This study was conducted over three growing seasons in the Southern Hemisphere (from September to March, between 2017 and 2020). The application of the RDI strategy and tree measurements were taken in the post-harvest period (January to March) in a commercial drip-irrigated cherry tree (Prunus avium L.) orchard located in Comalle, Curicó, Chile. (34.82° LS; 71.29° LW; 325 m.a.s.l.). The study site has a semiarid, warm, and suprathermal climate, with annual temperatures ranging from 5.2°C to 29.7°C (Santibáñez 2017). The soil is classified as Comalle series (fine, mixed, thermic Aquultic Haploxeralfs) (CIREN, 1997), with a volumetric soil water content (θ) of 40 % at field capacity (FC) and 28 % at wilting point (WP) at a depth of 1.0 m (Carrasco-Benavides et al., 2020a). The planting density was 3.0 m between trees per 4.5 m between rows (741 trees ha-1), oriented north to south. The height of the trees’ canopy was around 3.2 m by summer pruning using a Solaxe canopy training system. They were irrigated using two drip irrigation polyethylene pipes with in-line emitters per tree (4 L h-1 tree-1), spaced at 0.50 m (8,888 drippers ha-1).
Experimental design and treatments
The experimental design employed was a complete randomized design, comprising a row of 48 adult trees (planted in 2008) within a 2.4-hectare (ha) orchard of 'Regina' cherry trees grafted on 'Sour cherries' (Prunus cereasus L.).The leaf area index (LAI), estimated by the cover photography method, averaged 2.23 ± 0.55 m2 m-2 (Carrasco-Benavides et al., 2020a). Three post-harvest water stress-recovery treatments were applied during each growing season, guided by specific midday Ψs thresholds as follows: T0 (fruit grower management treatment, or control) (Ψs > -1.0 MPa, without-to-low water stress); T1 (low to mild water stress treatment = -1.0 > Ψs > -1.5 MPa); and T2 (mild-to-severe water stress treatment = -1.5 > Ψs > -2.0 MPa). These thresholds were selected based on previous cherry tree studies (Blanco et al., 2018; Carrasco-Benavides et al., 2020; Marsal et al., 2010; Podestá et al., 2010). Each experimental unit with a given treatment consisted of four trees (numbered from 1 to 4). Only the central trees (2 and 3) were measured, while the extreme (1 and 4) were used as isolation borders. Throughout the experiment, irrigation frequency and timing followed the standard practices recommended by the fruit grower. Irrigation scheduling for the orchard was planned weekly or biweekly, with guidance from the fruit manager and an external irrigation consulting service.The original dripper lines were replaced with new ones (H5 PC, Rainbird Co., Azuza, CA, USA) installed in the experimental plot, considering two pipelines per tree. For T0 trees, the dripper lines were installed without valves, following the fruit grower's irrigation practices. In contrast, plastic valves (indented 3157, Palaplast, Sindos, Greece) were inserted at the beginning of the drip lines for T1 and T2 trees to allow manual control of water flows based on Ψs measurements. Values of Ψs of all treatments were measured once or twice weekly, one day before each irrigation cycle, to maintain all treatments within the desired thresholds. The potential effects of irrigation from adjacent rows on the experimental plot were considered negligible, as no surface runoff was observed in the inter-row space. This assumption was confirmed by visual inspection, which showed that the wetted bulb zone of each emitter was restricted to its own tree row.
Meteorological measurements
An automatic weather station (AWS) (VANTAGE PRO2, Davis Instruments, Hayward, CA, USA), located 1.9 km from the experimental plot, provided the daily records of meteorological variables. The meteorological variables measured were solar radiation (Rs), air temperature (Ta), relative humidity (RH), wind speed (Ws) and direction (Wdir), and rainfall. The AWS data were used to compute the daily vapor pressure deficit (VPD) and short crop reference evapotranspiration (ETo), according to the Penman-Monteith model suggested by the FAO-56 manual (Allen et al., 1998). Additionally, a portable AWS was installed in the adjacent row of the experimental area during each post-harvest period. During the experiment, this second AWS was used to measure the micrometeorological variables in the cherry tree orchard. The data logger (Em 50, METER Group, Inc., Pullman, WA, USA) was used to store the averaged meteorological data from the experiment at 30-minute intervals. The sensors were connected to this datalogger as follows: Two Ta and RH probes (VP-4, METER Group, Inc., Pullman, WA, USA), placed at 1.5 m in the middle of the canopy and at 1.0 m above the top of tree canopies, respectively; a combined Ws and Wdir sensor (Wind, Davis Instruments, Hayward, CA, USA) at 2 m height in the inter-row space. One sensor of rainfall (ECRN-50, METER Group, Inc., Pullman, WA, USA), at 1.5 m height in the inter-row space, and a silicon pyranometer sensor to measure Rs (PYR, METER Group, Inc., Pullman, WA, USA), placed at 1.5 m above the top of the canopy. At the end of each growing season, the second AWS was removed and reinstalledin the next post-harvest period to protect the sensors from damage caused by routine during the pre-harvest period.
