Abstract
Vine water status is a major parameter for vine management because it affects both wine quality and yield. In order to optimize vineyard management and harvesting practices, it is necessary to characterize vineyard water status spatial variability. This work aims at establishing an empirical spatial model for stem water potential (ΨStem) with ancillary data based on vine water status. Carbon isotope ratio (δ13C) measured at harvest was selected as ancillary data because it reflects only the effect of vine water status variations integrated over the season and is not impacted by other factors such as vine nitrogen status. The proposed model was applied at the intrablock level. It is based on the spatial extrapolation of a ΨStem value measured at a reference site using δ13C values collected over the block. Measurements of ΨStem and δ13C were carried out over three consecutive years on 96 locations within the block. ΨStem values obtained with a spatial model were more accurate than ΨStem values obtained with a nonspatial model, indicating the relevancy of δ13C values to account for spatial variability of vine water status. Results show that operational maps of vine water status can be obtained by means of a spatial model, in which δ13C values from a previous season are used as ancillary data. Maps can be updated at any given time during the season by carrying out a limited number of ΨStem measurements in selected locations. This model offers a tool to monitor vine water status and to implement management practices while considering vine water status intrablock variability.
Vine water status is related to soil factors (water-holding capacity), climate factors (evaporative demand, rainfall), vine architecture, and management practices (irrigation, implementation of cover crop). It impacts shoot growth, vigor, yield, and grape ripening (Lebon et al. 2003). When vines are grown for wine production, a balance between vegetative and reproductive development is important to achieve high-quality wines (Ribéreau-Gayon et al. 1998). This balance is generally reached when a limiting factor contains vegetative development. Water deficit is a major factor potentially limiting vegetative development (Pellegrino et al. 2005). However, it adversely impacts grape yield and can increase or decrease grape quality potential for winemaking, depending on the severity of water deficit stress (Matthews and Anderson 1989, van Leeuwen et al. 2009). Moreover, the impact of water deficit on grape quality potential varies according to the timing of water deficit (Ojeda et al. 2002). Hence, assessment of vine water status is important for viticulture management.
Many methods have been developed over the past decades to assess vine water status and can be soil based (Seguin 1986, Koundouras et al. 1999), plant based (Cifre et al. 2005), or model based (Lebon et al. 2003). Among plant-based methods, water potentials, measured with a pressure chamber, are accurate indicators of water status because they are well correlated to physiological responses of plants to drought, such as stomatal conductance reduction (McCutchan and Shackel 1992). The pressure chamber technique can be used to measure (1) leaf water potential, (2) predawn leaf water potential, and (3) stem water potential. Among these three applications of the pressure chamber technique, the latter two provide the most accurate and robust estimations of vine water status. Predawn leaf water potential (ΨPD) provides a good estimate of vine water status when water deficits are severe, while stem water potential (ΨStem) is the most discriminating indicator for both moderate and severe water deficits (Choné et al. 2001a). The measurements of ΨPD and ΨStem have become the reference techniques for estimating vine water status (Santesteban et al. 2011). Nevertheless, two main drawbacks make them impractical for assessing intrablock variability of vine water status: they are time consuming and can only be performed at certain times. This time is limited to two hours at the end of the night for ΨPD and three hours in the early afternoon for ΨStem, resulting in a limited number of measurements per day. If vine water status is to be represented spatially, within a vineyard block or within an estate, many measurements have to be conducted, which can be difficult, if not impossible, to implement (van Leeuwen et al. 2006). Moreover, vine water status continually evolves over the season, depending on climatic conditions (evaporative demand and rainfall), phenological stages (Olivo et al. 2009, Bowen et al. 2011), or crop load (Bota et al. 2004). If vine water status is to be used for practical vineyard management (such as irrigation), water potentials measurements have to be made throughout the season.
Vine water status can be highly variable within a vineyard block (van Leeuwen et al. 2006, Taylor et al. 2010), inducing a wide variability of vine response at the intrablock scale in terms of vigor, yield, precocity, and grape composition (Tisseyre et al. 2008). Precision viticulture aims to modulate management practices at the intrablock scale in order to homogenize vine expression (Tisseyre et al. 2007, Arnó and Rosell, 2009). Hence, assessing spatial variability of vine water status is an important issue for fine-tuning vineyard management.
