Experimental design

This experiment was carried out in a greenhouse located in the Experimental Station of the Agricultural Water and Soils Laboratory of the Universidad del Valle, Cali, Colombia, where the variables ambient temperature and relative humidity were monitored (average temperature of 24.5°C, 79% relative humidity, and 30% poly shade). Young individuals of the plant species Heliconia psittacorum (He), not more than 0.15 m tall, were taken from a sub-surface wetland of horizontal flow from the was-tewater treatment plant of Ginebra-Valle, Colombia. The plants were acclimated for two months under saturated substrate conditions. The experimental design was completely randomized with one factor (irrigation). Three levels of the irrigation factor were established (100% Field Capacity (FC), 50% FC and 10% FC). The FC value of the substrate was estimated using the sand box method (^{Jaramillo, 2002}). Three treatments were performed with 4 replicates and 12 experimental units (EU). Each EU was an individual of the He species planted in 6 L pots with the substrate, which provided nutritional support, composed of soil with organic fertilizer, to which macro and micronutrients were added according to the requirements of nitrogen, phosphorus and potassium reported by ^{Sosa (2013)}. Tests of analysis of variances (ANOVA) and Honestly Significant Difference (HSD) of Tukey (P<0.05) were used prior to the normalization and verification of homoscedasticity to determine if there were significant differences between the treatments. The experiments began on November 20, 2016 and ended on January 27, 2017. Statistical tests were performed to evaluate significant differences between treatments on the days shown in Table 1.

TABLE 1 Start and end dates of experiments and days chosen to carry out hypothesis tests. 

* Substrate moisture was restored to FC for T2.

Irrigation control

The substrate moisture of each EU was measured using time domain refractometry (TDR) equipment (MPM160 ICT-International, Armidale-Autralia) three times per week at a distance equal to the effective root depth (0.07 m). Each measurement was taken three times in random points, and the average value was recorded. According to the volumetric moisture value of each EU, the moisture of the substrate was adjusted according to each treatment. The FC value of the substrate was estimated using the sand box method (^{Jaramillo, 2002}). All the EUs started at FC; treatments T2 and T3 were drained until reaching the respective depletion level for each group, while T1 was maintained at FC throughout the experiment.

Measurement of spectral reflectance

The leaf tissue reflectance (%) of each EU was measured using an STS-VIS Mini-Spectrometer (Ocean Optics®, Florida, USA). The equipment has a measurement range between 350 and 800 nm, with an integration time of 100 ms and a spectral resolution of 1 nm. The measurements were taken at a distance of 0.05 m from the leaf tissue, between 10:00 am and 12:00 midday. One measurement was taken per leaf in the central part and to one side of the midrib. The calibration was done using the reflectance blank prior to each measurement. A medium low pass filter was used to eliminate high frequency and noise components before the calculation of the spectral indices. The Normalized Difference Vegetation Index (NDVI) and the Variation of NDVI in the range 705-750 nm (NDVI705) spectral indices were estimated using Equations 1 and 2, respectively. The maximum value of the first derivative of the reflectance curve (Dmax) between 700 and 750 nm (Eq. 3) was also calculated:

where R650, R705, R750 and R800 are the percentages of light reflected with respect to the calibration blank at wavelengths 650, 705, 750 and 800 nm, respectively (%), and is the first derivative of the reflectance with respect to the wavelength (% nm^-1).

