1. Introduction
In the context of climate change and an increasing global water scarcity, water deficit stress (normally shortened to water stress) is one of the most critical abiotic stressors to plant growth. In order to increase the quantity and quality of food production using a reduced amount of water, the detection and quantification of plant stresses is of major interest for agriculture in general and precision farming in particular [
1]. Remote sensing provides powerful tools in different spectral domains for spatiotemporal monitoring of water stress [
2].
Plant response to water stress is expressed by a variety of physiological changes (e.g., stomatal behaviour, and leaf water content), biophysical changes (energy balance, leaf and canopy structure, and biomass and yield) as well as photochemical processes [
3,
4,
5]. Accordingly, various attempts have been made to detect these changes using remotely sensed signatures. To date, the main remote sensing imaging techniques for plant water stress detection are thermal imaging (TIR; 8–14 µm), visible, near- and shortwave infrared reflectance (VNIR/SWIR; 0.4–2.5 µm), and sun-induced fluorescence (SIF; 0.69 and 0.76 µm).
TIR imaging has been well studied for water stress detection, e.g., [
6,
7,
8]. The underlying principle is that plant temperature rises with increasing water stress in comparison to a well-watered plant due to decreasing evaporative cooling through stomatal closing [
9]. Since stomatal closure is one of the first responses to water stress, plant temperature as measured by TIR sensors can be used to detect water stress pre-visually [
7,
10,
11]. Plant temperature, however, is not solely governed by the plant water supply but also by the actual micro-meteorological conditions (i.e., solar radiation, wind speed, leaf boundary layer resistance and vapour pressure deficit (VPD)). Therefore, as an alternative to using absolute temperature, several temperature-based indices were developed over the last few decades with the aim of compensating for varying meteorological conditions. For example, the prominent Crop Water Stress Index (CWSI) [
12] by means of artificial references does not require any additional meteorological measurements to be calculated. The usefulness of temperature-based indices has recently been demonstrated in several airborne studies [
13,
14,
15,
16]. Currently, the main limitation of all temperature-based (i.e., absolute temperature and temperature-based indices) approaches arises from the use of broadband infrared cameras with erroneous temperature retrieval by assuming an emissivity value that is pre-determined (e.g., ε = 0.97 for vegetation) and constant over the spectral range from 8 to 14 µm. For example, an error of 1% emissivity results in an absolute temperature error of 1 K [
7]. However, very recently hyperspectral TIR airborne imagers such as Telops
Hyper-
Cam LW, Itres
TASI-
600, or Specim
AisaOWL have become available. These devices allow for stable temperature and emissivity separation (TES) and very accurate temperature retrieval (i.e., <0.5 K) by measuring the emitted radiation in many narrow bands [
17]. The importance of airborne (hyperspectral) TIR remote sensing lies in the possibility of bridging the gap between ground-based thermography e.g., [
11,
18] and proposed satellite missions, such as ECOSTRESS (Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station, [
19]), HyspIRI (Hyperspectral Infrared Imager, [
20]) or the hyperspectral mission concept HiTeSEM (High-resolution Temperature and Spectral Emissivity Mapping, [
21]).
In the VNIR/SWIR domain, various spectral vegetation indices (VI) have been developed, each for a specific purpose such as canopy water content (MSI; [
22]), greenness or fractional vegetation cover (NDVI; [
23]), or photosynthetic activity (PRI; [
24]). As these plant parameters are somehow related to plant water status (e.g., photosynthetic activity, chlorophyll and water content may decrease under water stress conditions [
3,
25,
26]), these VIs may serve as indicators of water stress [
13,
14,
15]. Among the large variety of available indices, the PRI (photochemical reflectance index; [
24]) is related to non-photochemical heat dissipation, which may be linked to water availability and photosynthetic efficiency. In fact, PRI is heavily influenced by canopy effects and pigment content, but Suarez et al. [
27,
28,
29] suggest that the use of PRI in combination with radiative transfer modelling, accounting for these effects, can effectively provide very good means for monitoring water stress in crops and natural vegetation at airborne level. However, the ability of the PRI to be used for water stress detection is not conclusive at a small-scale experimental ground and airborne level [
13,
14,
18].