In a review paper, the authors discussed the importance of developing a method for the spatial monitoring of plant water status, assessed by ΨPD measurements, and proposed a conceptual spatial model to predict ΨPD across a given domain (vineyard block, vineyard, region) (Acevedo-Opazo et al. 2008). The proposed model is an empirical approach which predicts ΨPD by combining local reference measurements (to take into account temporal variability) and a spatial model determined from ancillary data that are easily obtained with a high spatial resolution. By coupling high-quality, high-cost punctual vine measurements with low-cost medium-high density ancillary data, the model provides quick estimations of ΨPD at spatial resolutions that would be cost prohibitive and cumbersome to generate with classical punctual measurements. A subsequent paper demonstrated the possibility of providing estimations of ΨPD spatial variability using only two types of ancillary data: trunk circumference and exposed leaf area (Acevedo-Opazo et al. 2010). The study was conducted under severe water restriction corresponding to nonirrigated Mediterranean vineyards. It is likely that under those conditions, trunk circumference and leaf area are mainly influenced by vine water status, making these ancillary data relevant to establish the spatial model. However, in situations where water deficit is less severe, other factors, and particularly vine nitrogen status, are likely to influence vine vigor as well (Choné et al. 2001b). In such conditions, a spatial model based on vine vigor related ancillary data might not be appropriate and data directly related to vine water status should be preferred.
Carbon isotope discrimination measured on grape sugar at ripeness (called hereafter δ13C) is a potential candidate to provide water status-related ancillary data for establishing a spatial model to predict ΨPD. δ13C is based on the principle that ambient atmospheric CO2 contains 98.9% 12C isotope and 1.1% 13C isotope. 12C is more easily incorporated in hexoses during photosynthesis. Therefore, the sugar produced by photosynthesis contains a higher proportion of the 12C isotope than ambient CO2. This process is called “isotope discrimination.” When plants undergo water deficit conditions, isotope discrimination is reduced because of stomatal closure (Farquhar et al. 1989). Therefore, the 12C/13C ratio in photo assimilates provides a signature of plant water status over the period in which they were synthesized. When measured on grape sugar at ripeness, the 12C/13C ratio (so-called δ13C) provides an integrative measure of vine water status during grape ripening (Gaudillère et al. 2002), also related to meteorological variables (mean temperature and total rainfall) during this period (Aghemo et al. 2011). The greatest sensitivity of δ13C response to water deficit occurs during two critical periods of berry ripening: around veraison and at 4 to 6 weeks before harvest (Santesteban et al. 2012). Vine water status assessment with δ13C is conducted on grape juice sampled just before ripeness. Measurements can be delayed and samples can be frozen before analyses. Unlike assessment of ΨPD or ΨStem, the δ13C measurement is less subject to time constraints. The only factors limiting the number of measurements is the time required to sample and process the grapes and the cost of the analysis.
This study examined whether a previously proposed model (Acevedo et al. 2010) could be extended to a larger set of environmental conditions (i.e., a context of moderate water restriction) by using water status-related ancillary data. Thus, δ13C was examined as relevant ancillary data to establish a spatial model to predict variability of vine water status.
Materials and Methods
Site description.
The study was conducted in a 0.23 ha vineyard block located in Villenave d’Ornon (Pessac-Léognan appellation, Bordeaux region, France; lat. 44°45’, long. 0°33’) during the 2003, 2004, and 2005 seasons. 2003 was a very warm and rather dry season, 2004 was rather dry with temperatures close to average, and 2005 was a warm and very dry season (Table 1).
Climatic conditions of the 2003, 2004, and 2005 seasons in Villenave d’Ornon, France.
The soil varied in texture, ranging from gravel to heavy clay. Vitis vinifera L. Merlot grapes, clone 181, grafted on Fercal rootstock, were planted in 1991. Vine spacing density was 6,250 vines/ha (1.6 m interrow spacing; 1 m between vines in the row). Row orientation was north–south. Trunk height was 0.5 m. Vines were double-Guyot pruned, vertical shoot-positioned, and hedged at 1.5 m. The block was dry farmed.
The block was comprised of 18 rows of 80 vines each. Measurements were carried out on rows 2, 5, 8, 11, 14, and 17 and separated by 4.8 m in an east–west direction. Measurements were carried out on 96 sites of three vines each (16 in each sampling row, separated by 5 m in a north–south direction), regularly distributed in the block (Figure 1).
