A physiological signal derived from sun-induced chlorophyll fluorescence quantifies crop physiological response to environmental stresses in the U.S. Corn Belt

Data was collected from three platforms at different temporal and spatial scales (table 1). The plot-scale dataset was collected by a ground-based portable system in a warming experiment for soybean (Kimm et al 2021). The field-scale dataset was collected by ground-based stationary systems at adjacent irrigated and rainfed corn-soybean rotation fields. The regional-scale dataset was collected by MODIS and TROPOMI space-borne platforms covering the U.S. Corn Belt.

Table 1. Description of the datasets used in this study.

	Platform	Target	Stress	Spatial scale (footprint)	Temporal scale	Reference
	Ground-based portable system	Warming experiment plots	Temperature	<1 m radius	Daily (discrete)	Kimm et al (2021)
	Ground-based stationary systems	Irrigated vs rainfed crop fields	VPD	<5 m radius	Half-hourly (continuous)	Miao et al (2020)
	Space-borne system	U.S. Corn Belt	Temperature and VPD	0.05° × 0.05°	Daily (continuous)	Köhler et al (2018)

2.1.1. Plot-scale data

2.1.2. Field-scale data

2.1.3. Regional-scale data

We applied an advanced framework for interpreting a SIF signal (Dechant et al 2020, Zeng et al 2020), which disaggregates it into non-physiological and physiological signals. Following thorough testing based on simulations, we used either NIRvP or soil-adjusted NIRvP (SANIRvP) as a proxy for non-physiological (i.e. canopy structure and radiation) signals included in SIF, and they are obtained as follows:

$$\begin{align}{\text{NDVI}} = \left( {{\rho _{{\text{NIR}}}} - {\rho _{{\text{Red}}}}} \right)/\left( {{\rho _{{\text{NIR}}}} - {\rho _{{\text{Red}}}}} \right)\end{align} \tag{ 1 }$$

$$\begin{align}{\text{NIRv}} = {\rho _{{\text{NIR}}}} \cdot {\text{NDVI}}\end{align} \tag{ 2 }$$

$$\begin{align}{\text{NIRvP}} = {\text{NIRv}} \cdot {\text{PAR}}\end{align} \tag{ 3 }$$

$$\begin{align}{\text{SANIRvP}} = {\text{NIR}}{{\text{v}}_{{\text{max}}}} \cdot \frac{{{\text{NIRv}} - {\text{NIR}}{{\text{v}}_{{\text{min}}}}}}{{{\text{NIR}}{{\text{v}}_{{\text{max}}}} - {\text{NIR}}{{\text{v}}_{{\text{min}}}}}} \cdot {\text{PAR}}\end{align} \tag{ 4 }$$

where NIRv is an approximation of near-infrared reflectance of vegetation, PAR is incident photosynthetically active radiation, and NIRvmax and NIRvmin are long-term maximum and minimum of NIRv. SANIRvP accounts for soil impact through normalizing NIRv as shown in equation (4), but as it requires long-term records for the reliable consideration of soil impact, it was only used for satellite datasets. Proximal remote sensing-derived NIRvP and satellite-based SANIRvP used different PAR units. NIRvP used instantaneous photon flux density (μmol photon m⁻² s⁻¹) whereas SANIRvP used daily radiation energy flux (MJ m⁻² d⁻¹) (Jiang et al 2020a). We then derived Φ_F using (SA)NIRvP following the below calculations:

$$\begin{align}{\text{SIF}} = {\text{PAR}} \times {\text{fAPA}}{{\text{R}}_{{\text{chl}}}} \times {{\text{f}}_{{\text{esc}}}} \times {\Phi _{\text{F}}}\end{align} \tag{ 5 }$$

$$\begin{align}{f_{{\text{esc}}}} = {\text{NIRv}}/{\text{fAPA}}{{\text{R}}_{{\text{chl}}}}\end{align} \tag{ 6 }$$

$$\begin{align}{\Phi _{\text{F}}} = {\text{SIF}}/{\text{NIRvP}}.\end{align} \tag{ 7 }$$

The derivation of Φ_F is based on the consideration of non-physiological impacts on the canopy-scale observation of SIF that involves multiple factors, i.e. incident radiation, leaf-to-canopy scaling parameters such as LAI and leaf angle distribution. As (SA)NIRvP accounts for the photosystem-to-canopy scaling, the ratio of SIF to (SA)NIRvP allows for estimating photosystem-scale fluorescence yield, which is linked to photochemical yield and stomatal conductance (equations (3), (5)–(7)) (Zeng et al 2020).

