Precision agriculture is a useful tool to assess plant growth and development in vineyards. The present study focused on spatial and temporal analysis of vegetation growth variability, in four irrigation treatments with four replicates. The research was carried out in a vineyard located in the southwest of Spain during the 2012 and 2013 growing seasons. Two multispectral sensors mounted on an all-terrain vehicle (ATV) were used in the different growing seasons/stages in order to calculate the vineyard normalized difference vegetation index (NDVI). Soil apparent electrical conductivity (ECa) was also measured up to 0.8 m soil depth using an on-the-go geophysical sensor. All measured data were analysed by means of principal component analysis (PCA). The spatial and temporal NDVI and ECa variations showed relevant differences between irrigation treatments and climatological conditions.
Terroir is a French concept that means that “there are unique aspects of a place that shape the quality of grapes and wine”. Those aspects that impact on grapes and wine quality are usually associated with topography, soil, climate, plant management and plant genetics (Vaudour, 2002). According to several authors, the study of plant vegetative vigour is an essential parameter to successfully manage yield and grape/wine quality because plant growth integrates climate, soil, topography, available water and other plant controlling factors (Carbonneau, 1995; Cortell et al., 2005; Deloire et al., 2005; Smart, 1985). Consequently, appropriate management of soil and consideration of the main climatic variables are key factors to obtain good yields and, ultimately, quality wines. Vineyard canopy management practices such as pruning systems, shoot orientation, shoot thinning or leaf removal, have the capacity to modify climate factors around the plant and, consequently, to modify grape and wine quality (Dry, 2000).
Vineyard behaviours with regard to water management have been studied in recent decades in a wide range of environments and vineyard varieties because of the implications of irrigation on yield and quality of the final product (Smart and Coombe, 1983; Williams and Araujo, 2002; Mullins et al., 1992; Bravdo and Hepner, 1986; Intrigliolo and Castel, 2010). Previous authors also indicate that vine vegetative development is highly influenced by water availability, to the extent that this may become a limiting factor. However, with the same irrigation depth, sometimes the response between two close plants is not the same. This point should be considered when selecting methods to estimate crop water status in order to achieve better management and to meet the production objectives defined at the beginning of the growing season. On the one hand, covering all water needs is not recommended because this creates management problems, reduces crop quality and in general increases unnecessarily the cost of cultivation. On the other hand, increasing water availability to the vineyard causes not only grape production to rise, but also the costs of pruning and plant protection treatments, and usually results in reduction of grape quality. Thus, water stress has to be controlled to achieve good yield and quality of grapes as well as balanced growth while avoiding the problems of excess water. It is essential, therefore, to know the correct way to manage this crop.
Several studies related to spectral vegetation indices (VIs) have carried out analyses of vine canopy, shape, size and functional capacity in order to determine spatial and temporal management of vegetation as well as other productions factors such as water. Spectral VIs are able to predict a large number of plant features, such as leaf area index (LAI), vegetation fraction cover, fraction of absorbed photosynthetically active radiation (fAPAR), chlorophyll pigment concentration, plant stress and other related parameters (Baret and Guyot, 1991; Gitelson and Merzlyak, 2004; Jordan, 1969; Peñuelas et al., 1993; Rondeaux et al., 1996). These VIs, which are mathematical combinations of two or more plant reflectances at specific wavelengths, can be used in vine growth site-specific management enabling the optimization of grape quality and yield (Lamb and Bramley, 2001).
Nowadays, it is possible to obtain a plant spectral signature with a
multispectral proximal sensor (Tardáguila and Diago, 2008) suitable for
studying vine vegetation terroir. The normalized difference vegetation index
(NDVI), developed by Rouse et al. (1973), is one of the most extensive
indices used for vegetation growth analysis. It can be calculated as
Terroir is also affected by physical, chemical and biological soil properties; as a tool to interpret these soil property variations, soil apparent electrical conductivity (ECa) may be used. Soil ECa measurements can characterize soil spatial variability, with regard mainly to the physical features of the soil, and have been used by other authors to delineate homogeneous management zones (Terrón et al., 2013; Corwin and Lesch, 2003; Moral et al., 2010). Soil ECa measurements can be obtained through geoelectric sensors and this can be an easy and economical way of sampling the soil and guiding soil evaluators in their soil property analyses (Terrón et al., 2011).
According to Hall et al. (2002) the implementation of vineyard site-specific tools are needed in order to better manage vineyards. Thus, the present work makes use of precision agriculture tools to determine (i) the effects of different irrigation treatments on vine vegetation growth in two different climatic seasons and (ii) influence of the soil on vegetation growth expression.