Location of the within-block sampling sites; reference site is showed within a frame.
Intrablock variability.
Measurements were taken to assess intrablock variability of variables characterizing vegetative and reproductive development. Measurements were implemented on the 96 sites in 2003, 2004, and 2005, except for leaf area index (48 sites, in 2003 and 2004 only). Leaf area index (LAI, m2/m2) was measured with a LAI meter (Li-COR, Lincoln, NE; Ollat et al. 1998). Pruning weight was recorded immediately after winter pruning. Yield and berry weight were recorded at harvest. Grape sugar concentration at harvest (g/L) was determined with a TRAACS-800 analyzer (Technicon, Ontario, Canada).
Vine water status.
Stem water potential was measured with a pressure chamber on leaves that were covered with an opaque plastic bag one hour prior to measurement (Choné et al. 2001a). For each of the 96 sites, measurements were conducted on three vines (one bagged leaf per vine) on 31 Jul 2003, 7 Sep 2004, and 7 Sep 2005, during the driest period of each season. In 2004, a second measurement was carried out on 8 Sep on every second site (48 sites). To implement the measurements in the required time frame (12:30 to 3:30 pm solar time), six previously calibrated pressure chambers were used simultaneously.
Grape samples were taken each year prior to harvest. For each site, 66 berries were sampled on three consecutive vines (198 berries per sampling), taking berries from various parts of the bunch. Samples were transported in plastic bags to the laboratory and juice was extracted by crunching the berries inside the bag. Two mL of juice was introduced into an Eppendorf tube and centrifuged at 10,000 rpm. Tin capsules (TIN 6 × 4 mm) were introduced into a 96-well (8 mm) microplate (model 83.1835; Sarstedt, Nümbrecht, Germany). Five μL of grape juice was introduced in each tin capsule by micropipette. The microplate was placed in a nonventilated stove at 60°C for 24 hr. The microplate was then closed and sent to a laboratory specialized in stable isotope analysis to measure δ13C by isotope-ratio mass spectrometry (IR-MS).
Analysis of temporal stability of δ13C spatial patterns.
Kendall’s coefficient of concordance (W), proposed by Tisseyre et al. (2008) to test the temporal stability of spatial patterns, was used to assess the stability of δ13C along the 2003, 2004, and 2005 seasons. The analysis was conducted on a matrix, X, where the lines referred to the sites of measurement and the columns to the year. W varies from 0 in the case of total disagreement (i.e., no temporal stability) to 1 in the case of total agreement. The equation to compute W is given by Equation 1 (Kendall and Smith 1939).
(Eq. 1)
with
and where n is the number of sites of measurement, k is the number of year,
is the average rank of the measurement site over all the considered years, Xi,t is the δ13C value of site i of the year t, and R(Xi,t) is the rank of Xi,t among all the values of the year t.
Computation of the spatial model.
Following the previous empirical approach (Acevedo-Opazo et al. 2008), the spatial model tested here is
(Eq. 2)
where s corresponds to any of the 96 sites where a δ13C value was measured. The model requires the measurement of a ΨStem value (Ψ(sre,t)) at the reference site (sre) and time t. Its principle is to extrapolate the reference value Ψ(sre,t) using a function that relates Ψ(sre,t) to the δ13C values (δ13C(s)), over each site s of the block at the same time (t). Therefore, the model provides an estimation Ψ̂(s,t) of ΨStem value at any site s of the block where a δ13C is available. In this study, a linear relation defined by the coefficients b0 and b1 was considered. Equation 2 presents an empirical model that is intended to provide the best estimation as possible based on measured data. The coefficients b0 and b1 are not time dependant or site dependant. Therefore, the approach assumes that (1) the spatial model may be calibrated for the whole field, (2) once the coefficients are determined, they are stable over the time, and (3) climate (precipitation or irrigation and vapor pressure deficit) is homogeneous over the field.
Choice of a site of reference (sre).
The choice of the site of reference (Equation 2) could impact the accuracy of the prediction from the proposed model, particularly in high water restriction conditions (nonirrigated) when there is a wide range of variation in ΨStem over the block. Acevedo-Opazo et al. (2010) showed that, at the within-block level, the random selection of site of references did not affect the model output adversely. In a first approach, the site of reference was chosen randomly. The incidence of this choice was tested through a robustness analysis.