Measurement of thermal indices

The CWSI and GI values were calculated according to the model described by ^{Leinonen and Jones (2004)} (Eq. 4 and 5).

where T_leaf is the temperature of the leaf (°C), T_wet is the reference temperature of wet tissue (°C) and T_dry is the reference temperature of dry tissue (°C). The temperature measurement was taken with thermographs of each EU, using the FLIR E40® camera (Oregon, USA), with a thermal sensitivity of 0.07°C at 25°C, 160x120 pixels resolution, and a spectral range from 7.5 to 13 μηι. The images were taken at a height of 1.5 m directly above each EU. Two reference temperatures were used: T _dry (dry tissue) and T _wet (wet tissue). T_dry was obtained by taking 5°C above air temperature (^{Rud et al., 2014}; ^{Mangus et al., 2016}). The value of the reference temperature T _wet was measured using the most frequent value of the histogram of the thermography in wet leaf tissue. The thermal information of the images (Raw-Data) was exported in the .csv format using the FlirTools© (Oregon, USA). The data were then processed for the calculation of the thermal indices in free software developed by the main author in C++ using image processing libraries from OpenCV2.0.

Soil moisture

All EU started with the volumetric moisture of the substrate at FC (76% volume content). The substrate moisture was monitored and adjusted three times per week until it reached the values of each treatment T1, T2 and T3. Figure 1 shows the variation of the average substrate moisture (volume-basis) throughout the experiment. On d 19, the moisture of the T2 treatment was restored to 100% FC in order to determine if the changes in the thermal and spectral response of the EUs corresponded to the variation in the irrigation.

FIGURE 1 Variation of substrate moisture (volume-basis) for 3 treatments: T1 (100% FC), T2 (50% FC) and T3 (10% FC). 

Spectral reflectance and spectral indices

The leaf tissue reflectance of the EUs presented the typical response of plant tissue, in which the percentage reflected in the NIR band (around 800 nm) was higher than the percentage reflected in the visible band (400-700 nm). However, it was found that the NIR reflectance for the T2 and T3 treatments was lower than for the T1 treatment. Changes in morphology of the leaf and the structure of the mesophyll have a strong influence on the spectral properties of leaves in this range (^{Wójtowicz et al., 2016}). Decrease in reflectance in the near infrared could occur because of the structural modification of mesophyll, which is possibly related to a strong water deficit (^{Semenova et al., 2014}). Figure 2 shows the average of the reflectance curves for each treatment at d 14 of the experiment. These results are in agreement with those obtained in other species subjected to a water deficit. In a study with Zea mays L., ^{Genc et al. (2013)} found similar results for four treatments (100% FC, 66% FC, 33% FC and 0% FC). In the species Spinacia oleracea under water stress, ^{Corti et al. (2017)} reported the same effect of a decrease on NIR reflectance measured on green leaves.

FIGURE 2 Average reflectance for treatments T1 (100% FC), T2 (50% FC) and T3 (10% FC) at d 14 of the experiment. 

No significant differences were found (P>0.05) between any of the treatments for the NDVI and NDVI705 vegetation indices (Tab. 2); ^{Shimada et al. (2012)} reported that there were no significant differences between the NDVI index and the water potential of the species Hibiscus rosasinensis subjected to water stress. ^{Elvanidi et al. (2017)} conducted an experiment to study the water status of Solanum lycopersicum in an environment with a controlled microclimate. They found no significant differences in treatments using the NDVI index. However, the normalized difference indices showed greater sensitivity in changes generated by the structure at the canopy level (^{Duan et al., 2017}), especially when the leaf area index was low and in the face of lengthy follow-up times (^{Gamon et al., 2013}; ^{Morgounov et al., 2014}).

The vibrational modes of water molecules have associated wavelengths of 1,405 nm (2v1) and 1,896 nm (v1+v2); some weak vibrations are associated with the bands 1,150 nm (v1+3v2) and 960 nm (v1+4v2), the latter being the weakest (^{Yang et al., 2011}; ^{Steidle et al., 2017}). These wavelengths are all outside the range of the spectrometer used in this study. As such, it was not possible to directly associate the spectral indices in this range with the water status of the plants. However, one effect water stress produces is the inhibition of the synthesis of chlorophyll (^{Lisar et al., 2012}; ^{Etesami and Jeong, 2018}), a pigment that does have optical activity in the visible spectrum.

