The experiment was carried out in a 0.37 ha commercial Prunus domestica L. (plum) orchard located in the “Werder fruit production” area in Brandenburg, Germany (52° 28′ 1.56″ N, 12° 57′ 28.8″ E). The soil is typical for fruit production in temperate climate of Europe and Asia formed by glacial and post-glacial deposits after the last ice age about 10,000 years ago with typically small scale variability. The cultivar was “Tophit plus” with “Jojo” serving as a pollinator. One hundred and four 7 year old “Tophit plus” trees, located every 4 m in 4 rows spaced 5 m apart, were considered. On average, trees were 2.10 m tall and insertion height of the first branch varied between 0.46 and 0.96 m above the soil. Mean soil texture was 45% sand, 29% silt, and 26% clay with a mean pH of 7.72. Plum trees were irrigated using a drip system with one line per row and two emitters every 0.5 m. The irrigation laterals and drippers were mounted 50 cm above the ground to facilitate mechanical weed control. Independent of precipitation, trees were irrigated twice a week for 1.5 h with flow rate of 0.96 L h⁻¹.

Global radiation, wind speed, air temperature, air pressure, precipitation, and relative humidity were measured at 24 min intervals by a weather station (UNIKLIMA vario, Toss, Germany) positioned 100 m from the experimental orchard. Canopy temperature and relative humidity (Modul DLTi, UP GmbH, Germany) were recorded in 18 trees every 5 min. Water vapor pressure deficit (VPD) of the air was calculated according to the Goff–Gratch-equation (Jones, 1992; von Willert et al., 1995) from hourly averages of air temperature, relative humidity, and air pressure.

A resistivity meter (4-point light hp, LGM, Germany) was used to map the ECa of the soil at the experimental site on 16th August 2012 and 2nd August 2013. The four electrodes were arranged in a Wenner array with the tree trunk in the center to obtain ECa values representing 25 cm depth (Telford et al., 1990). Full details are given in Käthner and Zude-Sasse (2015). Soil water matric potential (pf-meter 80, ecoTech Umwelt-Messsysteme GmbH, Germany) was measured at 15, 35, and 45 cm depths. In addition, the gravimetrical soil water content (GWC) was ascertained by drying soil samples at 105°C for 48 h with n = 26 in 2012 and n = 6 in 2013.

Three mature leaves were randomly detached from the north-eastern side of each tree and rapidly transported to the laboratory. Here, projected surface area [cm²] was measured for each leaf with a portable area meter (CI-203, CID Bio-Science, Inc., USA). Leaf dry mass [g] was consequently obtained after oven drying at 65°C for 24 h and specific leaf area (SLA) was calculated as the ratio of leaf area and dry mass.

In the orchard, leaf water potential (Ψ_leaf) was measured with a Scholander bomb (Plant Water Status Console 3000, Soilmoisture Equipment Corp., USA) on three shaded leaves from the lower part of the canopy on the east side of the tree. In 2012, 44 trees were analyzed predawn and midday over 4 days (19th June–27th June). In 2013, 67 trees were sampled over 5 days (19th July–2nd August). Following determination of Ψ_leaf, the leaves were rapidly packed in plastic bags, transported to the laboratory, frozen at −30°C. After thawing, centrifuged tissue sap was analyzed for osmotic content (c_osmol) with a water vapor osmometer (Vapro 5520, Wescor Inc., USA). The osmotic potential (Ψ_π) of tissue sap was calculated according to the van't Hoff's equation (von Willert et al., 1995).

Thermal images of the canopies were taken with an uncooled infrared thermal camera (ThermaCAM model SC 500, FLIR Systems, Inc., USA) with resolution of 320 × 240 pixel and spectral sensitivity range from 7.5 to 13.0 μm in the temperature range of −50 to 60°C on 15th August 2012 and 25th July 2013. The camera was mounted on a tractor with z = 3.3 m above ground and pointed to the top of the canopies. Images were acquired with an opening angle (β) of 45°resulting in the length (l) of the imaged area (Equation 1).