Model calibration.
ΨStem measurements taken on 31 Jul 2003, 7 Sep 2004, and 7 Sep 2005 were used for the calibration of the model (calibration data set). Calibration refers to the estimation of the model coefficients. These coefficients relate δ13C with ΨStem for the studied block and for all the years. The accuracy of the calibration was estimated across all sites and dates using the standard error of calibration (SEC) and the proportion of variance (R2) explained by the model. The SEC was computed as
(Eq. 3)
where n is the number of sites and m is the number of available dates, Ψ(s,t) are the measured values of ΨStem for each location (s) and date (t), and Ψ̂(s,t) are the estimated values of ΨStem for each location (s) and date (t).
Model validation.
Using the calibrated model, ΨStem values were predicted at a date (8 Sep 2004), which was not included in the initial data set for calibration. The robustness of the model was tested using the standard error of prediction (SEP) and the proportion of variance (R2) explained by the model. The SEP was computed as
(Eq. 4)
where n is the number of sites and m is the number of available dates, Ψ(s,t) are the measured values of ΨStem for each location (s) and date (t), and Ψ̂(s,t) are the predicted values of ΨStem for each location (s) and date (t).
Sensitivity of choice of site of reference.
Sensitivity analysis was performed to determine whether a random selection for the site of reference was a valid method. Each site of the block was considered as a potential site of reference, with one spatial model calibrated per site. To show the benefit of using a spatial model instead of using a more classical approach, the results of the spatial model were systematically compared to a nonspatial model. For the nonspatial model, the average of three sites chosen at random was considered as the estimated value for the whole block. The random selection of three sites was repeated 1000 times, resulting in the computation of 1000 models. The SEC was computed to evaluate the accuracy of each model and a curve of probabilities of error was computed for each model.
Data mapping.
Data mapping was performed using 3DField Contour Map Software (ver. 2.9.0.0, Vladimir Galouchko; http://field.hypermart.net/). The interpolation method used in this study was based on a deterministic function (inverse distance weighting). For each map, three classes of value were created: low, corresponding to the 0 to 33% quantile; medium, corresponding to the 33 to 66% quantile; and high, corresponding to the 66 to 100% quantile. This classification is relative to each map. Maps have only been used to visualize the results of the analysis, for which three classes were considered sufficient.
Results
Intrablock variability.
Despite the limited size of the field, results showed significant within-field variability. Total leaf area varied among sites from 0.9 to 2.1 m2/m2 (Table 2). Leaf area standard deviation among sites ranged from 0.21 to 0.25 m2/m2 for different years. Pruning weight was similar among years, but varied more than 10-fold among sites, showing great spatial variability in vigor. Yield varied two-fold among years and three-fold among sites. Berry weight and berry sugar content showed greater variability among sites compared to variability among years. Berry weight varied more than 35% from one site to another. Sugar content varied up to 34 g/L from one site to another, which is approximately 2% in potential alcohol. These results agree with those observed in other plots and other conditions (Tisseyre et al. 2007, Arnó and Rosell 2009). They justify the relevance of a precision viticulture approach to take into account the spatial variability to produce a better decision support for viticultural management.
Variability of vine water status and vegetative and reproductive parameters in 2003, 2004, and 2005 in the experimental block.
Temporal stability of δ13C spatial patterns.
Seasonal effects on vine water status are directly reflected by δ13C measurements (Figure 2). The median δ13C value was the highest for 2005 season, indicating that plant water status from veraison to harvest was lower during that year. Median values of δ13C for 2003 and 2004 were lower (less water deficit) and similar, suggesting similar plant water status from veraison to harvest in 2003 and 2004. The dispersion of δ13C values was wider for the 2005 season.
Boxplot (median, quartile) of δ13C measurements taken in the block in years 2003, 2004, and 2005.
Despite a marked seasonal effect, the Kendall coefficient had a highly significant value over the three seasons (W = 0.87, p < 0.01), indicating the spatial patterns shown by the δ13C measurements were very stable from one year to another. The areas of the block with the lowest (0–33%), medium (33–66%), and highest (66–100%) δ13C values for each year are shown in Figure 3. These results suggest that δ13C measurements taken during any given season could be used to model the spatial variability of plant water status for any other season. Because measured values presented a wider variability in 2005, δ13C values measured during that season were chosen as ancillary data to estimate ΨStem for all seasons.