The study was carried out during the 2012 and 2013 growing seasons, in a
field belonging to the Agrarian Research Institute “La Orden –
Valdesequera”, in Extremadura (Spain) (38
The study area is located in a vineyard of 1.8 ha, varietal Tempranillo
(
The field is situated in the Guadiana River valley, whose soil morphology is typical of the Quaternary, carved in Tertiary sediments. Surface horizons have been artificially transformed for agricultural use with water management, which has resulted in destruction of these horizons and in burial of diagnostic horizons, even below 50 cm. These latter horizons are not differentiated and they have evolved edaphically from the sediments and materials of the lower terraces of the Guadiana River, which cover the coarser sediments. According to Soil Survey Staff (2006), the soil is in the order Entisol, suborder Orthent and the great group Xerorthent (Xeric).
There exists a soil study consisting of a 15-point database randomly distributed on the study area. Tables 1a and 1b show statistic values of soil samples of some physicochemical parameters at two depths (from 0 to 0.30 m and from 0.30 to 0.60 m respectively) which were analysed by official laboratory procedures.
From our measurements of a nearby well, we verified that there exists a water table placed at a depth of 6 m, in accordance with the river water level. However, from our experience in this area, we know that the level can vary due to rainfall and water use for irrigation.
The experimental design was randomized complete blocks, with four replicates
(plots) per treatment. Each plot had 108 vines in 6 rows with 18 vines per
row, where the distance between plants and rows were 1.20 and 2.50 m
respectively, placed on a trellis with an east–west row direction. Water
treatments were dependent on the growing season (Fig. 1):
2012 treatments were divided into
four levels of irrigation, corresponding to four levels of crop
evapotranspiration (ETc) rates:
fully watered, based on the application
of 100 % ETc; regulated deficit irrigation
(RDI) 50-20, based on the regulated deficit irrigation technique, with
50 % ETc before RDI 50-0, based on the regulated deficit
irrigation technique, with 50 % ETc before non-watered, based on rainfed treatment; and 2013 treatments were reduced to three levels of irrigation,
corresponding to three levels of ETc rates:
fully watered, based on the
application of 100 % ETc; RDI 30, based on the regulated deficit
irrigation technic, with 30 % ETc throughout the season; and non-watered, based on rainfed treatment.
Maps of treatments and respective plots:
The irrigation system was set up by drip irrigation with one emitter of
4 L h
Soil management was characterized by two annual cultivator treatments: one in
winter dormancy and another at the bud break phenological stage. Later,
spontaneous vegetation was controlled by herbicide treatments and
250–350 kg ha
The NDVI estimation was performed with two active proximal multi-spectral
sensors mounted on an all-terrain vehicle (ATV). These sensors (OptRx
ACS-430, Ag Leader Technology, USA) report directly the vineyard canopy NDVI
calculated with red (0.67
ATV with two multi-spectral sensors for NDVI mapping of vineyard canopy.
To validate the NDVI with the LAI, several measurements of the latter were taken throughout the ripening stage of the crop in both years. Measurements were recorded by a Plant Canopy Analyser LAI-2000 (LI-COR, Inc, USA), following the procedure of Mabrouk and Carbonneau (1996).
ECa measurements were conducted on 18 February 2011, with a VERIS 3150 Surveyor sensor (Fig. 3), simultaneously in two different soil levels: (i) shallow or ECs – to a depth of 0.30 m from the soil surface; and (ii) deep or ECd – to a depth of 0.80 m from the surface. Sampling details can be found in Moral et al. (2010).
The samplings shown in this work, corresponding to each data set of both growing seasons, were statistically analysed by means of ArcGIS v.10.1 software (ESRI, USA) for geostatistical analyses, and SPSS v.17 software (SPSS Inc., USA), for inferential statistics analyses.
Parameters corresponding to the theoretical semivariograms for the NDVI samplings in 2012 and 2013 growing seasons and for the CEa samplings in 2011 growing season.
The geostatistical analysis of the multi-temporal NDVI samplings included the
followings phases:
Voronoi map – a previous exploratory analysis of
the samplings was performed to extract outliers. ordinary Kriging
interpolation – the parameters used in the semivariograms of each sampling
to generate the corresponding maps are shown in Table 2. Once obtained, these
maps were rasterized using a pixel size of 2 m. principal component analysis (PCA) – in this work, a PCA process
was established separately for each of the years of study.
At each analysis, input raster data set included the five NDVI samplings of
the growing seasons, and the output data were distributed in five principal
components. Thus, the results of the PCA analyses obtained consisted of five
principal components for each year, where the first principal component shows
the NDVI spatial variability for all the mapping dates of each year.
