Assessing the Water-Stress Baselines by Thermal Imaging for Irrigation Management in Almond Plantations under Water Scarcity Conditions

: This work examines the use of thermal imaging to determine the crop water status in young almond trees under sustained deﬁcit irrigation strategies (SDIs). The research was carried out during two seasons (2018–2019) in three cultivars ( Prunus dulcis Mill., cvs. Guara, Lauranne, and Marta) subjected to three irrigation treatments: a full irrigation treatment (FI) at 100% of irrigation requirements (IR), and two SDIs that received 75% and 65% of the IR, respectively. Crop water monitoring was done by measurements of canopy temperature, leaf water potential ( Ψ leaf ), and stomatal conductance. Thermal readings were used to deﬁne the non-water-stress baselines (NWSB) and water-stress baselines (WSB) for each treatment and cultivar. According to our ﬁndings, Ψ leaf was the most responsive parameter to reﬂect di ﬀ erences in almond water status. In addition, NWSB and WSB allowed the determination of the crop water-stress index (CWSI) and the increment of canopy temperature (IT C ) for each SDI treatment, obtaining threshold values of CWSI (0.12–0.15) and IT C (~1 ◦ C) that would ensure maximum water savings by minimizing the e ﬀ ects on yield. The ﬁndings highlight the importance of determining the di ﬀ erent NWSB and WSB for di ﬀ erent almond cultivars and its potential use for proper irrigation scheduling.


Introduction
Irrigation performs an essential role in agriculture. As such, the increase in total irrigated area, coupled with scarce water resources, has encouraged the implementation of irrigation strategies that optimize the water-use efficiency. Specifically, in areas such as southern Spain, this supply is crucial for the proper development of woody crops, when the maximum evapotranspiration rates coincide with the rainfall absence. Considering the current scenarios of climatic change and water scarcity, the adaptation and sustainable strategies to boost the proper water management in irrigated crops is vital [1]. Among them, deficit irrigation (DI) has been implemented to enhance the yield, reducing the irrigation supplies and maximizing the crop productivity [2]. According to this, the implementation value, using the VPD of that particular day [14]. In this line, one step further towards to DI programming would be to obtain the most appropriate WSB, which would correspond to that obtained for the treatment, and ensure the maximum water saving and minimum yield loss. Moreover, this WSB would allow defining the threshold IT C , providing the advisable value for the maximum deviation from the NWSB.
Taking these points into consideration, the objectives of this study were (i) to determine the NWSB for three studied almond cultivars during the kernel-filling period; (ii) to define the WSB for two different water-stress levels; and (iii) to establish a protocol to manage the irrigation scheduling by means of these functions and its relation with the yield.

Experimental Site
The experiment was conducted during the kernel-filling and postharvest period (June to September) in two consecutive years (2018 and 2019), in a commercial almond (P. dulcis Mill. cvs. Guara, Marta, and Lauranne) orchard, grafted onto GN15 rootstock, and located in the Guadalquivir river basin (SW Spain, 37 • 30 27.4" N, 5 • 55 48.7" W). Trees were planted in 2013, spaced 8 m × 6 m, and drip irrigated using two pipelines with emitters of 2.3 L × h −1 , spaced at 0.75 m intervals. Canopy volumes were very similar within each cultivar, without differences between irrigation treatments. Thus, for Marta, canopy volumes ranged between 64 and 65 m 3 ; Guara trees, between 65 and 66 m 3 ; and Lauranne, trees between 72 and 74 m 3 .
The soil was a silty loam, a typical Fluvisol, more than 2 m deep, with organic matter <1.5%. Roots were located predominately in the first 50 cm of the soil, corresponding to the intended wetting depth. Soil water content values at field capacity (−0.33 MPa) and permanent wilting point (−1.5 MPa) were close to 0.40 and 0.15 m 3 × m −3 , respectively.
The climatic classification of the study area was attenuated meso-Mediterranean with a hot-summer Mediterranean climate (csa) in the Köppen climate classification [30], with an annual ET 0 rate of 1400 mm, an average temperature of 18 • C, and accumulated rainfall of 540 mm (average data corresponding to the last 15 years (2004-2019); obtained from the Andalusian Weather information Network).

