Introduction

One of the most practiced irrigation strategies is to replenish the soil’s water reservoir based on evapotranspiration (ET) rates. The irrigation volume (IRV) is usually calculated from the estimations of the crop evapotranspiration (ETC) since the last time of irrigation; and so, water is supplied to maintain constant water availability for the plant (Allen et al. 1998). ETC is normally calculated from the measurement of lysimeter evapotranspiration (ETlys; López-Urrea et al. 2012). However, since field-scale lysimeters are expensive and therefore scarce, they are normally used to develop empirical models that link the reference evapotranspiration (ETO) with ETC (Evett et al. 2012). The ratio between ETC and ETO, normally termed as the crop coefficient (KC), provides irrigators with a simple method to transform ETO—available through weather stations data—into ETC. From a physiological perspective, KC is a function of the crop’s evaporative surface (i.e., leaf area, LA) and its hydraulic resistance. Therefore, the calculation of ETC for a specific field is valid only under the assumption that the crop hydraulic resistances in that field and in the lysimeter are equal (Annandale and Stockle 1994). Such an assumption is reasonable when both field and lysimeter plants are grown under near optimal conditions, but when field plants are grown under water deficit, the crop resistance is likely to increase and reduce the actual ET.

As many tree crops (e.g., almonds, pistachio, citrus, and others) could benefit from deficit irrigation (Fereres and Soriano 2007), it is important to develop methods for deficit irrigation scheduling. Many past studies based the deficit irrigation treatments on multiplying the ETC of well-watered plants by an irrigation factor (IF), normally ranging between 0.2 and 1:

$${\text{IR}}_{\text{V}} = {\text{ET}}_{\text{C}} \times {\text{IF}}$$
(1)

For example, using an IF of 0.35, Shellie (2006) downregulated the irrigation amounts by 65 % to maintain grapevines (Vitis vinifera L.) at a leaf water potential of 0.3–0.5 MPa lower than the well-watered control treatment (100 % ETC). Specifically for grapevines, numerous studies successfully utilized the IF to calculate the IRV and its influence on berry characteristics (e.g., Intrigliolo and Castel 2010; Romero et al. 2010; Munitz et al. 2016).

However, there are still several fundamental gaps in our understanding of deficit irrigation scheduling according to the IF, especially when it comes to the ET of deficit-irrigated treatments. We know that the IF leads to a constant irrigation ratio between the deficit-irrigated and the well-watered plants. At the same time, it is not clear whether a similar ratio also exists between the ET of these treatments. As a result, it is not clear whether deficit irrigation according to the IF replenishes the soil water reservoir or leads to its gradual depletion. Moreover, the stability of KC under a deficit irrigation regime, with respect to the environment and plant physiology, requires an in-depth investigation.

While data on the whole plant transpiration are lacking, an examination of water relation studies and models suggests that the ratio of a single leaf ET between deficit-irrigated and well-watered plants depends on the atmospheric demand (Oren et al. 1999; Pou et al. 2008). From a hydraulic perspective, the stomata regulate the transpiration to avoid a critical water potential that would lead to xylem embolism or turgor loss (Tyree and Sperry 1989). Assuming that the soil water potential is high enough, maintaining high stomatal conductance (g s) would improve the carbon flow into the leaf without reaching this critical water potential. Accordingly, the response function of g s to stem water potential (Ψ s) is commonly described as sigmoidal (e.g., Tombesi et al. 2014). Therefore, any reduction of Ψ s due to high ETO should lead to a different downregulation of g s, depending on the soil water potential. This notion is reflected in the recent transpiration model by Sperry and Love (2015) that predicts very similar transpiration rates for both well-watered and water deficit plants under low ETO, but much lower transpiration rates for water deficit plants under high ETO.

In the present experiment we used lysimeters to compare the ETlys of well-watered (WW) grapevines with that of water deficit (WD) grapevines that were irrigated at 35 % of ETlys WW (i.e., IF = 35 %). We hypothesized that the ratio between the two treatments depends on ETO. The two treatments were compared one to another and also to midday physiological parameters in order to test the feasibility of utilizing these parameters to predict ET.

