late T2 period, followed by a rebound in the T3 period and declined again in the T4 period. VW3 started to decline in the T2 period. VW4 remained relatively stable in the first two periods and began to decline in the T3 period.

Fig. 3.

Daily variations in nonphysiological factors (NIRv, FCVI, and RENDVI), SIF, SIF_PAR, and physiological factors (ΦF) during the experimental period. Each column represents VW1, VW3, VW4, and HL1.

Fig. 4.

View above and below the canopy of VW1 in the T2 period. Pictures in each column were taken on 6 August 2023, 9 August 2023, and 13 August 2023).

Throughout the entire observation period (see Fig. 3), the SIF of all seven observed canopies, including HL1, showed a general declining trend, although with varying decline rates observed across different regions. The decline of SIF, following a period of initial stability, started at different times according to the varying onset times and degrees of VW stress. Among them, VW1 was relatively stable in T1 period, decreased in T2 period, and maintained its stability in T3 and T4 periods. The remaining VW observation points showed a slight decrease in T1 period, remained relatively stable in T2 period, and exhibited an obvious decreasing trend from T3 period. To eliminate the influence of PAR, we normalized SIF with PAR_proxy (SIF/PAR_proxy) as SIF_PAR, which exhibited a similar trend to SIF. The overall variation trend of ΦF, representing the physiological component of SIF, fluctuated and decreased at each observation point in different amplitudes. It also fluctuated between and within each subperiod.

B. Relative Importance of Individual Drivers to SIF Variations Under VW Stress

In Fig. 5, it can be found that the diurnal variation pattern of the relative importance of HL1 is consistent at each period of T1–T4. The diurnal variation pattern of the relative importance of PAR shows a “U” shape. The relative importance of ΦF exhibits an arch shape, which is lower in the morning and late afternoon, but higher near noon. The relative importance of NIRv is low and its variation is minimal. For observation points under VW stress, the pattern of variation for each factor in different periods was not as stable as HL1. In addition, the diurnal variation pattern of the relative importance of each parameter was found to be closely related to the degree of exposure to VW stress.

Fig. 5.

Results of diurnal changes in the relative importance of each subperiod. The x-axis coordinates have been converted to local time. Each row corresponds to subperiods T1, T2, T3, and T4 and each column represents VW1, VW3, VW4, and HL1.

In the T1 period, the diurnal variation patterns of the VW1, VW3, and VW4 are similar to that of HL1. In the T2 period, the importance of ΦF for VW1 on SIF exhibited substantial variation, with its maximum values declined by 28.1%. Its diurnal variation pattern was destroyed, with higher values in the morning gradually decreasing until noon, increasing in the afternoon, followed by a subsequent decrease. Compared with T1 period, the importance of NIRv increased remarkably, with a maximum value increased by 358.7%, and displayed a diurnal variation pattern with an arch shape. It is noteworthy that its importance surpassed that of ΦF after 11 o'clock due to the pronounced intensification of VW stress of VW1 in the T2 period. The abrupt temperature drop during T1|T2 led to the outbreak of VW in cotton plants within the observation area of VW1, resulting in a significant alteration of the canopy structure from dense to sparse (see Fig. 4). In T3, the diurnal pattern of ΦF in VW1 partially recovered, although the value was low. Moreover, the importance of NIRv surpassed that of T1 and exceeded ΦF in the morning. The diurnal variation pattern of VW1 in T4 was similar to that in T3, and the importance of NIRv increased. The diurnal variation of VW3 parameters in T2 period was similar to that in T1 period, but it changed greatly in T3 period. The diurnal pattern of the importance of ΦF no longer presented an arch shape, instead, the importance within each time period in the morning is similar, as is the importance in the afternoon. The relative importance of NIRv increased compared with the previous two periods and exceeded ΦF in the afternoon. During T4, the relative importance of NIRv further increased compared with T3, exceeding ΦF at most times of the day. Meanwhile, the maximum value of ΦF decreased by 47.7%, and its diurnal variation amplitude declined by 63.4% compared with T1. The diurnal variation of the relative importance of VW4 parameters in T2 and T3 was similar to that in T1. In addition, the relative importance of ΦF in T3 experienced a slight decrease of 4.9%, yet it remained higher than other parameters. During T4, the relative importance of ΦF for VW4 decreased by 16.5% with a lower amplitude, declined by 3.7% compared with T1 period. The relative importance of NIRv became higher than PAR, but still lower than ΦF during midday.