Irrigation Treatments
Three irrigation treatments were designed as follows: (i) a fully irrigated treatment (FI), which received 100% of the irrigation requirements (IR) during the whole irrigation period; (ii) a sustained deficit irrigation (SDI 75 ) treatment, which received 75% of the IR; and (iii) a sustained deficit irrigation (SDI 65 ), which received 65% of the IR.
In both seasons, irrigation was applied from the middle of March to the end of October, and these doses were calculated according to the methodology proposed by Allen et al. [31] (Equations (1) and (2)); obtaining the values of reference evapotranspiration (ET 0 ) by using a weather station installed in the same experimental orchard (Davis Advance Pro2, Davis Instruments, Valencia, Spain).
where ET C is the crop evapotranspiration; K C is the single-crop coefficient; Kr is the crop reduction coefficient, which depends on the percentage of shaded area cast by the tree canopy; ET 0 is the reference evapotranspiration; and IR is the irrigation requirements. The local monthly K C and K r used during the experimental period are shown in Table 1, as was determined by García-Tejero [32]. Additionally, the IR was reduced for SDI 75 and SDI 65 by multiplying it by 0.75 and 0.65, respectively.

Plant Measurements
During the kernel-filling period (162-225 days of the year (DOY) in 2018; and 162-217 DOY in 2019), crop water monitoring was done throughout the measurements of the leaf water potential (Ψ leaf ), stomatal conductance, water vapor (g s ), and canopy temperature (T C ); all these readings were taken between 12:00 and 13:30 GTM, and with a periodicity of 7-10 days.
The g s was measured using a porometer SC-1 (Decagon Devices, INC, WA, USA) in two leaves per tree (monitoring 8 trees per irrigation treatment) fully developed, and completely exposed to the sun, with the aim of monitoring the maximum values of g s and detecting the most detectable differences among the irrigation treatments. The selected leaves were at 1.5 m of height, approximately, and were SE facing. On the other hand, the Ψ leaf was measured using a pressure chamber (Soil Moisture Equipment Corp., Sta. Barbara, CA, USA), monitoring two leaves per tree, located on the north side of the tree and being totally mature, fresh and shaded, with the aim of minimizing the measurements variability. Selected leaves were at 1.5 m of height, approximately, and NW facing.
Considering the results obtained by García-Tejero et al. [33], who reported that the best moment for assessing the T C was between 11:30 and 14:30, and in the sunny side of canopy, thermal images were taken following this procedure: using a ThermaCam (Flir SC660, Flir System, USA, 7-13 µm, 640 × 480 pixels) and an emissivity ( ) of 0.96 ( Figure 1). Readings were developed at the sunny side of the canopy, placing the camera at a 4 m distance from the monitored tree, approximately. Afterwards, images were analyzed using the Flir Research Pro Software (Flir System, USA), which allows to select different zones of the images (in our case; 4 different sunny areas per image were selected); each pixel corresponding to an effective temperature value [19].
Once the images were obtained, T C was calculated for each treatment, cultivar, and monitoring day, and after this, the thermal index ∆T canopy-air was calculated. Taking into consideration the ∆T canopy-air values and the VPD registered during the data acquisition, the NWSB and WSB were defined according to Equation (3); these functions corresponding to trees that were subjected to different irrigation doses, and allow to estimate the optimum values of ∆T canopy-air for each treatment depending on the VPD values [29].
where b and a are the intercept point and slope of the linear function. Additionally, taking as reference the NWSB obtained for each cultivar, the CWSI along the monitoring period for each DI treatment was estimated, according to Equations (4) and (5): where ∆T canopy-air corresponds to the canopy readings obtained in each treatment and cultivar; ∆T wet is the lower limit calculated from the NWSB equation in each cultivar; and ∆T dry is the upper limit obtained according to the methodology proposed by Idso et al. (1981).
where a and b are the slope and the interception point for the NWSB; e s (T air ) is the saturated vapor pressure at air temperature; and e s (T air + b) is the saturated vapor pressure at the sum of the air temperature and interception point.