Materials and methods

Plant material and site description

The study was conducted during the two consecutive seasons of 2014 and 2015, at the University of Udine experimental station “A. Servadei” (46°02′N, 13°13′E; 88 m a.s.l.). The plant material consisted of four-year old Vitis vinifera cv. Merlot (clone R3), grafted to SO4 rootstock and potted (40 L) in a mixture of soil (49.0 % sand, 31.5 % silt, and 19.5 % clay) that was supplemented with 20 % perlite. The soil water holding capacity for each pot was 11.6 L. At harvest the vines had on average 1.8 kg of grapes/plant. To prevent natural precipitation, the trial was conducted under a transparent plastic (ethylene-vinyl-acetate) roof (4.5 m in height) that covered the entire experimental plot while the sides of the framing structure remained open (Herrera et al. 2015). The trial consisted of two rows of 16 plants each, spaced 2.5 m apart. Each row was divided into four blocks, for a total of eight plots. The in-row spacing between each vine within the block was 1 m.

The vines were cane pruned to a single Guyot (0.8 m high), with 8–10 nodes per cane, and trained to vertical shoot positioning. Hedging was not performed over the course of the experiment. The plants were provided with mineral nutrition (N-P-K Nitrophoska®), and fungicide was applied as commonly practiced in the area.

Lysimeter setup

Each four-vine plot consisted of three pots buried in the ground and a fourth (served as a lysimeter) that was placed in a hole (50 × 50 × 50 cm), to level with the other pots. The lysimeter was always in a central position within each plot. The outside surfaces of the lysimeter pots were thermally protected through shading, insulating foam, and aluminum foil in an attempt to minimize the effect of soil temperature fluctuation.

A self-constructed balance, based on two-beam single point load cells (©Tedea Huntleigh, Model 1042; maximum mass: 75 kg; accuracy: 0.02 % of rated output), was placed at the bottom of each hole, beneath the lysimeter pot. In 2014, the system was operated manually, starting on July 22. Lysimeters-grown vines were weighed at night, five times per week, both before and after irrigation was applied and drainage had ceased. In this way, it was possible to compute four evapotranspiration (ETlys) values per week for each lysimeter. In 2015, the scales operated automatically and pots’ masses were recorded every minute using a CR1000 data logger (Campbell Scientific, Inc., USA). ETlys was calculated as the difference between the average mass between 04:00 and 05:00 (after irrigation) and the average mass between 22:00 and 23:00 (before irrigation). In 2015, the system was activated in May, resulting in daily ETlys values for the whole season.

Irrigation treatments

Water was supplied by a drip irrigation system with a set of two emitters per pot (PCJ 2 L h−1, Netafim, Israel). Treatments were randomly assigned to each plot and were established at 50 % veraison, on July 24, 2014 and on July 31, 2015, and comprised two irrigation levels: well-watered (120 % of ETlys, WW) and water deficit (IF = 35 % of the WW treatment’s ETlys, WD). Irrigation treatments for each week were generally calculated from the average ETC of the previous week, but were modified in the case of an extraordinary daily ETO value.

Irrigation volumes were calculated through the linear regression (R 2 = 0.995) between irrigation duration and the volume dispensed by the drippers (data not shown). At the beginning of each season, the volume emitted by all drippers was compared to ensure that all drippers were within a 10 % interval. Abnormal drippers were replaced. Irrigation was applied daily and during hours of minimal transpiration (at 21:00 in 2014 and at 00:00 in 2015) to minimize significant effects (ET during the irrigation time was not considered) on the daily ETlys calculation

Following harvest (54 and 60 days after veraison in 2014 and 2015, respectively), the irrigation treatments remained in effect for another 10 days. After this interval, the WD treatment was abandoned, and all vines were irrigated to field capacity.

Weather data

Weather data were collected within the experimental plots to account for the possible impact of the plastic roof. In order to calculate the under-roof potential evapotranspiration according to the Penman–Monteith model (FAO56; ETO), measurements of temperature and relative humidity (HMP60, Vaisala, Finland), wind speed (Gill Windsonic, Gill Instruments, UK), and global radiation (LI-200SL, LiCor, Inc., USA) were recorded at a frequency of 10 s using a CR1000 data logger (Campbell Scientific, Inc., USA) and then averaged at 30 min.