Percentile maps representing the spatial variability of δ13C for the three seasons of the experiment (2003, 2004, and 2005).
Calibration of the spatial model.
Results were obtained with a randomly chosen site of reference located on site 83 (Figure 1). ΨStem values measured on different sites were related to ΨStem values measured on the site of reference (Figure 4). Three sites were chosen to be representative of the variability observed within the field: S4, S44, and S47, indicating where high, medium, and low water restriction was observed, respectively. Despite the limited dates, the relationship can be considered as linear, with R2 values ranging from 0.92 to 0.98 for sites 47 and 4, respectively. A linear relationship exists between ΨStem values observed on the site of reference and ΨStem values measured for each site in the field, with R2 values ranging from 0.89 to 0.99 for the whole data set (results not shown). Hence, the relation between ΨStem values taken at the site of reference and ΨStem values in any other part of the block are stable over time. However, the slope of this relation is different for each site in the block (Figure 4). The empirical model presented in Equation 2 was thus considered relevant.
Linear relationships between ΨStem values observed on the site of reference (site 83) and ΨStem values observed on site 4 (s4), site 44 (s44), and site 47 (s47). Symbols identified as A, B, C, and D represent ΨStem measurement dates of 31 Jul 2003, 7 Sep 2004, 8 Sep 2004, and 7 Sep 2005, respectively.
Measured ΨStem was plotted against estimated ΨStem values obtained through the spatial model (Equation 2) (Figure 5). The model was applied to every date of measurement on the calibration data set. R2 of the model was 0.91 and SEC was 0.14 MPa. On this data set, a nonspatial approach based on assigning the mean ΨStem to the whole block at each date gave a R2 of 0.79 and a SEC of 0.18 MPa. Therefore, the fact of implementing the spatial model reduces the error of the estimated values for the whole range of measured ΨStem values. However, under low water restriction (ΨStem > −0.8 MPa), the model provides a slight underestimation of ΨStem values. The same correlation can be applied over the three seasons, despite different climatic conditions and different levels of water deficit (Figure 5).
Observed ΨStem against ΨStem values estimated by the nonspatial model (mean ΨStem values) and the spatial model based on δ13C data of 2005 and applied to the calibration dataset. Symbols identified as A, B, and C represent ΨStem measurement dates of 31 Jul 2003, 7 Sep 2004, and 7 Sep 2005, respectively.
Validation of the spatial model.
Validation of the model on an independent data set is shown (Figure 6). The model predictions were performed for the second measurement date in 2004, which was not used for the calibration. The model R2 value was 0.80 and the SEP was 0.08 MPa.
Plot of observed ΨStem against ΨStem values predicted by the spatial model based on δ13C data of 2005 and applied to an independent data set (8 Sep 2004).
The proposed spatial model, based on a linear relationship, provided a better estimation of ΨStem than assigning the mean ΨStem to the whole block (nonspatial model). The standard deviation of ΨStem measurements on the date used for validation was 0.16 MPa. This value corresponded to the error produced on this particular date when spatial variability was not taken into account. In contrast, the SEP obtained by the spatial model was substantially lower, 0.08 MPa. When comparing those two results, it is important to note that all available ΨStem measurements taken on 8 Sep 2004 (48 sites) were used to compute the mean ΨStem used for the nonspatial model. It is not realistic in commercial management situations to compute a block average with such a large data set. In contrast, the spatial model was applied using ΨStem measured at a single location in the field—the site of reference—which, in practical terms, is very realistic.
Sensibility to the choice of the site of reference.
The probability of error with the spatial model, when the site of reference (mean of three ΨStem measurements from a single site within the block) was randomly selected, was compared to the probability of error with the nonspatial model, when an average of three sites (ΨStem measurements from three distinct sites within the block) were randomly selected, to estimate the value of the whole block (Figure 7). For both methods the choice of the reference site influenced the quality of the model. The SEC ranged from 0.12 to 0.30 MPa with the spatial model and from 0.15 to 0.39 MPa with the nonspatial model.
Study of the sensibility of the spatial and nonspatial model to the choice of the site(s) of reference. Plot of probabilities (y axis) of obtaining a standard error of calibration (SEC) lower than a given threshold value (x axis).