Meanwhile, the ECa samplings were also geostatistically analysed. In this case, only the ordinary Kriging interpolation tool was used, from which the ECs and ECd maps of 2011 were obtained. The parameters used to interpolate the ECa samplings are also shown in Table 2.
Mobile sensor platform Veris 3150 for ECa mapping.
The NDVI samplings from both growing seasons and the ECa samplings at both depths acquired by Kriging were statistically analysed in two phases: (i) firstly, descriptive parameters of each water treatment at each sampling date were acquired to obtain global knowledge of the behaviour of each of the components that make up the statistical design; (ii) secondly, variance analyses of each treatment at each sampling date were made. These analyses allowed comparison of the previously mentioned spatial and temporal behaviours.
CEa, shallow and deep, and first principal component of the NDVI at soil sample localizations.
To analyse relationships among all the variables studied, values of the first principal component (PC1) of the NDVI variables in both sampling sequences (years 2012 and 2013) and of ECa (ECs and ECd) in the sampling points from the raster maps (Figs. 6 and 7) were extracted and are shown in Table 3.
Finally, in order to determine the importance of the local soil
characteristics, given by the ECa and NDVI parameters respectively, in the
vegetative expression of the vineyard, the geographically weighted regression
(GWR) tool, included in ArcGIS v.10.1 software (ESRI, USA), was used. The
relationship between both variables resulted in maps of determination
coefficient (
Climatic variables logged by the weather station situated in a reference field near the tested vineyard recorded diverse behaviour during the 2-year test, with drier conditions in the first growing season. Figure 4a and b show cumulative annual rainfall, cumulative annual ETc, temperature parameters and growing degree days (GDDs) on both years. Focusing on the accumulation of precipitation, the total amount in the second year trial (2013) was more than double compared to the first year trial, where only in its first quarter it had the same amount of rainfall as the whole previous season. However, during the final stages of vegetative development and in the whole ripening phenologic stages, both years had a similarly low accumulation of precipitation. In fact, the temperature was not very different between both years. The observed climatological differences in both seasons influenced differentially the vineyard vegetative development with respect to the different irrigation treatments analysed in this study.
Despite the large difference in precipitations between the two growing seasons, however, for the second year of the test, which was the wettest, hydric demand was similar to the previous year. This result permitted comparison of vegetative response of two consecutive years that were very different in their climatology. Furthermore, if this premise remains constant over the years, it could be possible to know the total needs of the culture of vineyards under whatever climatological conditions, and appropriate reductions could be made in ETc for a watering schedule based on precipitation occurring at each moment of the campaign. Obviously, as Wample and Smithyman (2002) have reported, increases in hydric necessities at each phenological stage must be taken into account, as shown in the slope changes of the accumulation curve of ETc (Fig. 4a), and care must be taken in dry seasons not to bring about unwanted water stress.
Correlation matrix (
In this study, Fig. 5 shows the relationship between LAI estimations and NDVI
measurements at the ripening stage of grapes in both years. It can be
confirmed that they are well related (
NDVI–LAI relationship of both 2012 and 2013 years.
Soil properties, spatial variation of ECa, both ECs and ECd, and PC1 of NDVI
(both 2012 and 2013 growing seasons) were statistically analysed (Table 4).
On the one hand, ECs and ECd results were correlated with the clay content
from 0 to 0.30 m (
Results of ECa spatial variation (Fig. 7) seem to be a pattern consisting of a variation in ECa from the northern and southern boundaries of the assay up to the centre, and also from east to west, coinciding with some physicochemical parameters of soil. There exists, too, a pattern in the variability of soil characteristics due to the good relationship with ECa, in particular with clay content (Moral et al., 2010). Spatial variability of ECa, both shallow and deep, had also shown significant differences among the locations of the plots of the different irrigation treatments (Table 5), with different values for the soil properties that influenced vegetative growth of the grapevines. In general, the spatial variability pattern mentioned above was observed in the plots with the different treatments, with higher ECs or ECd values in those plots near the northern and southern boundaries of the vineyard test site. Because of this spatial variability, it was necessary to perform, even within plots of the same treatment, geostatistical analyses between NDVI and ECa to determine the extent of the influence of soil properties on vegetative growth of the vineyard in each of the irrigation treatments and their respective plots.
NDVI first principal component of
Interpolated apparent electrical conductivity maps of 2011 growing
season:
Statistic descriptive analyses of shallow and deep soil ECa interpolated data.
Sampling was carried out on 18 February 2011.