For extraction of temperature values, the raw thermal images were obtained in the FLIR systems' proprietary format and converted to text file format for the processing with MATLAB® (R2010B, MathWorks, USA). Crop water stress index (CWSI_J) was calculated (Equation 2) according to Jones (1992) ranging from 0 to 1:

where Tc is actual canopy temperature, Tw is temperature of a fully transpiring leaf with open stomata obtained from a wet paper leaf analog, and Td is temperature of a non-transpiring leaf. When using references Td_ref was obtained from a dry and Tw_ref from paper leaf analog (Jones, 2004). For this purpose, green paper leaves were cut to the formerly measured mean leaf area of 6 cm², mounted on a 2 m stick, and manually placed in the center of the canopy in each tree.

In addition, CWSI was also calculated according to three alternative methods. Irmak et al. (2000) calculated the CWSI_I (Equation 3) setting non-transpiring leaf temperature at 5°C higher than air temperature (Td+5) and Tw as the minimum temperature found in the canopy.

As described by work groups of Jones (1999) and Ben-Gal et al. (2009), Tw and Td was obtained analytically (Appendix in Supplementary Material) to calculate CWSI_JB (Equation 4).

CWSI_R was determined according to the work of Rud et al. (2015). Likely the most suitable for automated readings, this method calculates (Equation 5) the canopy temperature (Tc_histo) and reference temperatures of dry (Td_histo) and wet (Tw_histo) leaves from the histogram of pixels from each image itself.

In the CWSI_R approach, before processing histograms, extreme values above air temperature representing Fresnel reflection from the sun were removed from the further analysis. In the histogram of pixels, thresholds were determined for separating temperatures of soil, grass, and canopy. Dry reference, Td_histo, was defined as the minimum temperature of soil visible as a peak with high values in the histogram. Wet reference, Tw_histo, was taken as the minimum temperature of canopy. Since the canopy and grass partly coincided, pixels were spatially compared considering equal values as grass and varying values as canopy. This threshold was found with Wiener filter to enhance the contrast (Honig and Goldstein, 2002; Chen et al., 2006). After removing the soil and grass data, the Td_histo and mean canopy temperature (Tc_histo) were extracted and averaged for each tree.

On the day of CWSI measurement, a portable porometer (CIRAS-1, PP Systems, Hitchin, UK) was used to monitor the diurnal course (n = 3) of CO₂ exchange and transpiration. The instantaneous water use efficiency (WUE_i) was calculated as ratio of these parameters (von Willert et al., 1995) in μMol CO₂ m⁻² s⁻¹/mMol H₂O m⁻² s⁻¹.

The WUEc (Equation 6) of the production system represents the ratio of yield (y) and water volume supplied to the plants [g L⁻¹],

with i = irrigation water, pp = precipitation from the start of vegetation period until harvest time.

In 2012, the accounted period lasted from 17th April to 30th August during which 182 mm of irrigation and 273 mm rain with a total of 455 mm water were supplied. In 2013, irrigation water was given from 22nd April to 9th September accounting for 168 mm of irrigation water and 248 mm of rain was recorded summing up to 416 mm water supply.

Soluble solids content [%] of fruit was analyzed using a digital refractometer (DR 301-95, A. Krüss Optronic, Germany). Dry matter content of fruit [%] was calculated as the ratio of fruit dry mass and fruit fresh mass. Fruit flesh firmness [N cm⁻²] was analyzed as maximum force measured with a convex plunger at a velocity of 200 cm min⁻¹ (TA-XT Plus Texture Analyzer, Stable Micro Systems, UK). Fruit size measured as height [mm], fresh mass [g], and yield as number of fruits per tree and fresh mass per tree, was measured at harvest. In 2012, the analysis of fruit quality was carried out on all fruit of every tree, while in 2013, 3 fruits per tree were analyzed.

Statistical analyses were carried out using the statistical package for MATLAB® (R2014b, MathWorks, U.S.). Multi-way analysis of variance (ANOVA) was used for testing the effects of multiple factors on the plant variables. Therefore, the ECa data were grouped in 8 classes (Käthner and Zude-Sasse, 2015), while CWSI and WUEc were grouped according to the results of hotspot analysis.