At the same time, remote sensing of sun-induced fluorescence (SIF) has become increasingly popular over the last decade with a variety of studies ranging from ground-based experiments to airborne campaigns and even towards satellite missions (see Meroni et al. [
30] for a comprehensive review). In principle, radiative energy absorbed by leaf chlorophyll is processed along three competing pathways: (i) conversion of photosynthetic active radiation (PAR) to sugars through photosynthesis, (ii) the re-emission of non-used energy through chlorophyll fluorescence or (iii) dissipation of heat [
31]. Hence, these three processes are in competition with each other; variation in one of the processes affects the others [
32]. For example, plant photosynthetic efficiency is reduced under environmental stress conditions (i.e., water or nutrient shortage, heat stress, and other types of stress) due to plants protective mechanisms (e.g., leaf rolling reduces the plants surface and therefore the PAR (Photosynthetically Active Radiation) absorbance; stomatal closure reduces water loss and CO
2 uptake). In this context, SIF is expected to be a direct indicator of photosynthetic efficiency and plant stress, although further studies are needed to establish a consistent basis for robust assessment [
31,
33]. The
HyPlant sensor facilitates sub-nanometre airborne acquisitions from the red (0.68 µm) towards the far red (0.78 µm) spectral range. Current experiments using the
HyPlant sensor demonstrated the capability of quantitative plant stress detection at airborne level using both red and far red chlorophyll fluorescence peaks [
32].
These preceding studies suggest that plant water stress symptoms may be detectable by means of the three approaches (TIR temperature indices, VNIR/SWIR VIs and SIF indices). However, since they are based on different models and physiological processes, the sensitivity and suitability of these approaches may vary depending on the target and environmental conditions. Therefore, to examine the detectability of water stress symptoms by the three approaches, we conducted a field-scale experiment specifically designed to tentatively evaluate the potential of state-of-the-art remote sensing at airborne level. Hyper-Cam LW as well as HyPlant sensors were flown over a commercial grass farm. Water stress symptoms were simulated by treating grass surfaces with two different chemical agents and comparing them to untreated Control (CR) plots.
The first agent, the anti-transpirant Vapor Gard
® (VG; Miller Chemical and Fertilizer, Hanover, PA 17331, USA), is composed of di-1-
p-menthene, a natural terpene polymer. The emulsion surrounds the leaves with a thin film and is supposed to reduce the plants water transpiration by limiting stomatal conductance. Thus, VG is recommended for farming to reduce water loss and prevent water stress in times of limited water availability [
34,
35,
36]. Because of decreasing stomatal conductance, transpiration is also reduced and consequently plant surface temperature is expected to be significantly increased in comparison to a CR plot. Furthermore, VG does not only have an effect on the permeability of water but also reduces CO
2 uptake rates [
37]. The second agent, kaolin (KA) is a highly reflective white powder and can be dissolved in water and sprayed on plants. The white color of KA increases the albedo of the plant surface and therefore reduces the absorbed light energy (APAR).
Thus, we assume that VG treated plants (in comparison to CR plants) have: (i) decreased transpiration, (ii) increased leaf temperature, and (iii) decreased photosynthetic activity, while keeping relatively high leaf water and chlorophyll content (in case of enough soil moisture). We assume further that KA treated plants (in comparison to CR plants) have: (iv) decreased absorbed radiation resulting in decreased transpiration under energy-limited conditions (in case of sufficient soil water), but also (v) little change in stomatal conductance, leaf temperature, photosynthetic efficiency, leaf water and chlorophyll content.