In summary, at the early stages of VW development, SIF was mainly influenced by ΦF representing the photosynthetic physiological state of cotton. Its contribution increased in the morning, reached the maximum at noon, and then gradually decreased. Nonphysiological factor had a minor effect on SIF, with a limited range of change in a single day. As the VW stress intensified, the influence of ΦF on SIF decreased further, with the diurnal pattern destroyed, while the impact of NIRv increased, surpassing that of PAR but still lower than ΦF at noon. Under more severe VW stress, the contribution of ΦF on SIF was similar to that of NIRv, and its value remained stable throughout the day. NIRv increased further than before, was potentially higher than ΦF at noon, and even at most of the day under more severe stress.

C. Relative Change Induced by Random Error

Due to the limitations of the estimation method for ΦF in the existing studies, we used FCVI to calculate ΦF, which meets the requirement of the LMG model. In order to verify the capacity of NIRv and FCVI to represent the same physical meaning, and consequently the feasibility of this method in estimating ΦF, we conducted a linear regression analysis on both (see Fig. 6).

Fig. 6.

Linear regression results for NIRv and FCVI.

It was found that the relative changes caused by the retention of 7 decimal places (Round7) were not too large (see Fig. 7), and within the 15 parallel samples, the absolute value of the maximum relative change was 7.4%. Among the four types of Gaussian random noise, the one with a standard deviation of 0.01 (GS_001) caused the smallest relative change, but it still caused a change of 12.7% [see Fig. 7(f)]. With the increase of the standard deviation, the relative change in the result gradually expanded, accompanied by a gradual decline in stability. It led to large relative changes and poor stability of the results to set the standard deviation to 0.1 times the standard deviation of the parameter (GS_Std01). Similarly, setting the standard deviation as parametric standard deviation (GS_Std) caused remarkable changes in the relative importance results, with the highest relative change reaching 216.1% [see Fig. 7(b)], and the results exhibited substantial instability [see Fig. 7(c)].

Fig. 7.

Variation introduced by random error in relative importance results. S1–S15 represent 15 parallel samples, with each sample undergoing five repeated tests. Results in the left column represent the average of relative changes caused by errors and results in the right column represent the standard deviation of relative changes caused by errors.

In conclusion, the relative changes of results caused by various random errors indicate that retaining seven decimal places (Round7) led to relatively minor alterations among five types of random errors. Gaussian random noise with a standard deviation set at 0.01 (GS_001) resulted in small and stable relative changes. Conversely, the relative changes caused by Gaussian random noise with a standard deviation set at its own standard deviation (GS_Std) were remarkable and highly unstable. Based on the above experimental findings, it is recommended that when using the LMG model to assess the relative importance of each parameter in the SIF conceptual model, alternative parameters representing the same physical characteristics can be employed to calculate ΦF. For instance, in this study, both NIRv and FCVI serve as the representations of nonphysiological factors, exhibiting a notably high coefficient of determination in their linear regression. Thus, when NIRv was utilized for estimating fPAR × f_esc, FCVI was used for estimating ΦF, presenting a practical calculation strategy aligned with the model's principles.