Soil sensors

Soil volumetric water content (v/v; θ) was monitored in each lysimeter pot using EC-5 sensors (Decagon Devices, USA) installed at 10–15 cm deep with the waveguides in vertical position. Measurements were recorded every 30 min using the EM50G data logger (Decagon Devices, USA).

Leaf area

Leaf area (LA) for the eight lysimeter-grown vines was calculated six and nine times over the course of the seasons in 2014 and 2015, respectively, based on the number of leaves and their average area (Montero et al. 2000). The latter was computed using a linear model based on the average length of all the leaves in one representative shoot per vine (Rapaport et al. 2014). The model was developed by collecting 60 leaves and measuring their area using a Li-3100 (Li-Cor, Inc., USA; R 2 = 0.89).

Evapotranspiration (ET) parameters

Measurements of evapotranspiration from lysimeters (ETlys) are usually normalized to the soil area (Allen et al. 1998), facilitating their implementation for agricultural use. However, the adopted experimental setup (40-L lysimeters not in a field) questions the validity of such normalization. Instead, ETlys values were directly normalized to ETO (as in Ben-Gal et al. 2010), resulting in KC (m2 plant−1):

$${\text{K}}_{\text{C}} = \frac{{{\text{ET}}_{\text{lys}} }}{{{\text{ET}}_{0} }}$$
(2)

Physiological parameters

Stem water potential (Ψ s) and stomatal conductance (g s) were periodically measured (8–9 times in each year) during the experiment and compared with ET. Both measurements were conducted on all the lysimeter vines (four repetitions per treatment) at midday, during sunny days, on a young fully expanded leaf.

To determine the stem water potential (Ψ s) the leaves were bagged and covered with aluminum foil for 1 h before they were excised from the shoot using a sharp blade. While still bagged, the leaves were placed into the pressure chamber (Soil Moisture Co., Santa Barbara, CA, USA) with the petiole protruding from the chamber lid. The chamber was pressurized using a nitrogen tank, and Ψ s was recorded when the initial xylem sap was observed emerging from the cut end of the petiole.

Stomatal conductance (g s) was measured using an infrared gas analyzer (LI-6400, LiCor, Inc., USA). The measurements were taken at a constant light intensity (1000 μmol m−2 s−1) and CO2 concentration (400 μmol mol−1).

Results

The lysimeter measurements illustrated the large variability in grapevine evapotranspiration (ETlys) throughout the growing season. With respect to the evaporative demand and vine physiology, the transpiration ranged between 0 and 6 L plant−1 day−1 in 2015 (Fig. 1a). The 2014 data collection only began at veraison, when the irrigation treatment started, but the 2015 data shows that the ETlys was minimal shortly after bud break and quickly increased as soon as the canopy developed (Fig. 1), reaching the maximal ETlys on August 13. The application of the WD treatment at veraison resulted in a quick reduction of ETlys to a daily consumption of 0.65–1.3 L.

Fig. 1
figure 1

a Seasonal pattern of the lysimeter evapotranspiration (ETlys) and b leaf area (LA) of the well-watered (WW) and water deficit (WD) treatments in the 2015 season. Data are the averages of four plants ±SE

Full size image

The irrigation treatments led to significant differences of Ψ s and g s (Table 1). The physiological differences were largest between 10 and 20 days after veraison with a difference of ~0.8 MPa in Ψ s and of ~150 mmol m−2 s−1 in g s between the treatments. Surprisingly, in 2014 there was a substantial reduction in g s of the WW treatment toward harvest to values of 70 mmol m−2 s−1 (Table 1). The WD treatment led to a fast decline in θ for the first 6–8 days after the initiation of the treatment in both seasons (Fig. 2a, b). Following this short period, there was no further decline of θ, which ranged from 0.05 to 0.2 (v/v), significantly lower than the θ of the WW treatment (0.3–0.5 v/v). While the soil conditions were fairly similar between the two seasons, the evaporative demand (ETO) was lower in 2014 than in 2015, as it was exceptionally cool and rainy. During the irrigation period, in 2014, the maximal daily ETO was 3.7 mm, while in 2015, it was 5 mm (Fig. 2a, b). Additionally, a difference between the KC of the WW treatment was observed between the two seasons. In 2014, an unexpected reduction in KC during mid-August led to low KC values (~0.6 m2 plant−1) in the WW vines until harvest. Conversely, and as expected, in early August of 2014 and throughout the duration of the WD treatment in 2015, KC was significantly larger in the WW treatment (1 m2 plant−1) than in the WD treatment (0.4 m2 plant−1; Fig. 2c, d).