For example, when the spatial model is used, the probability of obtaining a SEC <0.15 MPa is 0.45, while a similar error cannot be obtained with the nonspatial model (the probability of a SEC <0.15 MPa is 0 for the nonspatial model). With a threshold of 95% probability, the spatial model provides a SEC <0.22 MPa while the nonspatial model provides a SEC <0.32 MPa. In general terms, the spatial model always provides lower probabilities of reaching a given error value than the nonspatial model (Figure 7). Thus, the probability of obtaining a better estimate is always higher with the spatial model, whatever the choice of the site(s) of reference.
Sensibility to the choice of δ13C measurements.
According to the results (Table 2), δ13C measurements taken on any single season could be used to model spatial variability of plant water status during any given season. δ13C was measured during three seasons. In order to test the model sensibility to the choice of the season during which δ13C measurements were performed, the spatial model was tested using ancillary data δ13C measurements acquired each season (Table 3). The best performance of the model was obtained when using δ13C measured in 2005 as ancillary data, which corresponded to the year when the vines were most stressed (Figure 2). However, the spatial model remained accurate even when δ13C from other seasons was used as ancillary data.
Sensitivity of the spatial model to the season of δ13C measurements. Standard error of calibration, SEC (MPa), and coefficient of determination, R2, obtained by the spatial model based on δ13C from each season.
Discussion
This study was carried out in a vineyard block with high spatial variability of vegetative and reproductive parameters (Table 2). We hypothesized that the modeling approach proposed by Acevedo et al. (2010) could be applied under moderate water deficit conditions by adapting both ancillary and reference plant water status measurements. The results demonstrated the validity of this hypothesis. Ancillary vigor-based measurements were replaced by δ13C measurements (water status-based). Reference ΨPD measurements were replaced by ΨStem measurements. As a result, calibration of the model provided a good estimation of ΨStem measurements (R2 = 0.91; SEC = 0.14 MPa) under moderate water deficit conditions (Table 1).
Using this modeling approach, δ13C measurements, taken at harvest on grape juice, can be used as ancillary data to extrapolate plant water status in subsequent seasons (Figure 5). Once the model coefficients have been determined, within-block spatial variability of plant water status can be estimated combining ΨStem measurements obtained at one single site with δ13C measurements taken on a regular grid at a previous date or previous season. This approach further increases the informative value of δ13C measurements since they can be used over several seasons. δ13C measurements have typically been used to estimate plant water status spatial variation during a given season (van Leeuwen et al. 2010), providing one single, integrative map per season. The use of δ13C maps as a tool to improve vineyard management has been limited by the delay needed for δ13C analysis (well after harvest) and by the cost of the maps, given the cost of each analysis and the numerous δ13C measurements needed to create a map.
In this study, δ13C measurements were taken during three seasons and tested as ancillary data for a spatial model. The best results were obtained with δ13C measurements taken during the driest year (Figure 2, Table 3). The practical implication is that spatial models should be preferably based on δ13C measurements taken during drier seasons. Furthermore, δ13C measurements could be combined with other ancillary measurements. This combined approach would improve the modeling of plant water status estimates, thus complementing δ13C measurements when δ13C data are obtained during a year of low water restrictions. It has been shown that δ13C values measured on grape sugar at ripeness best correlate with ΨStem measured 4 to 6 weeks before harvest (Santesteban et al. 2012). However, because spatial variability of vine water status is stable over the season, the spatial model can be applied at any given time after water deficits have developed.
Using the spatial model, a slight underestimation of the ΨStem values was observed when ΨStem > −0.8 MPa (Figure 5), suggesting that the relationship between δ13C and ΨStem might be better fitted with a nonlinear function. However, according to a study on physiological thresholds for irrigation, targeting ΨStem values near −1.2 MPa seems to be optimal for fruit phenolic composition and water use efficiency (Romero et al. 2010). Consequently, the model underestimation occurs for a range of values that are not relevant to trigger irrigation and as such, this model limitation should have no negative impact on production strategies.
A key challenge to developing a practical application from the proposed model is that it requires a specific calibration for each block, as observed elsewhere (Acevedo et al. 2010). Three points are important to focus on to optimize the model calibration: (1) the choice of the reference site, (2) the number of calibration sites, and (3) their location. In this work, the location of the reference site was randomly chosen. Results indicate that model quality is affected by the choice of the reference site (Figure 7; SEC varied from 0.12 to 0.30 MPa). A simple and practical recommendation to overcome this drawback would be to select several sites as potential reference sites during the first year of the study and then choose the most appropriate one thereafter.