SECTION IV. Discussion

A. Effects of VW on Cotton Canopy

Various vegetation parameters reflect the state of vegetation canopy in different aspects. As depicted in Fig. 3, VIs used to depict canopy structure exhibit different responses in the canopy of healthy cotton plants compared with those under varying levels of stress. Comparing the values at the end with the beginning of the experimental period, it can be found that the decrease in VW3 is much greater than that in VW4. This phenomenon occurs due to the progression of VW, which is characterized from the bottom up, with the leaves in the lower part of the canopy first turning yellow and shedding. With the aggravation of the stress, the upper leaves may also gradually fall off, resulting in sparse cotton canopy [43]. Comparing Fig. 8(b) and (d), it can be observed that VW3 in the T4 period exhibited a reduction in canopy leaves and a notable increase in the proportion of sky compared with T1 period. Differently, VW4 showed a minor variation, which possibly indicates its lower susceptibility to VW stress than VW3, leading to a lower level of stress during the same time period compared with VW4. Leaf senescence and subsequent shedding, resulting from VW stress, initially manifest in the lower canopy leaves. However, these changes under the canopy are challenging to be detected under the vertical downlook observation mode of this experiment. RENDVI, indicating the canopy chlorophyll content, showed a declining trend across all observation points. The comparison of the RENDVI value of VW3 and VW4 in Fig. 3 revealed a more pronounced decline in VW3 than that in VW4. This finding is consistent with previous studies that conducted physical–chemical experiments on leaves exposed to various stress levels, which concluded that chlorophyll content was lower in more severely stressed leaves [6], [44], [45]. This is also evident in Fig. 8, where during the T4 period, leaves at the upper canopy of VW3 displayed yellowing [see Fig. 8(c) and (d)], while those of VW4 and HL1 still appeared green [see Fig. 8(g) and (h)]. This phenomenon occurs due to the accelerated yellowing of leaves at equivalent leaf ages or positions in both the upper and lower sections of cotton plants affected by VW compared with healthy counterparts. This leads to a substantial reduction in chlorophyll content, especially in severely affected cotton plants [46]. The RENDVI of VW4 and HL1 canopy exhibited a relatively small decrease, potentially attributed to the fact that, during the T4 period, the cotton plants had reached the late stage of their growth period, naturally leading to an overall decrease in chlorophyll in the canopy as the natural progression of cotton plant growth toward senescence. In addition, the yellowing and aging of leaves below the canopy due to VW stress in VW4 would further contribute to a decline in canopy chlorophyll content compared with HL1.

Fig. 8.

View above and under the canopy of VW3 and VW4 during T1 and T4.

The SIF and SIF_PAR of VW3, VW4, and HL1 presented a decreasing trend from T1 to T4, with VW4 and HL1 showing a smaller decrease compared with VW3. This suggested a reduction in the photosynthetic performance of cotton plants under VW stress. This observation is consistent with previous studies on leaf scale chlorophyll fluorescence [5], [27], [44], as well as existing researches on vegetation under disease stress [7], [8], [19]. Further observation revealed a slight decrease in the SIF of VW3 in T1–T2 period, with an increased amplitude of decline in T3–T4 period. Conversely, VW4 exhibited a minimal decrease throughout the whole experimental period. The above variations may stem from differences in disease tolerance among the cotton plants. Although under the same growing environment, VW4 experienced milder stress compared with VW3, with the stress degree of VW4 in T4 similar to that of VW3 in T2. This phenomenon may be attributed to the fact that, when the stress level is mild, the leaves have not been directly infected by Verticillium dahliae, but the physiological state of the entire cotton plant undergoes changes due to the stress. As the stress degree intensified, photosystem Ⅱ (PSⅡ) was damaged, resulting in weakened photosynthesis. With further intensification of VW stress, the chlorophyll content of the leaves decreased, and the vascular bundle of the cotton plant was blocked, hindering the transport of water and nutrients inside cotton plants, leading to a significant inhibition of the primary reaction of photosynthesis [45], [47], [48]. The fluctuation of ΦF during T1–T4, combined with the decrease of SIF_PAR, suggested that the decline of SIF during the whole period following VW stress was the outcome of both physiological and nonphysiological changes [20], [22], [25], [49], [50].