Table 1 Midday stem water potential (Ψ s, MPa) and stomatal conductance (g s, mmol m−2 s−1) in well-watered (WW) and water deficit (WD) grapevines during the experimental seasons (from veraison to harvest) in 2014 and 2015
Full size table
Fig. 2
figure 2

Soil water content (θ; a, b), reference evapotranspiration (ETO; a, b), leaf area (LA; c, d) and the crop coefficient (KC; c, d) of the well-watered (WW) and water deficit (WD) treatments during the application of the water treatment (starting at veraison, dashed line) in 2014 (a, c) and 2015 (b, d). Data are the averages of four plants ±SE

Full size image

During canopy development, there was a strong correlation between LA and KC (R 2 = 0.89; Fig. 3a). However, soon after the application of the WD treatment, since KC values were significantly reduced (Fig. 2c, d), no correlation between LA and KC was found (Fig. 3b). The WD treatment did result in a decrease in LA from 2.3 to 1.51 m2 plant−1 in 2014 and from 2.83 to 1.78 m2 plant−1 in 2015 (Fig. 2c, d). However, apparently this decline had a negligible effect on KC compared with stomatal regulation, leading to the low correlation between KC and LA (R 2 = 0.02).

Fig. 3
figure 3

Crop coefficient (KC) as a function of leaf area (LA) during canopy development (before treatments were imposed; a) and leaf shedding (following the water deficit treatment; b). Data are the averages of four plants

Full size image

To assess whether the irrigation scheduling based on the irrigation factor (IF = 0.35) supplied the actual ET of the WD vines, the ETlys of the two treatments were regressed (Fig. 4). As expected, assuming the regression is linear, the slope (S) was 0.347, a value very close to the scheduled IF (0.35), but with a poor correlation coefficient (R 2 = 0.343, P < 0.001; data not shown). Instead, the better-fitting logarithmic relationship between the ETlys of the two treatments (R 2 = 0.69, P < 0.001; Fig. 4) emphasizes that the relation between the treatments’ ET is dynamic, possibly depending on environmental conditions and on the degree of water stress developed by the plant during the imposed deficit irrigation conditions.

Fig. 4
figure 4

Lysimeter evapotranspiration (ETlys) of the water deficit treatment (WD) as a function of the ETlys of the well-watered treatment (WW). Data were taken only from post-veraison, during the application of the irrigation treatments. Data are the averages of four plants

Full size image

As hypothesized, ETlys was highly correlated to ETO, with a very different regression for each of the treatments (Fig. 5a). Under low ETO (<2 mm), both treatments exhibited similar ETlys, meaning that water limitations had little effect on transpiration in the WD treatment. Conversely, under an ETO of 3 mm or above, the ETlys of the WD treatment was constant at around 1.2 L plant−1 day−1, meaning that under such an ETO, plant resistance was modified to limit transpiration. At the same time, the response curve of the WW treatment maintained a linear increase of ETlys as a function of ETO (Fig. 5a). Since both treatments had similar ETlys values under low ETO, but very different ETlys under high ETO, the evapotranspiration ratio between the treatments, ranged from 1 (i.e., similar water consumption) under low ETO down to 0.25 under the high ETO of 5 mm (Fig. 5b).

Fig. 5
figure 5

Lysimeter evapotranspiration (ETlys) as a function of the reference evapotranspiration (ETO) in the well-watered (WW) and water deficit (WD) treatments (a), and the ratio between the two treatments (b). Data are the averages of four plants

Full size image

In light of the great effect the treatment imposed on KC, different parameters (θ, Ψ S and g s) that could represent the plant limitation to transpiration were compared with KC. These parameters were correlated with KC, each with a different response function (Fig. 6). The logarithmic response of KC to θ flattens at the θ of 0.3, probably since higher θ values do not consist in any water limitation for the plant (Fig. 6a). In addition, the response curve of the Ψ s flattens at values higher than −0.7 MPa, probably for the same reason (Fig. 6b). Both θ and Ψ s were not correlated with the unexplained decline of KC in the WW treatment during 2014 (in red; Fig. 6). On the other hand, the entire data set of g s (including WW 2014) was well correlated with KC, resulting in the highest correlation coefficient (R 2 = 0.79). The intercept of the g s linear regression suggested that even when the transpiration component is 0, soil evaporation will result in a KC of 0.3.