Descriptive statistics of spatially resolved data was carried out using hotspot analysis according to Peeters et al. (2015), who used ArcGIS (ESRI, Redlands, CA, USA). In the present study, the algorithm was adapted for using the free spatial Matlab toolbox (Spatial Filtering, Max Planck Institute for Biochemistry, Germany). The method is based on the general (G) statistic for testing the effect of spatial autocorrelation (Getis and Ord, 1992) of the variables. Thereby a locally weighted mean around each observation is separately compared with the mean of the whole data (Anttila and Kairesalo, 2010). The outputs of the statistic are the z-score and the p-value, which indicate whether an observed pattern of clusters is statistically significant. Spatial clusters with statistically significant positive z-score are called hot spots, whereas the clusters with statistically significant negative z-score are called cold spots (Getis and Ord, 1992; Ferstl, 2007).

The ECa of soil at 25 cm depth indicated small-scale variability (Figure 1). The values of soil ECa reached a maximum of 24 mS m⁻¹ with a pattern of reduced values pointing to a sand lens visible in the center-eastern part of the experimental field (Figure 1) and neighboring area. Another sandy area was located in the south-west of the orchard. Values of soil ECa measured in 2013, increased compared to those obtained in 2012. This may be due to wetter soils, which was caused by the relatively high precipitation occurring in July and August 2013. This assumption is further supported by the close correlations found between the gravimetrical soil water content and ECa with R = 0.45 and R = 0.68 in 2012 and 2013, respectively (Table 1). In contrast, correlation coefficients of soil matric potential (pF) and soil ECa were only R = 0.15 and R = 0.44 in 2012 and 2013, respectively (Table 1). Repeated analyses showed similar pattern in different years with R = 0.88 considering 2011 and 2012 and R = 0.71 for years 2012 and 2013.

In 2012, during the acquisition of thermal images on 15th August from 13:40 until 16:19 (Figure 2), the mean global radiation was 641.1 W m⁻². August was the warmest month of the year with a mean maximum temperature of 25.6°C. The maximum air temperature on the day of measurement was 25.4°C. The diurnal increase of air temperature coincided with increasing VPD. The mean wind speed was 0.9 m s⁻¹. In 2013, mean global radiation of 306.8 W m⁻², maximum air temperature of 25.4°C, and wind speed of 1.2 m s⁻¹ were measured.

Compared to the free air temperature recorded by the automatic weather station, maximum temperature measured within the tree canopy occurred with a 3 h-delay. Maximum instantaneous water use efficiency was calculated just before noon and then declined during the rest of the day. In general, it ranged from 4.42 to 1.28 μMol CO₂ m⁻² s⁻¹/mMol H₂O m⁻² s⁻¹ (Figure 2). Inside the canopy, the VPD increased midday reaching a maximum in the afternoon at 17:00.

The instantaneous Ψ_leaf at midday varied between −0.40 and −2.14 MPa and Ψ_π between −1.93 and −2.56 MPa. At predawn, Ψ_leaf varied between −0.12 and −1.48 MPa and Ψ_π between −1.50 and −2.46 MPa.

Thermal images were acquired on partially cloudy days and wet and dry leaf-references were moved with the camera for each tree record within the orchard. With our camera set-up, l = 2.734 m and thus one pixel corresponded to 8.543 mm in width. This resolution was, thus, high enough to differentiate leaves, and to select the pixels that represent the wet and dry leaf-references (Table 2). In 2012, Vaseline® covered leaves were additionally used as dry leaf-references; however, the fingerprints of the application procedure remained visible on thermal images thus producing artifacts (data not shown).

The reference temperatures calculated with the analytical method (Ben-Gal et al., 2009) were always lower (Tw_ana 12–15 and Td_ana 17) compared to those measured on paper references or obtained from the histogram of images (Tw_histo 16–21°C and Td_histo 21–25°C). Furthermore, the correlations between leaf water potential (ψ_leaf) or osmotic potential (ψ_π) and the different crop water stress indexes were analyzed using data that were all obtained on the same day. Of all tested approaches, correlation coefficients for both ψ_leaf and ψ_π were highest for CWSI_JB, i.e., when the dry and wet temperatures were calculated analytically. In contrast, correlation between CWSI_J and ψ_leaf was low, showing enhanced variability caused by the appearance of clouds (Table 2). The use of air temperature plus 5° as Td and minimum temperature in the image as Tw for calculating CWSI_I (Irmak et al., 2000) resulted in a bias with overestimated values and also tremendously high variability due to clouds and, therefore, data were not used further. The CWSI_R ranged from 0.15 to 0.88, while the CWSI_J and CWSI_JB ranged from 0.03 to 0.78 and from 0.47 to 0.51, respectively. For CWSI_JB correlation with ψ_leaf was high, while the automated analysis of CWSI_R resulted in slightly reduced, but significant (p < 0.001) correlation coefficient of R = 0.52 (Table 2). However, the latter approach provided the advantage of feasible analysis of Td and Tw based on the individual images taken in the varying environment. Consequently, all further analyses were based on CWSI_R.