B. Dominant Drivers of the Variations of SIF Under VW Stress

It is worth noting that the diurnal variation of relative importance of each component was a composite result influenced by two factors: differences in the same time period within a day and among days within subperiod. As shown in Fig. 5, in the T1 period, the relative importance of PAR obtained higher values in the morning and evening, and lower values near noon. This is because the absolute value of PAR undergoes an arch-like change, exhibiting large changes in the morning and evening, and minor changes near noon in the same time interval due to the minimal variation in PAR among days within subperiod.

The results showed that the diurnal variation of each parameter's importance to SIF variations could reflect the degree of VW stress. At the early stages of VW development, the SIF variation was prominently driven by physiological factors (ΦF) for the following reasons. On the one hand, the cotton plant demonstrated a strong capacity for self-regulation of its photosynthetic processes. The rate of photosynthesis fluctuates in response to variations in light intensity, air temperature, and water conditions [51], [52], [53], [54]. On the other hand, when initially subjected to VW stress, reversible inactivation of the PSⅡ reaction center happened and nonphotochemical quenching (NPQ) was activated as a photoprotective mechanism in protecting plants from photodamage [55], resulting in the decline of ΦF. This is similar to drought and heat stress, which induce a decline in fluorescence emission due to an increase in NPQ [20], [22], [56], [57]. This occurs in this study because VW stress causes vessel occlusion, leading to water deficiency in cotton plants [58]. As VW stress intensified, the internal physiological structure of cotton leaves was damaged [5], [47], leading to a reduced photosynthetic activity in mesophyll cells. Meanwhile, some nonphysiological impacts of VW stress on the canopy appeared from bottom to up, such as leaf chlorosis, necrosis, and finally leaf abscission [58], causing the decline of leaf chlorophyll content and leaf area at the canopy scale [8].

The effect of physiological and nonphysiological on SIF variations is closely related to the severity and type of stress. SIF variations were influenced by both physiological and nonphysiological aspects under VW stress, with the former initially greater than the latter. As VW stress intensified, the nonphysiological impacts increased and eventually surpassed the physiological effects, for the typical bottom-up damage characteristic of VW. The joint regulation of physiological and nonphysiological information was also detected in the case of water stress, which is slow and sustained, similar to VW stress. The prominent response of physiological information was detected at the early stages of water stress on sugar beet plants [20]. The importance of nonphysiological information increased to a similar value as the physiological in determining SIF variations when data were collected in a potato crop two weeks after the presence of water exclusion, for dynamic changes in leaf angular distributions in response to water stress [22]. However, the mechanism of herbicide stress is different. It can induce a rapid and well-characterized impact on the photosynthetic apparatus by inhibiting the photosynthetic electron transport from PSII to photosystem I (PSI) [59]. Therefore, the importance of physiological information to SIF variations increased rapidly and significantly within just several hours after herbicide treatment. After the early stage of treatment, it gradually decreased and eventually stabilized at values similar to those before the treatment within approximately two weeks [29]. When plants are subjected to different types of stress, they respond adaptively to these stresses by adjusting their morphology, physiology, and other means [60], [61], [62]. Thus, when analyzing the relative importance results, it is important to deeply comprehend the mechanism of stress impact on plants.

C. Influence of Random Error on Relative Importance Results

According to Fig. 7, we found that the relative change caused by Round7 and GS_001 was relatively small among the five types of errors, but they still have the potential to alter the trend of result. If the standard deviation of Gaussian random noise was enlarged to 0.05, the relative change resulting from human error would be unacceptable and even worse when the standard deviation was further increased. Our results indicated that, although human error could make the LMG model work, it introduced uncertainty into results. It is worth noting that some errors might be introduced into data during the period of observing and processing data. During the observing period, some errors might be caused by environmental condition, such as poor weather, unstable illumination, and large solar zenith angle, which will influence the quality of observed data [63], [64]. As some research reported, variations of SIFyield (SIF/PAR/fPAR) were entirely overshadowed by the noise present in the spectral measurements and the retrieved values at the canopy scale [65]. In addition, during data processing, such as rounding retention, variations in default data accuracy across different programming languages may introduce errors in data. In order to obtain reliable and accurate results, it is important to be cautious of factors that may potentially introduce errors in the observed data.