Experimental Design and Statistical Analysis
The experimental design was of randomized blocks, with four replications per irrigation treatment and cultivar. Each replication had 12 trees (3 rows and 4 trees per row); the two central trees for each replication were monitored. Thus, eight trees per irrigation strategy treatment were used. Statistical analysis was done by using the Sigma Plot statistical software (version 12.5, Systat Software, Inc., San Jose, CA, USA). For each measurement day, an exploratory and descriptive analysis of the data (TC, Ψleaf, and gs) was developed, applying a Levene's test to check the variance homogeneity of the variables studied. Significant differences among irrigation treatments (p ≤ 0.05) were identified by applying a one-way ANOVA, and a Tukey's test to identify the significant differences. Additionally, there were defined the NWSB and WSB for each irrigation treatment and cultivar, analysing the differences by applying an ANCOVA to evaluate the differences in the interception points and slopes, and obtaining the threshold values of the CWSI and ITC for each cultivar that ensure minimum yield loss and the highest water saving. For this, at the end of each season, the effects on kernel yield in relation to irrigation treatments were analyzed by applying a one-way ANOVA, and a Tukey's test to identify the significant differences. Table 2 shows the climatic conditions during the two studied seasons. During the irrigation period (from April to October), the cumulative rainfall was 326 and 85 mm for 2018 and 2019, Figure 1. Thermal images at the sunny side of the studied almond canopies. FI, fully irrigated treatment; SDI 75 , sustained deficit irrigation treatment at 75% of the irrigation requirements; SDI 65 , sustained deficit irrigation treatment at 65% of the irrigation requirements; G, Guara; M, Marta; L, Lauranne.

Experimental Design and Statistical Analysis
The experimental design was of randomized blocks, with four replications per irrigation treatment and cultivar. Each replication had 12 trees (3 rows and 4 trees per row); the two central trees for each replication were monitored. Thus, eight trees per irrigation strategy treatment were used. Statistical analysis was done by using the Sigma Plot statistical software (version 12.5, Systat Software, Inc., San Jose, CA, USA). For each measurement day, an exploratory and descriptive analysis of the data (T C , Ψ leaf , and g s ) was developed, applying a Levene's test to check the variance homogeneity of the variables studied. Significant differences among irrigation treatments (p ≤ 0.05) were identified by applying a one-way ANOVA, and a Tukey's test to identify the significant differences. Additionally, there were defined the NWSB and WSB for each irrigation treatment and cultivar, analysing the differences by applying an ANCOVA to evaluate the differences in the interception points and slopes, and obtaining the threshold values of the CWSI and IT C for each cultivar that ensure minimum yield loss and the highest water saving. For this, at the end of each season, the effects on kernel yield in relation to irrigation treatments were analyzed by applying a one-way ANOVA, and a Tukey's test to identify the significant differences. Table 2 shows the climatic conditions during the two studied seasons. During the irrigation period (from April to October), the cumulative rainfall was 326 and 85 mm for 2018 and 2019, respectively. In relation to ET C , similar values for 2018 and 2019 (~880 and 840 mm respectively) were registered. This fact, together with the high differences in terms of rainfall, promoted that the irrigation doses applied in the studied treatments were much greater in the second experimental season. In this sense, FI, SDI 75 , and SDI 65 received 4974, 3713, and 3342 m 3 ·ha −1 , respectively, in 2018; and 7700, 5744 and 5159 m 3 ·ha −1 , respectively, in 2019. R, rainfall (mm); T max , maximum air temperature ( • C); T min , minimum air temperature ( • C); T av , average air temperature ( • C); RH max , maximum relative humidity (%); RH min, minimum relative humidity (%); RH av , average relative humidity (%); Rad, solar radiation (W·m −2 ); R, rainfall (mm); ET 0 , reference evapotranspiration (mm); ETc, crop evapotranspiration (mm). Table 3 displays the physiological response found for Ψ leaf , g s , and T C during 2018. The main significant differences among the irrigation treatments were detected for Ψ leaf . In this sense, cv. Marta showed differences at 190, 197, 211, 218, and 225 DOY. For the case of cv. Guara, these differences were detected at 166 and 225 DOY. Finally, regarding cv. Lauranne, significant differences were observed at 190, 197, and 211 DOY. For the remaining variables, only punctual days showed significant differences. Table 3. Temporal evolution of the physiological variables measured throughout 2018. A similar pattern was observed during the second experimental season as shown in Table 4, the Ψ leaf being the physiological parameter that displayed the most perceptible effects in response to the different irrigation treatments.