Hotspot analysis of ECa revealed one cold spot representing extreme low conductivity and 5 hot spots showing soil of high conductivity. Around the cold spot with critical z-value of < −1.65 (90% confidence level) a sand lens with an extension of ~20 × 25 m was found (Figure 1), while at the hot spots with critical value >1.65 (90% confidence level) water logging was observed after heavy rain fall indicating soil with lower particle size (Figure 3). The soil ECa was correlated with the number of leaves per tree. Consistently, spatial variability of canopy VPD within the orchard was found in the x-direction, which pointed to an influence of geographical position in the orchard (Rx = 0.31, Ry = 0.03, Rz = 0.20). This is the same direction as found for extreme values of soil ECa.

The CWSI_R ranged from 0.15 to 0.88. The hotspot analysis of CWSI_R revealed 5 cold spots occurring at z-values < −1.65 representing trees with no water shortage. The 3 hot spots appeared at critical value >1.65 referring to high CWSI_R. Here, the hot spots refer to unfavorable conditions with enhanced water deficit. The hot spots appeared on the east side of the orchard, within and adjacent to the position of the central sand lens (Figure 3). The cold spots were found in the western positions of the orchard.

The comparison of spots considering soil ECa and CWSI_R pointed to no correlation. Also, no correlation was found between canopy size dimension and CWSI_R considering the canopy length parallel to the row (R = 0.010), canopy width perpendicular to the row (R = 0.015), and volume calculated from length, width, and distance between first branch and last shoot (R = 0.001).

In 2012, the average leaf number per tree was 2,362. The SLA ranged from 32.00 to 59.76 cm² g⁻¹ and showed no correlation with soil ECa. The fruit size was correlated with soil ECa at R = 0.223 considering the hot and cold spots. However, other fruit quality variables did not correlate with soil properties.

No correlation between leaf water potential and fruit quality was found in the few trees measured. CWSI_R was correlated with SLA, but no significant difference was found for the number of leaves or fruit quality (Table 3). WUEc obviously depends primarily on the degree of crop load, because the water supply was kept uniform in the orchard. Mean WUEc was 2.362 g L⁻¹ in 2012 and 2.521 g L⁻¹ in 2013. In 2012, WUEc seemed only slightly, if at all, affected by soil ECa (R = 0.133), while in 2013, the correlation increased (R = 0.274).

The WUEc showed a correlation of R = −0.367, R = 0.183, and R = −0.270 with the fruit size, dry matter, and fruit flesh firmness, respectively, in 2012 (Table 3). Particularly, larger fruit size was correlated with low WUEc, and consequently with decreased crop load (Figure 4). Correlation between the above parameters seemed to be stronger in 2013. However, the reduce sample size in 2013 hampered the statistical comparison of the influence of slight drought stress on fruit quality in the different years.

WUEc showed no correlation with CWSI_R with R = 0.071 and R = 0.093 in 2012 and 2013, respectively. However, fruit quality was strongly affected considering the interaction of both variables (Table 4). Grouping according to WUEc and the instantaneous values of CWSI_R resulted in highly significant differences for fruit size and dry matter.

Shortly before harvest, the spatial patterns of soil ECa appeared closely related to soil water content with decreased ECa values at the positions of a sand lens found in the experimental orchard. This finding is consistent with earlier investigations carried out in areas with arid conditions (McCutcheon et al., 2006). The low correlation between soil matric potential and soil apparent electrical conductivity could be expected because previous chemical analyses of soils (Käthner and Zude-Sasse, 2015) at 10 spots of the same experimental site indicated only marginal <5% variations of phosphorus and potassium content, salinity, and pH. Increased, but still <10% variation was found for magnesium, calcium, sodium, and chloride contents. Nevertheless, the analyses of the variations of soil ECa during fruit development may provide data and information for the evaluation of spatial distribution pattern of water, which could potentially affect the quality of the mature fruit.