Both temporal and spatial evolution of the NDVI index of the irrigation
treatments and their respective plots in the 2012 growing season is shown in
Fig. 8. At first glance, the results of NDVI mapping for this year show how
all the treatments had a temporal evolution similar to a Gaussian function,
the mean value of the index increasing as the campaign advanced, reaching a
maximum value around the phenological stage of
The intermediate RDI 50-20 and RDI 50-0 irrigation treatments also showed significant differences between NDVI values with regard to the previous ones, with intermediate values. Both RDI treatments kept their NDVI values similar up to January, and then they became different as a result of the change in the water dose of the experimental design. At that moment, the RDI 50-0 treatment had a greater decrease in NDVI mean value and, consequently, in vegetative expression of the vineyard. Taking into account these aspects, and bearing in mind the existing relationship between vegetative growth of the vines and NDVI value, it can be considered that the latter increased its value when water doses were higher, and that variations in doses will result in changes in the vegetative expression of the vineyard.
Statistic descriptive analyses of the NDVI interpolated data sets for 2012 and 2013 growing seasons (dimensionless).
On the other hand, despite the relation found between the water doses applied in the assay and the vegetative development of the vines, there were significant differences among the various plots of each water treatment (data not shown), indicating that spatial variability of the NDVI index existed, and consequently of vegetative growth too, which was dependent on other factors, even though the characteristics of the management were identical. In this respect, it can be observed in Fig. 8 how vegetative expression was not homogeneous in all the plots within a specific water treatment, but rather variations were found in the NDVI value depending on the geographical location of each of the plots. Thus, on a specific mapping date, some plots with different water treatments had similar mean values of NDVI, even between plots with fully watered and non-watered treatments. A factor associated with geographical location, therefore, has had some influence on the vegetative growth. The terroir effect, in which the physicochemical parameters of soil are included, could be one of the factors that have caused a certain influence on vegetative development, as indicated by van Leeuwen and Seguin (2006).
A priori, the global results on the relationship between NDVI and ECa indicate a low association when compared in the first 0.30 m of soil depth (ECs, Table 7), and a relatively high one when a large section of soil is considered (ECd, Table 7). These results suggest that the soil surface layer has little influence on vegetative expression of the vineyard because of the deeper distributions of the roots, although it does influence other crops with shallow roots (Fortes et al., 2014). Furthermore, in the year when climatic quality involved drought (2012), ECa and NDVI values were lower, suggesting that the soil properties seem to be an influential factor but not a limiting one in vegetative expression, the availability of water resources being the principal limiting factor.
Interpolated NDVI maps of 2012 growing season:
NDVI maps year 2013:
Figure 9 shows the spatial and temporal evolution of NDVI in the watered
treatments and their respective plots in the 2013 growing season. In the same
way as the previous year, increased water doses applied to the vineyard were
associated with a higher NDVI mean value. However, in this season, the
differences in this mean value were closer, being no higher than 0.10 points
of index value. The intense precipitations between post-harvest of 2012 and
flowering of 2013 decreased the possibility of water stress in the vines, so
vegetative development was very similar at the beginning of the NDVI
mappings, the only difference being the RDI 30 treatment that came from the
RDI 50-20 of the previous growing season (Table 6). On the other hand, in the
2013 season, temporal evolution of the mean NDVI value of the whole
treatments was more homogeneous for most of the season. Generally speaking,
there was an initial increment of the NDVI value in all treatments up to the
phenological stage of
Local
With regard to temporal behaviour of NDVI between water treatments, the mean
value of the index resulted in slightly higher significant differences as the
season progressed, with two different groups of treatments at
The irrigation treatments of the 2013 growing season also had significant differences in mean NDVI values among their respective plots spatially (data not shown), with a pattern of reduced values from north to south of the vineyard test area. Thus, for the same water treatment and mapping date, the mean NDVI values of each plot decreased the further south the plot was located, with, furthermore, significant differences among them. This result was already shown by Blanco et al. (2012), who reported that vegetative growth of vines under the same management had different behaviours due to spatial changes in some influential factor, such us spatial variability of the physicochemical properties of soil. On the other hand, the influence of terroir, taking into account its climatic and edaphic factors, was so high in the 2013 season that it gave rise to similar mean NDVI values, with some exceptions, in plots with different irrigation treatments. Thus, for example, northern plots of fully watered and non-watered treatments gave similar NDVI values, as did the southern plots, but these values were statistically different between the two geographical locations. This behaviour can be seen in Fig. 9.