2.3.1. Data preprocessing

We preprocessed the data for our analyses to reduce variations of incoming radiation and canopy structure, which are non-physiological and directly associated with SIF, and to focus on the impacts of high temperature and high VPD on crop physiology. For the two ground-based sensing datasets, time of day and time of year are restricted to peak radiation hours and peak growth season. Particularly for the field-scale data, time of day was set corresponding to the overpass time of TROPOMI data, and the spectral data were aggregated at the daily timescale (table 2). For the TROPOMI data, we only selected the data from a three week period of the highest NIRv on a pixel basis in each year to minimize the pixel-to-pixel difference caused by different phenological stages and to focus on crop responses in their peak growth period. We obtained the three week period mean values and analyzed spatial patterns.

Table 2. Description of data preprocessing.

	Target	Spatial scale	Temporal scale	Time of day	Time of year
	Warming experiment plots	<1 m radius	Daily (discrete)	10 AM–3 PM	20 July–15 September → 20 July–4 September
	Irrigated vs rainfed crop fields	<5 m radius	Half-hourly (continuous) → Daily	9 AM–6 PM → 10 AM–2 PM	May–October → July–August
	U.S. Corn Belt	0.25° × 0.25°	Daily (continuous)	10 AM–2 PM	January–December → Peak three week period

2.3.2. Relative importance analysis

To quantify relative importance, we calculated the relative sum of squares from analysis of variance (ANOVA), i.e. a sum of square for each term (${\text{S}}{{\text{S}}_{{\text{NIRvP}}}}$, ${\text{S}}{{\text{S}}_{{\text{SANIRvP}}}}$ or ${\text{S}}{{\text{S}}_{{{{\Phi }}_{\text{F}}}}}$) divided by the total sum of square (${\text{S}}{{\text{S}}_{{\text{Total}}}}$). The ANOVA was conducted to the below regression model:

$$\begin{align}{\text{SIF}} = {c_1} + {c_2} \cdot {\text{NIRvP}} + {c_3} \cdot {\varphi _{\text{F}}}\end{align} \tag{ 8 }$$

where ${c_1}$, ${c_2}$, and ${c_3}$ are regression coefficients. To keep the consistency of our analysis across different datasets, we focused on temporal variability rather than spatial variability. ANOVA was conducted for each plot, each site, and each pixel, respectively, and then we averaged the quantified relative importance across different plots, sites, and pixels and obtained a 95% confidence interval from their variability (figures 2(a) and (c)).

2.3.3. Plot-scale data

2.3.4. Field-scale data

2.3.5. Regional-scale data

We first evaluated the importance of the non-physiological information for the spatiotemporal variation of all the available SIF data in each dataset. The determination coefficient (R²) of the relationship between SIF and (SA)NIRvP was >0.82 in the three datasets (figure 1). Less than 20% of the SIF variation was potentially attributable to plant physiological information. Considering the uncertainty of the measurements and SIF retrieval, this value maybe even less than 10%. To quantify and evaluate clearer physiological signals from SIF, we controlled temporal variations in radiation and seasonality of canopy development. Before and after the preprocessing, we disaggregated SIF into (SA)NIRv and Φ_F that represent a non-physiological and physiological signal, respectively, and quantified the relative importance of each component from the three datasets (figure 2). Although (SA)NIRvP explained the majority of SIF variability in both cases, our relative importance results showed that a physiological signal of SIF occupied a significant portion of SIF variability after the preprocessing and indicated potential applications of SIF-derived physiological signals. With all data included, the relative importance of (SA)NIRvP explained 64%–91% of SIF variability, and Φ_F explained 0%–20% (5%, 20%, and 0% for the plot-scale, field-scale, and regional-scale datasets, respectively, figure 2). After preprocessing the data, (SA)NIRvP explained less (60%–82%) and Φ_F explained more of SIF variation (28%, 31%, and 17%, respectively for the plot-scale, field-scale, and regional-scale dataset). In both cases, the plot-scale and field-scale datasets showed greater importance of Φ_F than the satellite dataset.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For each of the three different datasets, we tested whether (if so, to what extent) the SIF-derived physiological signals indicate physiological stress to high-temperature and water deficit stress. First, linear regression analysis of the plot-scale dataset showed that both NIRvP and Φ_F were significantly correlated with GCVI and canopy temperature (Tc), and NIRvP-GCVI and Φ_F-Tc showed the greatest association suggesting that NIRvP and Φ_F contained information on non-physiological variation and physiological stress, respectively (correlation coefficient for NIRvP-GCVI: 0.47, NIRvP-Tc: −0.28, Φ_F-GCVI: 0.35, and Φ_F-Tc: −0.50, p< 0.001 for all cases) (figures 3(a)–(d)). In partial correlation analysis, including GCVI and canopy temperature as an explanatory variable, collinearity between the two variables was accounted for, and we evaluated more strictly whether Φ_F were indicative of physiological stress. We found from the partial correlation results that the SIF-GCVI correlation was largely attributed to NIRvP (partial correlation r_p for NIRvP-GCVI: 0.4, p< 0.001, and for Φ_F-GCVI: 0.16, p> 0.01), while the SIF-canopy temperature correlation was entirely attributed to Φ_F (r_p for Φ_F-Tc: −0.4, p< 0.001, and for NIRvP-Tc: not significant, p> 0.1), which showed even greater sensitivity to canopy temperature than SIF (r_p = −0.28, p < 0.001) (figure 3(e)).