However, it is still unknown how well they allow the detection of plant water stress symptoms at airborne level. Thus, the overall aim of this study was to evaluate the capability of selected TIR and VNIR/SWIR hyperspectral remote sensing indices, as well as SIF remote sensing for detection of plant water stress symptoms. The specific objectives were: (i) examining if different temperature-based indices (e.g., CWSI, Ts, Ts–Tair), traditional VNIR/SWIR indices (e.g., PRI, NDVI, MSI), and SIF indices reveal differences over grass plots treated with chemical agents (VG and KA); (ii) an assessment of why the different indices may or may not have changed from treatments with respect to the underlying physiological processes; (iii) specifically examining diurnal changes in temperature-based index values regarding different treatments.
4. Discussion
As shown by the results of the meteorological measurements (i.e., SWC, VPD;
Figure 2) no actual plant water stress occurred during the experiment due to sufficient water supply from the soil. Nevertheless, the VG treatment had a modest but clearly measurable effect on H
2O flux relative to PAR (reduction of relative transpiration of grass by 20% on average) as well as on CO
2 flux (reduction by nearly 21%). This proved that real, but rather mild effects on transpiration and CO
2 uptake arose from the VG treatment.
The diurnal change of
Ts can be explained by the effects of plant transpiration. At late morning, as a consequence of increasing net radiation and VPD (
Figure 2), plant transpiration and thus, evaporative cooling increased and prevented a rise in
Ts. Due to the same process,
Ts–
Tair also increased from early to late morning. However,
Ts–
Tair slightly decreased from late morning to midday (
Figure 5) because
Ts remained constant due to higher transpiration rate while
Tair continued to rise due to increased net radiation (
Figure 2). Furthermore, the diurnal changes in CWSI demonstrated the same effect as described for
Ts–
Tair. Since the upper boundary (
Tdry) was determined by adding 5 K to the current
Tair,
Tdry increased in the same manner as
Tair resulting in overall lower CWSI values at midday. Negative CWSI values in KA resulted from the fact that
Ts was lowest for KA and the lower boundary (
Twet) was determined by the coolest 5% of CR plots, which were partly warmer than KA.
Indeed, the mild physiological manipulations through the chemical agents (i.e., 20% reduced transpiration in VG) only induced very small effects to temperature-based indices. VG reduced plant transpiration and induced symptoms of water stress, i.e., an overall increase in
Ts compared to CR plots. Contrary, the KA treatment highly increased the plant albedo, and thus reduced the overall energy uptake by the plant causing a decrease in
Ts. In this study, only absolute temperature differences of less than 1.0 K were observed between the treatments. Nevertheless, these minor differences could be distinguished by TIR remote sensing, as observed by a distinct pattern for all temperature-based indices (
Figure 5). These results demonstrate that TIR remote sensing indices were sensitive to small
Ts differences induced by the chemical treatments. In comparison, a recent study has demonstrated that temperature-based indices can detect water stress from airborne multispectral TIR data under distinct water deficit conditions with large temperature difference of up to 4.7 K [
13,
14]. However, our results indicate that the TIR-based indices can be used to detect minor or moderate level of water stress. It is well known that the leaf temperature can be increased by various stressors including water deficit and diseases [
64,
65]. Therefore, TIR imaging of high spatial and spectral resolution greatly contributes to the detection and/or quantification of the physiological anomaly caused by abiotic and biotic stresses.