Red edge variations

The treatments with a water deficit showed a decrease in the slope between 700 and 750 nm. As a consequence, the maximum value of the first derivative of the reflectance curve (Eq. 3) decreased for T2 and T3. Figure 3 shows the average behavior for d 14 of the measurements in the three treatments.

FIGURE 3 First derivative curve of reflectance treatments T1 (100% FC), T2 (50% FC) and T3 (10% FC). 

Significant differences were found between the T1 treatment and the T2 and T3 treatments from d 9 of the measurements for values with a significance of P<0.05 (Tab. 2). The inhibition in chlorophyll synthesis in plants subjected to water stress can explain the statistically significant difference in the spectral response in the 700-750 nm range between the T1 treatment and the T2 and T3 treatments. Authors such as ^{Mielke et al. (2012)}, ^{Dian et al. (2016)}, and ^{Liu et al. (2016)} have considered spectral indices in the red edge band as good indicators of the leaf chlorophyll content of different plant species. ^{Yang et al. (2015)} established a correlation model for chlorophyll content that considers the maximum of the first derivative of the reflectance in the range 670-780 nm as one of its variables.

TABLE 2 ANOVA and Tukey test results for GI, CWSI, Dmax, NDVI, NDVI705, between the T1 treatment and the T2 and T3 treatments. 

*Significant difference between treatments P≤0.05.

Leaf temperature and thermal indices

The temperature of the leaf tissue was found to increase for treatments with a water deficit (T2 and T3). Figure 4 shows the leaf tissue temperature of the EUs in the three treatments on d 14 of the measurements. The data from 1 to 4 correspond to the EUs of T1, data 5 to 8 to T2, and data 9 to 12 to T3.

FIGURE 4 Leaf temperature of the EUs in treatments T1 (100% FC), T2 (50% FC) and T3 (10% FC). 

The mean temperature difference between treatments T1 and T3 was 2.2°C, for which there were significant differences. A similar result was presented by ^{Farifteh et al. (2013)}, who found that the difference between the control group and the treatment with induced water stress was 2.7°C after 22 d, a time period in which significant differences were found between the treatments. The increase in leaf temperature of the treatments with a water deficit may correspond to stomatal closure, a decrease in gas exchange and a reduction in transpiration, which occur in response to low water availability (^{Lisar et al., 2012}; ^{Etesami and Jeong, 2018}). All these phenomena, in principle, contribute to a lower water circulation and contribute to a process of raising the temperature of the tissue since the rate of heat transmission to the environment is reduced. Authors such as ^{Lima et al. (2016)} correlated measurements of stomatal conductance (gs) and transpiration (E) with increased leaf temperature. The above is valid if it is assumed that leaf temperature is not strongly influenced by variations in wind speed, resistance to heat flow and vapor pressure deficit. In this way, the stomatal resistance model described by ^{Jones (1999)} has an inverse relationship with leaf temperature, which would explain the relationship between leaf temperature and a possible stomatal closure resulting from water stress.

The T _wet reference was calculated using the most frequent value of the histogram in the region of interest (ROI). Figure 5 presents the ROI for measurement of the T_wet and its histogram. The circle shows the most frequent value of the histogram of the ROI temperature field. Similarly, the leaf temperature of the EUs was estimated with the temperature image.

FIGURE 5 T _wet reference temperature using the most frequent value of the thermography histogram. 

The CWSI and GI showed significant differences between the T1 treatment and the T2 and T3 treatments. After restoring irrigation to FC for T2, no significant differences were obtained between T1 and T2, which shows that the thermal response of the leaf tissue of the EUs was influenced by the irrigation levels (Tab. 2). Figure 6 presents the estimated thermal indices for the thermographs taken for the three treatments on the 14th d of experiment.

FIGURE 6 CWSI and GI for treatments T1 (100% FC), T2 (50% FC) and T3 (10% FC). 