D. Applications and Prospects

SIF has been proven as a quick and powerful tool for detecting vegetation stress. However, the observed canopy SIF is influenced by various factors during the radiative transfer process, leading to uncertainties in its direct relationship with vegetation photosynthesis. This greatly impacts the understanding and application of SIF in response to disease stress. It is of great importance to disentangle the physiological and nonphysiological components of canopy SIF to precisely depict the impact of stress on vegetation during the onset and progression of disease stress. Our results indicated that SIF variations were initially regulated by physiological information and, subsequently, by nonphysiological information as the impact of VW stress intensified, enhancing our comprehension of change process in cotton plants under VW stress. These findings are of great significance for accurately monitoring the degree of stress on cotton, as well as for the comprehensive prevention and control of cotton VW.

In future studies, it would be beneficial to incorporate more comprehensive experimental data, including photosynthesis and chlorophyll at leaf level, and structural parameters (leaf area index and leaf inclination angle distribution) at canopy level to provide more comprehensive and more intuitive insight into changes of cotton plants. More specific canopy structure data provide a more detailed description of fPAR and f_esc individually, which although is considered as the combined effect of nonphysiological information in this study. From the physical mechanism, the combination of fPAR and f_esc would absorb the changes in each of them. For example, the decrease in leaf area index results in a decrease in fPAR and an increase in f_esc [20]. Therefore, it is necessary to analyze fPAR and f_esc separately for better depicting nonphysiological impacts of VW stress. Furthermore, if actual measurement data are lacking or difficult to acquire, it is advisable to utilize radiative transfer model simulation results.

A number of studies have proved the capacity of SIF in estimating GPP [18], [66], [67], and it is beneficial for cotton yield estimation under VW stress to understand SIF variations better and more comprehensively. With the fast development of technology, these findings may contribute to estimating cotton yield at larger spatial scale, such as at field or county level based on the data acquired from UAV or satellite-based sensors.

SECTION V. Conclusion

In this study, we conducted continuous observation on both healthy and stressed cotton plants during the peak incidence period of VW stress. We compared and analyzed the changes in the physiological, nonphysiological factors, and SIF of the cotton canopy under different degrees of VW stress, followed by an exploration of the factors contributing to the reduction of canopy SIF. Results indicated that the regulatory role of SIF components in its variation varied under different levels of VW stress, reflecting the severity of VW. At the early stages of VW development, the changes in canopy SIF were significantly affected by physiological factors with maximum contribution close to 0.7, and the diurnal pattern exhibited an arch shape. As the VW severity increased, the contribution of physiological factors declined greatly, with the maximum contribution decreasing by 47.7%, and nonphysiological factors played a more substantial role in the changes of canopy SIF. In addition, the arch shape of physiological factors' diurnal change pattern was gradually destroyed, with the diurnal variation amplitude declining by 63.4%. Besides, we proposed a practical strategy for estimating ΦF when calculating the relative importance of parameters to SIF variations using the LMG model. This study helps to further understand the impact of VW stress on cotton plants, enhancing our comprehension of each SIF component's contribution to SIF variations in cotton plants under varying VW stress severity, thereby contributing to the accurate diagnosis of VW stress severity.

ACKNOWLEDGMENT

The authors would like to thank the Director C. You from the Shihezi Institute of Agricultural Sciences for the support in coordinating the experimental field.

Appendix

Fig. 9.

Daily variations in nonphysiological factors (NIRv, FCVI, and RENDVI), SIF, SIF_PAR, and physiological factors (ΦF) during the experimental period. Each column represents VW2, VW5, and VW6.

Fig. 10.

Results of diurnal changes in the relative importance of each subperiod. The x-axis coordinates have been converted to local time in Xinjiang. Each row corresponds to subperiods T1, T2, T3, and T4 and each column represents VW2, VW5, and VW6.