Zoom In Zoom Out Reset image size

Second, the field-scale data included different water-deficit levels by different management practices (i.e. irrigation or rainfed). We first compared SIF, NIRvP, and Φ_F between the two fields and found a statistically significant difference only in Φ_F (p < 0.001), which showed an 18% relative difference (two sites' difference normalized by the value of the irrigated site) (figure 4(a)). The smaller Φ_F in the rainfed site suggested water deficit stress of the crops, and the lack of SIF difference indicates that Φ_F could have a better ability to detect stress than SIF. Further, both Gs and Φ_F showed smaller values in the rainfed site, which supported stronger stomatal regulation and photosynthetic depression caused by water deficit. The comparison between Gs and Φ_F showed a positive linear relationship only in the rainfed site (figure 4(b)). The discovered positive linear relationship between Gs and Φ_F indicated that the observed Φ_F difference resulted from the physiological stress by water deficit in the rainfed site.

Zoom In Zoom Out Reset image size

Third, we analyzed the regional-scale data, i.e. TROPOMI SIF, MODIS-based SANIRvP, and SIF-derived Φ_F. All three variables negatively responded to VPD, and Φ_F showed a greater response to VPD than SANIRvP and SIF (figure 5(d)). For air temperature, the three variables showed a similar pattern with weaker responses than to VPD (figure 5(d)). Only SIF and Φ_F showed a negative response to air temperature, and again, Φ_F showed greater sensitivity than SIF. Considering the findings of the two ground-based sensing datasets, the relationship of satellite-derived Φ_F with air temperature and VPD suggested that Φ_F variation could be indicative of plant physiological responses. The correlation between SIF and environmental variables showed that SIF may capture plant responses in terms of both canopy structure and plant physiology, but Φ_F separately quantified physiological signals and showed its strength for stress detection through greater sensitivity than SIF.

Zoom In Zoom Out Reset image size

We first showed a high correlation between SIF and (SA)NIRvP (figure 1), which is consistent with the dominant role of structural information in explaining SIF from previous studies, e.g. SIF was highly correlated with APAR (Miao et al 2018, 2020, Yang et al, 2018a) and NIRvP (Baldocchi et al 2020, Dechant et al 2020). The agreement between SIF and NIRvP was predictable considering the pronounced importance of diurnal and seasonal variation of radiation and seasonality of canopy development in SIF. The relative importance of (SA)NIRvP and Φ_F in this study, however, further showed that data preprocessing that controlled temporal variation in radiation and canopy structure amplified Φ_F signals and allowed for investigating Φ_F and plant physiological responses. With all data included, the relative importance of Φ_F was no greater than 20% (5%, 20%, and 0% for the plot-scale, field-scale, and regional-scale datasets, respectively). The highest importance found in the field-scale dataset could be due to the water deficit-induced physiological variability. Meanwhile, the plot-scale dataset showed smaller importance of Φ_F although it also included a large physiological variability due to the warming treatments. The different results of the plot-scale and field-scale datasets might be the coupling of NIRvP and Φ_F. In the plot-scale dataset, the continuity of the warming treatments affected both canopy structure and plant physiology whereas the field-scale dataset showed decoupling of NIRvP and Φ_F (figure 4(a)). After the data preprocessing, the importance of Φ_F increased to 17%–31% in the three datasets (figure 2). Specifically, the importance of Φ_F was greater in the plot-scale and the field-scale datasets than the regional-scale dataset (28%, 31%, and 17%, respectively for the plot-scale, field-scale, and regional-scale dataset). A probable reason is stress factors included in the two ground-based datasets (i.e. high-temperature stress and water deficit stress in the plot-scale and field-scale data, respectively).

We found from the three datasets that the quantified Φ_F contribution to the SIF variation after the data preprocessing was relevant to plant physiological responses. Chlorophyll fluorescence measurements including SIF have long been used to test high-temperature and water stress impacts at the canopy scale (Ač et al 2015). However, SIF-derived Φ_F has been only recently evaluated, and in the majority of the previous canopy-scale studies, a lack of explicit consideration of canopy structural impact might have influenced the interpretations of SIF or SIF yield (i.e. SIF divided by APAR). The data of this study showed that SIF-derived Φ_F allowed for quantifying plant physiological variation and contributed to quantifying plant physiological responses under stress conditions.