In the current study, the calibration of the model was computed using 96 ΨStem measurements taken at three dates during three subsequent seasons. The acquisition of such a large database is not realistic for commercial applications. Theoretically, the calibration of the spatial model used in the current work (Equation 2) requires the measurements of ΨStem at the reference site and at only two additional sites within the block. This problem was addressed by researchers who have developed and successfully tested a sampling method based on ancillary data to optimize the number and location of ΨPD measurements needed for model calibration (Herrero-Langreo et al. 2012). This method allowed the calibration of a spatial model for plant water status estimates (assessed by ΨPD) even when ΨPD was measured at only three locations within the block.
Two additional aspects, related to temporal resolution of plant water status estimates, could be further improved. In this work, temporal resolution of plant water status measurement is limited by the frequency at which plant water status can be measured at the reference site by means of water potential. Consequently, the inherent nature of water potential measurements (destructive, short period of measurement during the day, manual reading) limits the temporal resolution of plant water status estimates. One way to overcome this limitation would be to substitute pressure bomb measurements performed in the reference site by automated measurements of vine water status. Real-time and continuous measurements of vine water status at the reference site could be obtained by automated sensors, such as sap flow sensors. Such an approach would be a good step toward the real-time monitoring of plant water status at the whole vineyard scale.
In addition, the spatial resolution of the estimated plant water status is limited by the number of δ13C measurements taken in the field. In the current work, δ13C was measured at 96 locations in a 0.23 ha block. In a commercial situation, the affordable resolution of measurements of δ13C per block will necessarily be lower. For practical applications, other ancillary data with high spatial resolution, such as NDVI images, could be used to minimize the number of δ13C measurements per block. First, high-resolution ancillary data could be used to identify uniform management areas within the block. Second, plant water status measurements taken from a single reference site could be extrapolated to each uniform management area by measuring δ13C in a single location within each area.
Moreover, δ13C measurements open the possibility to apply the spatial modeling approach (Acevedo et al. 2010) to a whole vineyard or denomination scale, as δ13C is directly related to plant water status and only marginally modulated by cultivar (van Leeuwen et al. 2001). It is very likely that δ13C is not influenced by rootstock, vine age, or training system in another way than their effect on vine water status, suggesting that the proposed modeling approach could be applied to extrapolate vine water status from a single reference point to a series of vineyards blocks using δ13C. A necessary condition to enable such wide-scale extrapolation imposes that climatic conditions are similar over the whole area. Further research will investigate this approach.
This method to characterize spatial variability of vine water status should enable the implementation of differential management practices within the vineyard according to vine water needs (Centenari et al. 2012, Palliotti et al. 2011). Under irrigated conditions, differential irrigation can be applied to target only the vineyard areas where the water deficit is too severe. This application could help to limit the use of irrigation water and may help in improving berry composition, which is adversely affected by under- and overirrigation.
Conclusions
This study offers a new decision-support tool to improve the management of intrablock operations aiming at optimizing vine water status. The proposed approach enables updating stem water potential maps from plant water status measurement obtained at only one reference site in combination with δ13C measurements taken in a regular grid during a previous season. This tool will help to identify when and where to perform vineyard operations, such as leaf pulling, fruit thinning, or irrigation. Several limitations remain before a practical application of the method is offered and further research is needed to investigate how to implement an optimized sampling method to minimize the number of measurements required to calibrate the model, to suggest a rational choice of the reference site based on high spatial resolution data, and to improve the temporal resolution of the model by using real-time monitoring systems of vine water use (such as sap flow sensors) located at the reference site.
Acknowledgments
Acknowledgments: The authors are grateful to Ghislaine Hilbert (INRA Bordeaux) and Jean-Philippe Roby, Elisa Marguerit, and Encarna Cuevas (Bordeaux Sciences Agro) for help with data collection.
- Received October 2012.
- Revision received January 2013.
- Revision received April 2013.
- Accepted May 2013.
- Published online December 1969
- ©2013 by the American Society for Enology and Viticulture
Literature Cited
Vol 64 Issue 3
Jump to section
Related Articles
Cited By...
More from this TOC section
Similar Articles