Water-Stress Baselines for Each Cultivar and Irrigation Treatment, and Their Interactions
Taking as reference the T air values registered during the data acquisition (Figure 2), the ∆T canopy-air was calculated, and afterwards, the relationships with measured VPD were defined for each treatment and cultivar, considering the whole dataset obtained in both experimental seasons ( Figure 3, Table 5).    As shown in Table 5, within each cultivar, the ANCOVA did not manifest differences in terms of the slope and the interception point for any of the studied cultivars. This absence of differences is in accordance with the previous results noted in relation to T C and g s , parameters without differences during the monitoring period. Moreover, this difference could be associated with the inherent variability of the experiment, especially, in T C readings. In this agreement, within each treatment it was observed Tc variations of ±0.5, ±0.9, and ±1.5 • C in the FI, SDI 75 , and SDI 65 , respectively. This variability was also higher the more remarkable the imposed water stress was. Moreover, a higher variability was found in cv. Guara while cvs. Marta and Lauranne showed lower and similar variability trends.
Considering this absence of differences between the irrigation treatments, there was defined a single WSB for each cultivar with the whole dataset (Table 6). These reference water-stress baselines (rWSB) would allow knowing an optimum ∆T canopy-air , establishing the lower and upper limits from the NWSB and WSB previously defined for the FI and SDI treatments (Figure 4).  According to our findings, the maximum IT C reported for each cultivar was~1.0 • C ( Figure 4); that is, the highest differences between the FI and SDI 65 strategies would report increases beyond the lower limit around a degree in the sunny side of the almond canopy. Moreover, this deviation would be different depending on the cultivar. For the case of cvs. Marta and Lauranne, the maximum IT C were detected in the lower ranges of the VPD that contrasts with cv. Guara.
Finally, taking into consideration the NWSB and WSB for each treatment and cultivar, and with the aim of establishing a useful threshold limit that ensures the maximum water savings, the CWSI on a monthly basis was estimated ( Figure 5). It is noticeable the progressive increase along the kernel-filling period, especially in cvs. Marta and Lauranne, displaying a progressive rise because of the water-stress accumulation. Moreover, in cv. Guara and cv. Marta, the SDI 65 reported a CWSI higher than those obtained in cv. Lauranne, where in the latter the SDI 75 registered similar values of the CWSI. For cv. Guara, the maximum CWSI was reached under SDI 65 , with values close to 0.14. Similar results were found for cv. Lauranne (~0.15), whereas in cv. Marta these values were somewhat lower, roughly 0.12.

Linking the Yield with Water-Stress Baselines Defined for Each Cultivar and Irrigation Treatment
After estimating the different WSBs, the final yield was analyzed for each irrigation treatment and cultivar ( Figure 6). This fact is necessary to define the threshold values canopy-air and CWSI to minimize the yield losses and maximize the water savings (in case of obtaining significant differences between irrigation treatments). On average, for cvs. Marta and Lauranne, no differences Figure 5. The crop water-stress index (CWSI) on a monthly basis for the water-stressed treatments (SDI) and studied cultivars. SDI 75 , irrigated at 75% of the irrigation requirements; SDI 65 , irrigated at 65% of the irrigation requirements. Vertical bars are standard deviation.