Field campaigns for PI measurements (Ψs, gs) and TIR images shot
Weekly field campaigns were conducted before each irrigation cycle, on daylight days around noontime (1100 – 1500 h at local time). At that time, Ψs were measured using a pressure chamber (model 600, PMS instruments, Albany, OR, USA). The samples consisted of two healthy leaves per tree from the middle of the canopy. At least one hour before the measurement, leaves were wrapped in plastic and aluminum (Carrasco-Benavides et al., 2020b). In parallel, two mature, healthy, and fully developed leaves were used to measure gs with a portable leaf porometer (SC-1, METER Group, Inc., Pullman, WA, USA).
Thermal images of cherry tree canopies were obtained simultaneously with Ψs and gs measurements (in a range of 2 to 4 min) to ensure no changes in those parameters, as Jones (1999) recommended. The thermal images were shot from the side of the trees, perpendicular to the canopies, at ≈ a 3.5 m distance using a thermal camera (TIS60, Fluke Corporation, Everett, WA, USA). This camera had a 260 × 195-pixel resolution, 7.5–14 µm spectral range sensitivity, and an HFOV/VFOV of 35.7°/ 26.8°, respectively. Similarly to García-Tejero et al., (2018), the camera was calibrated at a factory, considering an air emissivity (εa) of 0.95 and a background temperature of 25 °C. At each image, the two extreme points, Twet and Tdry, were artificially generated by painting (1-2 min before each camera shot) two representative healthy and non-detached mature leaves on both sides (adaxial and abaxial) with water mixed with dishwashing soap (0.01 percent v/v) and liquid petroleum jelly (Fuentes et al., 2012). According to Jones et al., (2002), this is a reliable technique to effectively provide reference temperatures reflecting the radiometric and aerodynamic properties of the actual leaves being studied.
Each resulting TIR image was exported into a comma-separated values (CSV) file using SmartView (version 4.3.79.0, Fluke Corporation, Everett, WA, USA). These CSV files were post-processed using a script developed in Python (version 3.9, Python Software Foundation. Available at http://www.python.org) by implementing the different formulations of CWSI and Ig (Eqs. 1 – 5) (Fuentes et al., 2012; Jackson et al., 1981; Jones 1999; Jones et al., 2002; Idso et al., 1981; Poirier-Pocovi and Bailey 2020; Poirier-Pocovi et al., 2020). Pixels with non-leaf material, such as branches, stems, trunks, soil, and sky, were automatically removed from each image because they were outside the Twet and Tdry range.
where Tc, Twet, and Tdry are the canopy temperature, the full-transpiring leaf temperature (wet-bulb temperature), and the non-transpiring leaf temperature (dry-bulb temperature), respectively (all in °C); all of the versions of CWSI and Ig are dimensionless. Theoretically, Eq. 1 represents a non-normalized index considering that Tdry > Tc; thus, a CWSI1 = 0 represents a water-stressed plant. This index gradually increases as the water stress reduces (Poirier-Pocovi and Bailey 2020). In the case of Eq. 2, CWSI2 linearly moves from 1 for well-watered plants to 0 for water-stressed plants. Eq. 3 presents a non-normalized index ≥ 0, where CWSI3 linearly increases as the plant dries out. From Eq. 4, since Tdry ≥ Tc ≥ Twet, the resulting CWSI4 = 0 indicates that the plant is transpiring at full capacity and not showing signs of water stress. On the contrary, a CWSI4 = 1 means that the plant is water-stressed. Finally, Ig (Eq. 5) is a non-normalized to the unity index related to gs. Ig = 0 represents non-transpiring leaves (plants under water stress), whose values increase as the water stress decreases.
Statistical analysis
The statistical analysis considered first the Analysis of Variance (ANOVA) among the treatments for each physiological indicator, including the effect of the growing season and their interactions. When the differences among the treatments from ANOVA were significant, Tuckeys's Honest Significant Differences (HSD) test was applied (α > 0.05). Then, to evaluate the effects of the different treatments on the other PI, considering the possible abrupt changes in the slope of the linear relations among them, the piecewise linear regression (PLR) was used (Toms and Lesperance, 2003). In this broken-stick model, two lines with various slopes combine at a breakpoint. Applying this method allows for objectively identifying the threshold changes (Boys et al., 2016) of the different relations among the different PI, considering water stress in each treatment. Finally, the TIR-image-based PI values were linearly compared using the Pearson correlation coefficient (r) linear regression approach to determine the degree of correlation between pairs of thermal-based PI. All Figures and data analyses were made using the R Core Team (2022) and the RStudio software ("Elsbeth Geranium" Release, RStudio Inc., Boston, MA, USA). We used the generic functions and the packages' ggplot2' (Wickham 2016), 'tidyverse' (Wickham et al., 2019), ‘devtools’ (Wickham et al., 2022), ‘PerformanceAnalytics’ (Peterson et al., 2020), and 'segmented' (Muggeo, 2008).