Our plot-scale data from the canopy warming experiment revealed that NIRvP indicated a non-physiological impact, and SIF-derived Φ_F was indicative of physiological stress with even greater sensitivity than SIF (figure 5(a)), which is consistent with our expectations based on the theory behind the Φ_F derivation. Meanwhile, the field-scale data showed a positive linear relationship between Gs and Φ_F, which is consistent with a leaf-level finding (Flexas et al 2002). Even the irrigated site also showed positive linearity between Gs and Φ_F, which suggests not only soil moisture but also VPD regulated crop physiology (Kimm et al 2020a). A recent study, on the other hand, reported asymmetry between Φ_F and Gs (and photochemical quantum yield, Φ_P) responses, which disagrees with our finding (Magney et al 2020, Marrs et al 2020). However, the asymmetry was found under artificial treatments, i.e. abscisic acid application and pressure cuff-induced xylem embolism, while our result was based on less artificial abiotic stress impacts, and target plants' stress coping mechanism might have differed under artificial treatments.

The relationship between Gs and Φ_F of our analysis suggests that Φ_F is indicative of plant physiological responses because, under the water deficit conditions of our measurements, Gs behavior suggests physiological stress and further, Gs could be mechanistically connected and synchronized with Φ_P. The relation between Φ_P and Gs might be partially explained by mechanisms related to photosynthetic processes. First, there have been investigations on the direct connection between stomatal behavior and photosynthetic electron transfer chain (Sharkey and Raschke 1981, Messinger et al 2006, Busch 2014, Głowacka et al 2018). The connection between the two components has not been fully understood, but changes in light, i.e. different spectral intensity or modulations in photosystem II, were found to induce stomatal behavioral changes suggesting a direct connection. Second, optimality theory may explain the link between stomatal behavior and light reaction. Stomatal behavior was often understood from a perspective of optimization between carbon gain and water loss (Katul et al 2012), and photosynthetic processes were also seen from an optimality perspective utilizing the balance between rubisco carbon fixation and photosynthetic electron transfer rate (Jiang et al 2020b). A rate of photosynthesis involves multiple processes including stomatal behavior, and to optimize resource and energy use, plants are expected to balance between the processes resulting in the linearity between Gs and Φ_P.

The SIF-derived Φ_F application also showed the potential with the satellite data analysis and supported the large-scale applicability of Φ_F for SIF-based physiological investigations. Existing studies of satellite-based SIF have found that SIF showed stronger and earlier plant responses to droughts and heat stress than conventional vegetation indices (Sun et al 2015, Song et al 2018, Li et al 2020a). However, what the SIF responses actually suggested remained unclear, e.g. how much of SIF responses were due to the physiological responses or non-physiological variability. Our analysis showed that SIF responses to temperature and VPD were attributed more to Φ_F than SANIRvP, indicating physiological stress than canopy structural depressions. Especially, temperature impact on SIF was largely attributed to Φ_F while almost no response was found in SANIRvP (figure 5(f)), and we found greater sensitivity of Φ_F to climate variables than SIF. The aforementioned relationship of Φ_F with temperature and VPD showed that, despite the large uncertainty, the derived Φ_F suggested plant physiological responses. Its sensitivity to climate variables indicates Φ_F could be more useful than SIF in detecting physiological stresses possibly because Φ_F is unassociated with canopy structural variation, unlike SIF.

In this study, we quantified SIF-derived Φ_F and demonstrated its capability for physiological investigations in crops. Our results addressed the unique contribution of SIF to remote sensing of crops and showed the potential of SIF for large-scale crop stress quantification. Previous crop SIF studies, however, reported that physiological signals may have a marginal contribution to SIF variations (Miao et al 2018, Yang et al 2018a, Dechant et al 2020), and particularly, studies that investigated SIF-derived Φ_F found the physiological signal relatively constant throughout their datasets (Dechant et al 2020, Liu et al 2020). We speculate that the different findings about Φ_F were due to two major aspects, i.e. if physiological stresses existed and how tightly physiological stress was coupled with canopy structural variation. First, crop physiological stress tends to be well managed through agricultural management practices such as irrigation and fertilization application (Peng et al 2020), and previous crop SIF studies observed crops without stress, and therefore, observed no clear signals of physiological variation. This study, on the other hand, included data of crops with and without high-temperature or high-VPD conditions, and by comparing their difference, we captured strong physiological down-regulation of crop productivity. Second, crop physiological status is largely coupled with canopy structure, and crop productivity largely depends on canopy structure (Wu et al 2019, Liu et al 2020). The opposite case, for example, is evergreen ecosystems where canopy structural variation and physiological status are largely decoupled and SIF can quantify physiological variations better than in other ecosystems (Walther et al 2016, Magney et al 2019, Kim et al 2021). In this study, we processed the data constraining variations in radiation and canopy structure to decouple structure and physiology, which allowed the extraction of signals that are related to physiological variations.