The CWSI and GI made it possible to obtain greater statistical differentiation between the T1 treatment and the T2-T3 treatments than the NDVI, NDVI705 and Dmax spectral indices studied in the 350-800 nm range (Fig. 7).

FIGURE 7 Statistical diagram of the CWSI, GI, Dmax, NDVI and NDVI705 

Figure 8B shows the histograms associated with the ROIs of the thermographs shown in Figure 8A, taken on experiment d 14. Information was obtained on the thermal indices of the plant species for each pixel associated with the leaf tissue of the EUs using Equations 4 and 5, whose gray level is proportional to the value of the index in each pixel (Fig. 8 C and D).

FIGURE 8 Thermal response of He to induced water stress in treatments T1 (100% FC), T2 (50% FC) and T3 (10% FC). A) Gray scale image of the temperature of the experimental units, B) Thermography histogram, C) Gray scale image of the CWSI thermal index, and D) Gray scale image of the GI thermal index. 

From the image analysis, it was found that the higher intensity values in the CWSI images and the lower intensity values in the GI images were associated with the treatments with a water deficit. Similar results were found by ^{Rud et al. (2014)}, ^{Gómez-Bellot et al. (2015)}, and ^{Bellvert et al. (2014)}. The EUs of the T2 treatment, once they were taken to field capacity, had higher stomatal conductance indices and lower water stress indices than those seen when there was a water deficit (Fig. 9).

FIGURE 9 CWSI and GI thermal indices of two experimental units of treatment T2 before (d 14) and after (d 20) restoring the moisture of the substrate to field capacity. 

It was also found that the response of the plants to a water deficit was not uniform. It was observed that the EUs under a water deficit showed an increase in leaf temperature, an increase in the CWSI and a decrease in the GI, initially in the central part and then spreading to the rest of the leaf. Figure 10 shows an EU with a water deficit and non-uniform temperature field, in which the histogram did not have a single defined maximum. Figure 11 shows a control EU and uniform temperature field. The associated histogram had a single maximum in the area of the leaf. The evidence shows that, under water stress, plants have an anisotropic warming in the tissue of their leaves, which, as mentioned above, would be related to the characteristic of thermal instability when there is a low rate of evapotranspiration and stomatal closure as a physiological state of response to water stress.

FIGURE 10 Thermal image and ROI histogram. Plant under water stress, non-uniform temperature field. 

FIGURE 11 Thermal image and ROI histogram. Plant before water stress, uniform temperature field. 

Faced with an increase in temperature, the plants showed temporary wilting as a natural adaptation mechanism, decreasing the effective area and the temperature in the leaf tissue (Fig. 12).

FIGURE 12 Response of the He plant to the excessive temperature increase. 

The CWSI and GI made it possible to differentiate the EUs at FC and those subjected to a water deficit because of the fact that they have a relationship with the thermoregulation affected by variations in transpiration and gs, which, in turn, is associated with the water status of the plant. ^{Bellvert et al. (2016)}, ^{Egea et al. (2017)} and ^{Santesteban et al. (2017)} found a relationship between thermal indices and gs. Figure 13 shows the average value of the thermal indices between d 5 and 20 for each treatment. For d 19, the moisture was restored to FC for T2.

FIGURE 13 Average value of CWSI and GI for treatments T1, T2 and T3. 

The maximum values of the CWSI index were associated with the treatments with a water deficit, which is related to the tendency of leaf temperature to approach the reference temperature, inhibiting the transpiration process. On the other hand, the maximum values associated with the GI were related to the EUs at FC. This corresponds, according to Equation 4, to the tendency of leaf temperature to approach the reference temperature, favoring the process of perspiration. The CWSI went from a mean value of 0.9 for d 18 to 0.45 for d 20, after resetting the moisture of the EUs of T2 to FC. Likewise, the GI went from an average value of 0.1 to 1.2, showing that restoring irrigation in the EUs of T2 favored stomatal conductance and decreased water stress in the plants.