Fig. 6
figure 6

Crop coefficient (KC) as a function of the daily average soil water content (θ; a), midday stem water potential (Ψ S; b), and stomatal conductance (g s; c) of the well-watered (WW) and water deficit (WD) treatments. Values of unexpected decreases in KC during the late summer of the WW 2014 treatment are highlighted in red. Data are the averages of four plants ±SE

Full size image

Discussion

The high correlation between the seasonal ET and the LA, as the one calculated for the measurements during canopy development (Fig. 3a), is supported by previous studies (Williams et al. 2003; Netzer et al. 2009; Picón-Toro et al. 2012). These authors found that under WW conditions, the LA is a good predictor of KC and suggested practical ideas for its incorporation into irrigation models (Williams and Ayars 2005; Netzer et al. 2009; López-Urrea et al. 2012). It is important to mention that application of deficit irrigation early in the season could lead to a substantial effect on the plant growth (Matthews et al. 1987). Conversely, our results show that under post-veraison WD, restrictions derived from the plant hydraulic resistance were substantially more prominent than LA modification, and led to a significantly lower KC (Fig. 2). The dominancy of the plant hydraulic resistance over LA was expected, since the downregulation of gs under WD, up to 40-fold (Fig. 6c), was considerably larger than the modifications of LA. As a consequence, KC should be modeled with respect to the modifications in plant resistance that the WD imposes. Previous studies have incorporated this “stress factor” (i.e., a fraction that is multiplied with the calculated ET) into their models (Raes et al. 2009; Steduto et al. 2009). Nonetheless, the usage of a constant fraction is probably not accurate, since the stress factor depends on the environmental conditions (Fig. 5).

If deficit irrigation applied through IF provides the plant with its ET (IRV = ETlys), we can expect an apparent steady θ; and indeed, following the initial reduction in θ over the first 6–8 days, the stability of θ suggests that the ETlys of the WD treatment reached an apparent equilibrium with the IRV. Under field conditions, in which the water holding capacity is much larger than the 40-L lysimeters, the reduction in θ could take a much longer period than 6–8 days (Green et al. 2008). Nonetheless, since the plant hydraulic resistance is expected to grow with the decline in θ, an apparent steady state should be reached once the plant resistance limits the ET to be equal to the IRV. Accordingly, in the current experiment, after 6–8 days, an apparent steady state was reached, maintaining a relatively constant θ and KC. The term “apparent” is used to describe the steady state condition, because the change in environmental conditions resulted in a subsequent modification of KC and θ.

Our results show that the relation between the ETlys of the WW and WD treatments is not linear (Fig. 4). Therefore, using a constant IF for deficit irrigation scheduling will result in an IRV that could be lower or higher than the actual ETlys, leading to continuous small fluctuations of θ. The ratio between the ET values in WW and WD conditions depends on ETO (Fig. 5b), meaning that the evaporative demand determines the relative transpirational sensitivity of deficit-irrigated grapevines. A similar phenomenon was measured by Collins et al. (2010) when comparing the ET of WW and WD plants under different ETO using sap flow meters. High ETO will likely drive WD plants to a higher stress degree than WW and as a consequence WD plants will further reduce their transpiration. Therefore, deficit irrigation scheduling according to the IF under high ETO will probably result in excessive water loads, while under low ETO, it is probably deficient (compared with the actual ET). The overall ETlys (during long periods) should be equal to the IRV, but short periods of unusual ETO might lead to inaccurate irrigation, leaving a margin for optimizing the irrigation volumes. Plants with a large root volume, such as under most field conditions, could probably buffer the deficit or excess irrigation; however, crops that are drip irrigated with high frequency or grown in sandy or shallow soils could be sensitive to several days with extreme ETO.