Basically, canopy reflectance spectra in VNIR/SWIR are affected by water content, pigment content and composition (chlorophyll, carotenoid, xanthophyll), leaf internal structure, and canopy geometry e.g., Reference [
66]. Thus, in turn, the stress symptoms are detected via the spectral change associated to the response of these physiological traits to environmental stressors. In this study, we assumed little change in water and chlorophyll content as well as in structure among CR, KA and VG treatments. The response of all analysed VIs (PRI, NDVI, WI, LWI, MSI) that are closely related to either structure, water and pigment, or physiological functioning showed little difference between CR and VG. This fact suggests the consistency of these VIs in various conditions. On the other hand, the significant differences found between CR and KA imply that care should be taken in the application of these VIs to some specific leaf surface conditions (e.g., after yellow-sand dust and volcanic ash fall). In such cases, the surface reflectance spectra could be altered (as in KA treatment) without any physiological changes. Generally, the applicability of normalised VIs is based on the consistent relationship of multiple bands to the internal physiological changes. Accordingly, alteration of the relationship between spectral bands (i.e., shape of spectra) caused by external factors would strongly hamper the consistency of VIs. In fact, KA treatment increased plant albedo, but did not increase reflectance at the same ratio in all wavebands. Thus, the significant difference in KA was not caused by physiological changes, but rather indicated abnormal alteration of reflectance spectra. Since reflectance spectra is altered unusually by plant diseases such as powdery mildew and rust diseases e.g., Reference [
67], similar unusual changes of VIs as in KA would be detected by hyperspectral measurement. Therefore, our results from the unique experimental design provide useful insights on the consistency and usefulness of hyperspectral reflectance measurements, especially for discrimination of causes of detected changes.
According to previous research [
30], SIF decreases with a reduction of photosynthesis under high light and stressed conditions (e.g., shortage in water availability). However, in our experiment, SIF indices showed no significant difference between VG and CR treatments while the photosynthetic efficiency should have been lowered in VG plots because of reduced H
2O and CO
2 fluxes (as proved by chamber flux measurements). This fact suggests that either the stress was too subtle to reduce photosynthesis, and thus had no influence on the efficiency of photosynthetic light reaction or that the used SIF indices were not sufficiently sensitive to detect the slight photosynthesis reduction. On the other hand, the significant reduction in SIF indices (F
687, F
760) found under KA treatment is proportionally related to the reduced amount of APAR (Absorbed Photosynthetically Active Radiation) by increasing surface reflectance, and thus lowered the photosynthetic rate. Nevertheless, it is not clear whether the reduction of SIF was directly related to the change of photosynthetic photon use efficiency because the fraction of APAR (fAPAR) as well as water and chlorophyll content remained unchanged. In addition, one has to be careful about appropriate conclusions regarding the attenuation of SIF and incident PAR at leaf surfaces. It is still not obvious how SIF can be related to photochemistry and identifying a unique relationship between SIF and photochemical efficiency is challenging [
31]. The slightly negative values in F
687 originate from uncertainties in the F
687 retrieval due to the largely modified surface reflectance by the KA treatment. However, our results clearly show that the subtle change in SIF indices for KA treatments was detected by the
HyPlant sensor. Therefore, optical hyperspectral techniques bear the ability to detect both rapid photochemical change (e.g., photon use efficiency) and integrated physiological change (e.g., photosynthetic capacity), and to identify the causes for the changes by using multiple wavelengths. Since remotely sensed SIF is affected by the non-linear interactions of photochemical, physiological and biophysical factors, further experimental studies in combination with process-based modelling e.g., Reference [
68], are needed for both scientific and industrial applications.
In fact, the interpretation of the results presented here with regard to real water stress occurrence in natural conditions is limited due to: (i) logistical challenges related to synchronous flight plan of both airborne sensor systems HyPlant and HyperCam-LW which complicated the comparison of TIR versus VNIR/SWIR and SIF water stress indices, and (ii) instead of a real water stress occurrence only water stress symptoms were simulated by the chemical agents (i.e., VG and KA). Nevertheless, the study allowed us to deduce new findings about the sensitivity and the interplay of different indices in relation to water stress symptoms as induced by chemical agents at the airborne level.
In general, biotic and abiotic plant stressors (e.g., water stress, heat stress, and diseases) often occur simultaneously and cause similar plant physiological responses (e.g., increase in plant temperature, reduced photosynthetic efficiency, and change in canopy structure). Therefore, multi-sensor and multi–temporal approaches have a great potential to obtain useful information not only about the current plant status but also on the causes of biophysical, physiological and photochemical changes. In particular, airborne remote sensing with its high spatial and temporal resolutions can bridge the gap between in situ and satellite observations.