Linking the Yield with Water-Stress Baselines Defined for Each Cultivar and Irrigation Treatment
After estimating the different WSBs, the final yield was analyzed for each irrigation treatment and cultivar ( Figure 6). This fact is necessary to define the threshold values of ∆T canopy-air and CWSI to minimize the yield losses and maximize the water savings (in case of obtaining significant differences between irrigation treatments). On average, for cvs. Marta and Lauranne, no differences were observed, evidencing that water withholding close to 35% of the irrigation requirements would not promote yield losses, at least during two consecutive seasons. Something different was determined for cv. Guara. In this case, in spite of not finding significant differences, there was a trend between the yield loss and water stress imposed; that is, the obtained values for SDI 75 and SDI 65 were notably lower than those observed under FI, with yield reductions of 11% and 15%; respectively. Figure 5. The crop water-stress index (CWSI) on a monthly basis for the water-stressed treatments (SDI) and studied cultivars. SDI75, irrigated at 75% of the irrigation requirements; SDI65, irrigated at 65% of the irrigation requirements. Vertical bars are standard deviation.

Linking the Yield with Water-Stress Baselines Defined for Each Cultivar and Irrigation Treatment
After estimating the different WSBs, the final yield was analyzed for each irrigation treatment and cultivar ( Figure 6). This fact is necessary to define the threshold values canopy-air and CWSI to minimize the yield losses and maximize the water savings (in case of obtaining significant differences between irrigation treatments). On average, for cvs. Marta and Lauranne, no differences were observed, evidencing that water withholding close to 35% of the irrigation requirements would not promote yield losses, at least during two consecutive seasons. Something different was determined for cv. Guara. In this case, in spite of not finding significant differences, there was a trend between the yield loss and water stress imposed; that is, the obtained values for SDI75 and SDI65 were notably lower than those observed under FI, with yield reductions of 11% and 15%; respectively.

Discussion
The focus of this paper was to assess the use of thermal data as indicator of crop water status instead of discontinuous measurements, such as Ψleaf or gs, which are highly time-consuming with a huge number of measurements that are needed for taking decisions.
Considering the results showed in this work, the leaf was the parameter that showed the highest differences between treatments in the two-year experiment, relative to gs and TC (Tables 3 and 4). It is remarkable that the decreasing pattern in leaf was not followed by gs, likely because of the lower capacity of almond trees to regulate their stomata under mild water-stress situations [3,34]. These Figure 6. Average kernel yield for the studied almond cultivars during the study. FI, fully irrigated at 100% of irrigation requirements; SDI 75 , irrigated at 75% of the irrigation requirements; SDI 65 , irrigated at 65% of the irrigation requirements. Vertical bars are standard deviation.