The Ψ_leaf measured predawn showed high variability and minimum value of −1.48 MPa indicating at least slight drought stress in some trees. Based on the measurements of weather and tree canopy microclimate, stable environmental conditions (Bellvert et al., 2014) during thermal imaging between 13:40 and 16:19 can be assumed for both years. Only the variation of radiation due to changing cloud cover could have slightly impaired thermal imaging due to the different dynamics of surface and air (ambient) temperatures (Agam et al., 2013). On the other hand, the analysis of the instantaneous WUEi, performed at the same time, revealed diurnal changes in a value range reported in other investigations on plum trees under similar conditions (Flores et al., 1985). Consequently, consistent sets of thermal readings may have been obtained on each measurement day.

The influence of clouds indeed appeared as a perturbing factor in the present study, especially when using dry and wet paper as references for obtaining Tw and Td. The use of air temperature plus 5 K for setting Td with low difference of Td and Tw resulted in high bias of CWSI_I. The analytical analysis of Td and Tw, as well as the automated approach resulted in significant correlation of CWSI and Ψ_leaf. Calculating Td and Tw analytically (Jones, 1999; Ben-Gal et al., 2009) had the disadvantage that weather data were needed. However, this method provided some insurance against artifacts. In the approach of intrinsic analysis of thermal images to calculate CWSI_R (Rud et al., 2015), references are directly obtained from the images, which is presumably the most feasible approach for an application of thermal imaging in a real world orchard avoiding the need for additional measurements. The approach appeared to be appropriate for the semi-humid summer rain region with cloudy conditions of the current study. The correlation coefficient of R = −0.52 considering Ψ_leaf and CWSI_R was at least encouraging to estimate the water stress of the plum trees.

Hotspot analysis (Getis and Ord, 1992) was applied to identify geographically located trees that differ from the mean. The spots found in the ECa data set point to significantly different clusters of trees appearing in the orchard. This small scale variability of soil ECa is typical for postglacial deposits which are common sources of soils in fruit production regions in temperate areas of Europe and Asia.

The CWSI_R varied between 0.15 and 0.88 presumably indicating a range of unstressed to stressed trees in the orchard. As for the ECa patterns, the appearance of significant clusters considering instantaneous CWSI_R points to a possible impact of tree water status on plant growth. However, we can certainly make no a-priori assumption on stable CWSI patterns, since crop load, stage of fruit development, and vegetative growth are all expected to influence water demand. This said, neither ECa nor crop load in the current study showed a correlation with CWSI_R.

Bellvert and co-authors identified an influence of the fruit development stage on the correlation coefficient of leaf water potential and CWSI in peach and nectarine (Bellvert et al., 2016) with increased correlation shortly before harvest, which is developmental stage 3. In olive fruits, less severe but equally directed correlation was found (Martin-Vertedor et al., 2011). In the current study, CWSI_R was similarly measured in stage 3 of plum fruit development corresponding to the second peak of fruit growth rate with high water demands.

In plum production, fruit size is of highest economic importance. In the present study, no effect of instantaneous tree water status as indicated by CWSI_R on fruit size was found. However, at high crop load, fruit size was reduced and water required to produce high quality (large enough) fruits may have been deficient. While the instantaneous canopy transpiration based CWSI_R alone did not indicate this level of potential water deficit, cumulative data of WUEc was correlated with fruit quality.

The reducing effect of crop load on Ψ_leaf or stem water potential has been pointed out previously (Naor et al., 2001; Marsal et al., 2010), particularly under very high crop load (Sadras and Trentacoste, 2011). An impact on the fruit size is consequent. WUEc, by definition, was dependent of crop load, since, as said, the water supply was uniform in the orchard. However, the variability of soil ECa might point to differences in effective water supply, which would be worthwhile to consider in future studies for calculating the effective WUEc.

Considering the spatial variability measured in the present study, the factor combination of the cumulative WUEc and instantaneous CWSI_R resulted in highly significant interaction with fruit quality. The effects of WUEc and CWSI outweighed the effect of soil ECa on the fruit quality. However, these findings certainly need additional experimentation and confirmation before development as a practical management tool.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.