The logarithmic behavior of ETlys compared with ETO in the WD treatment was predicted by Feddes et al. (2001) and is probably the result of stomatal closure under a high vapor pressure deficit (VPD; Pou et al. 2008). High transpiration rates lead to lower Ψ s and the downregulation of g s, resulting in a saturation of ET values despite the increased VPD. The differential response of ETlys compared with ETO in the two treatments is well explained by the model of Sperry and Love (2015). This model assumes that gs regulation is a function of the “soil-canopy conductance vulnerability curve,” which should be similar in both treatments. Since this function is sigmoidal in nature, increased VPD under WW conditions will maintain Ψ s close to the upper asymptote, leading to a similar g s and a linearly growing ETlys. Conversely, increased VPD under WD conditions will drive Ψ s into the steep slope of the sigmoid, leading to a large reduction in g s and the saturation of ETlys (Sperry and Love 2015). While the model (Sperry and Love 2015) explains well the qualitative differences between treatments and environments, generalizing it to predict ETlys might prove to be very difficult as g s regulation largely depends on the soil and genotype (Soar et al. 2006; Tramontini et al. 2014).

This difficulty in calculating reliable estimates of ET based on environmental models highlights the advantage of incorporating direct plant or soil measurements into the models. All of the three measured parameters (θ, Ψ s, g s) showed a promising potential for predicting KC in this study and others (Rana et al. 2004; Williams et al. 2012). Since all three are good indicators for the plant water status (Cifre et al. 2005) and all correspond to one another (Hochberg et al. 2013), the correlation with KC was expected. However, the nature of the coordination between θ, Ψ s and g s highlights the disadvantages of using θ or Ψ s.

The main control over transpiration—and accordingly over KC under WD—is the regulation of g s (van den Honert 1948). Following a water limitation and a reduction of θ and Ψ s, a combination of hydraulic and chemical signals will downregulate g s and KC (Franks 2013). Therefore, we expect KC to be correlated with θ and Ψ s as long as water limitation is the key factor in g s regulation. Previous experiments showed these high correlations in grapevine (Rana et al. 2004; Williams et al. 2012), olives (Ben-Gal et al. 2010) and peach (Mata et al. 1999). However, g s is not only regulated in response to water limitation, but also by plant development, light, temperature, mesophyll CO2 concentration, and several other known and possibly unknown factors (Schroeder et al. 2001); accordingly, KC could be modified without a correspondent change in θ or Ψ s. Even more so, a reduction of KC due to factors other than water (for instance, low light) would lead to an increase in both θ and Ψ s. A good example is the unusual reduction of KC in the WW treatment during August of 2014 (Fig. 2) that highlighted the potential problem of using θ or Ψ s to predict KC (Fig. 6).

On the other hand, the strong linear regression between g s and KC (Fig. 6c), even in the 2014 WW treatment, should not come as a surprise. Both parameters are similarly calculated (from ET or transpiration normalized to ETO or VPD for KC and g s, respectively) and physiologically resemble each other. Their good correlation suggests that the instantaneous response of a single leaf at midday provides a good indication of the entire vine daily water use. It is important to mention that in denser canopies with a larger number of shaded leaves, the relationship between g s and KC might be different. At the same time, even under denser canopies a strong KC ~ g s correlation is expected (Williams et al. 2012). Assuming a similar canopy density, the relation should be similar across genotypes and environments. Conversely, the regression between θ or Ψ s and transpiration largely varies with soil type, genotype and atmosphere (Rogiers et al. 2009; Tramontini et al. 2014), complicating the ability to generalize transpiration models to different vineyards.

Conclusion

The results provide an insight into the ET of deficit-irrigated grapevines and show their increased relative transpirational sensitivity under high ETO. Calculating the ET of deficit-irrigated vines as a fraction of the ET of well-watered vines might lead to either over or underestimation of irrigation volumes, and thus, deficit irrigation scheduling should consider its differential response to ETO. Furthermore, the high correlation between g s and KC, and the difficulty in predicting both suggest that the measurement of g s (both direct and indirect) could provide a reliable estimation for the ET of crops under deficit irrigation regime.