Discussion
The focus of this paper was to assess the use of thermal data as indicator of crop water status instead of discontinuous measurements, such as Ψ leaf or g s , which are highly time-consuming with a huge number of measurements that are needed for taking decisions.
Considering the results showed in this work, the Ψ leaf was the parameter that showed the highest differences between treatments in the two-year experiment, relative to g s and T C (Tables 3 and 4). It is remarkable that the decreasing pattern in Ψ leaf was not followed by g s , likely because of the lower capacity of almond trees to regulate their stomata under mild water-stress situations [3,34]. These findings were in agreement with other works [35,36], showing that under mild stress, almond decreases Ψ leaf significantly more than g s , which remains fairly constant until severe water stress. As g s tightly controls plant transpiration, this, in turn, determines to a great extent the leaf temperature. The lack of significant differences in g s among the irrigation treatments and for none of the cultivars support why there were also no differences between T C and WSBL. In addition, plant transpiration, in which g s determines photosynthesis, in conjunction with turgor, is liable for growth and yield. Accordingly, fruit yield did not show relevant differences among the irrigation treatments for cvs. Marta and Lauranne, although these were more evident for cv. Guara. In accordance with our data and to previous works, it seems that to detect a higher response of g s to water stress it would be necessary to impose more severe water-stress conditions; then the stomatal response would be mainly governed by the crop water status [10,22].
The use of thermal data as indicator of crop water status has been implemented in different works to solve the drawback that Ψ leaf or g s carried out with their development [27,36]. In order to define the most proper strategy, many authors have discussed the best time to capture the images, the tree area or the time range to take the images. In this sense, González-Dugo et al. [37] concluded that, for the case of citrus trees, the best moment to capture the thermal images would be between 11:20 and 12:00. They also observed that the maximum differences between the control and stressed trees ranged between 1.5 and 2.5 • C. Their results agree with those obtained in this experiment, in which the maximum difference between the FI and SDI treatments is ±1 • C (Figure 4). In the same line, García-Tejero et al. [33] in an experiment with almond (cv. Guara) concluded that the best moment to capture thermal images was between 11:30 and 14:00 in the sunny exposed side of the tree, when the maximum differences of T C between the FI and DI treatments were reached. Therefore, these differences were always from 0.5 to 1.5 • C when a water restriction close to 50% of the irrigation requirements was imposed, similar to findings that was obtained in the present work.
Despite Tc not always having a direct relationship with Ψ leaf or g s , due to the large environmental variability, the use of different thermal indexes that normalize this parameter to the meteorological conditions make this tool suitable to determine the crop water status [24]. In this study the use of the index ∆T canopy-air allowed to establish the NWSB and WSB for three almond cultivars, adjusting these values with those of the VPD registered. In this context, Bellvert et al. [38] outlined that different WSB can be obtained, and their main differences could be associated with their intercept point; these differences being associated to variation in the crop water status [19,20] or the crop phenological stage. Similarly, García-Tejero et al. [28], for mature almond trees, reported differences in the interception point between different WSBs within a cultivar subjected to different irrigation doses. These results agree with that found in this work ( Figure 3, Table 5). In this line, although the ANCOVA did not evidence significant differences in the slope and interception point among the irrigation doses imposed in each cultivar, we observed maximum differences between the NWSB and WSB close to 1.0 • C, comparable to findings by García-Tejero et al. [28] or García-Tejero et al. [33]. The main differences among these results and those reported by the authors would be mainly in the slope of the functions calculated for the studied cultivars. Thus, González-Dugo et al. [37] or García-Tejero et al. [28] reported similar slopes for mature almond trees, cv. Guara, which were growing under similar climatic conditions. In our case, the obtained slopes were substantially different; this being an important fact to be considered in future works. Thus, this fact could be due to the tree age and this work being defined in young trees, whereas the previous works were developed in mature almond trees, in which the transpiration capacity could have substantially changed.
Authors such as Romero-Trigueros et al. [39] largely discussed the advantages of this type of functions when these are applied in isohydric crops, with a higher capacity of stomatal regulation when they are subjected to water withholding. This is not the case for almond, with a downregulation of stomatal conductance, resulting in similar T C values for trees subjected to different irrigation doses. Considering that no differences were found among the irrigation treatments, the rWSB defined for each cultivar would be a suitable option for irrigation scheduling under moderate scenarios of water scarcity, knowing that there were no differences in productive terms with water around 2000 m 3 × ha −1 ( Figure 6).
Finally, in spite of to the absence of significant differences in yield for the three studied cultivars, cv. Guara was affected with a progressive depletion in relation to the water stress imposed. Confronting these results with the maximum IT C registered, cv. Guara was the unique in which IT C increased for major values of VPD, and it could demonstrate a higher sensitivity to the SDI strategy than the remaining cultivars, especially when atmospheric demand is higher. Likewise, the absence of differences in terms of yield has been widely stated by several authors [10,[40][41][42] and, therefore, this reaction ratifies the advantages of this agronomic practices for almond cultivation in arid and semi-arid environments.

Conclusions
From the research that has been performed in this paper, it is possible to conclude that the ∆T canopy-air and its related thermal indexes (CWSI and IT C ) are precise indicators of the crop water status in young almond trees. In detail, the use of ∆T canopy-air to establish the NWSB and WSBs for different cultivars and water-stress levels would offer an optimum tool for irrigation management differentiated by cultivar and water restrictions. On the other hand, considering the three cultivars studied, cv. Guara offered a higher sensitivity to water stress, as in yield reductions in terms of its physiological response. Following the proposed methodology of this study, using thermal data, it would be possible to materialize other WSB for different cultivars and tree ages for alternative irrigation programming, especially when DI is used. However, taking into consideration that there were no differences found in yield between the water-stressed and non-stressed treatments, future essays imposing more severe water stress should be considered, in order to ensure obtaining the maximum threshold value (in terms of the CWSI or ITc) that would not significantly impact yield, explicitly under long-term irrigation periods.