Figure 10 shows the local relationship between the PC1 of NDVI in each
growing season and the ECa in 2011, both shallow and deep, throughout the
test area, which is the level of influence of soil features on vegetative
development in each water treatment. The highest ratios prevailed, again, in
the northern and southern limits of the test area, in agreement with those
zones where ECa reached the lowest values. Thus, the maximum values in the
relationship between soil properties and vegetative growth were obtained
during the 2013 season; values of
Correlation matrix (
Correlation matrices between 2012 and 2013 NDVI surfaces of each irrigation treatment.
The results of each mapping date of NDVI in both growing seasons (Figs. 8 and 9) show the behaviour of the vegetative development of the whole treatments established in the experimental designs. As has already been stated, NDVI values and, accordingly, vegetative growth of the vineyard were influenced by soil properties (including groundwater level) in spatial components, and by climatic features in temporal ones.
With regard to the temporal variability, Fig. 6 shows the results obtained in the first principal component (PC1) of each PCA made for the different mapping dates in each growing season. This PC1 shows spatial variability of NDVI for the whole of NDVI mapping dates for each year. Thus, each PC1 map for 2012 explains 80.57 % of the temporal variability of each geographical location within the assay area, and 85.92 % for the 2013 growing season. Thus, the PC1 for each year shows more than 80 % of the mean variability of the NDVI values throughout both seasons in each of the irrigation treatments and their respective plots. In general, the PC1 map for 2013 shows higher and more homogeneous values than the 2012 one, indicating a higher and more homogeneous vegetative growth of grapevines.
Table 8 shows the level of relationship of NDVI values between the different
mapping dates for each irrigation treatment. Generally speaking, in both 2012
and 2013 there was an increase in the determination coefficient (
With respect to differences in spatial variability of vegetative growth between the years tested, the 2013 season showed greater homogeneity, where the highest increase was found in the northern half of the test area, regardless of the water dose applied. Conversely, this vegetative development was lower in the south, where the southern plot of non-watered treatment did not have lower vegetative growth, although it did respond to a spatial pattern. Thus, the vegetation response in 2012 was more dependent on the irrigation treatments, while in 2013 it was more dependent on soil characteristics or other edaphic–climatic variables. In 2013, RDI 50-20 and RDI 50-0 treatments became RDI 30a and RDI 30b respectively, with water doses of 30 % of ETc during the whole irrigation period. In the same way that the rest of the treatments had higher NDVI values in 2013, RDI 30 also showed higher NDVI values than the RDI treatments of the previous season. However, despite having the same water dose, RDI 30b gave lower values than RDI 30a during most of the season (data not shown), thus suggesting once again that water dose must be redefined taking into account climate and soil properties.
According to Howell (2001), there must be an optimal method of crop management in any situation, in order to obtain the desired yields and qualities, but intra-year and between-year management must be performed in accordance with the terroir features of each year.
Water level and vegetative growth were clearly related; greater availability
of water resources gave rise to greater vegetative development of the
vineyard. However, spatial–temporal changes in climatic quality or in soil
properties also affect vegetative expression. To the already estimated
differences in vegetative growth of grapevines between different water doses,
one must add the effects that climate and soil properties have on plants.
Consequently, the application of the same cultural practices in each growing
season makes it unfeasible to attain stable goals, i.e. the same level of
quality in grapes and wines or similar yields every season. The application
of some precision agriculture techniques to the vineyard crop, through
real-time measurements of NDVI and ECa, makes it possible to determine
homogeneous zones of growth and development in the vineyard dependent on
climatic and soil characteristics for a specific irrigation treatment. Thus,
according to the results of this study, the following can be concluded:
In global terms, the higher the
water doses the higher the NDVI values and, hence, the greater the vegetative
growth of the vineyard. Vegetative development is not homogeneous, even
when the same cultural practices are being used, but spatial and temporal
variability occur depending on climatic and soil characteristics and their
interactions. It is necessary for crop management to adapt to the
variability of agronomic factors in order to achieve homogeneous vegetative
growth even in zones where the soil characteristics are different.
An irrigation schedule based on real-time NDVI results, and knowledge of the
variability of soil characteristics could be the basis for improved vineyard
management.
This work was carried out with funding from the RITECA Project, Transboundary Research Network Extremadura, Center and Alentejo, co-financed by the European Regional Development Fund (ERDF), by the Spain–Portugal Border Cooperation Operational Programme (POCTEP) 2007–2013 and by the Government of Extremadura.
This research was also co-financed by the Government of Extremadura and the European Regional Development Fund (ERDF) through the project GR10038 (Research Group TIC008).
The vineyard irrigation project which has complemented this work is INIA RTA2009-00026-C02-02 and was co-financed by the European Regional Development Fund (ERDF). Edited by: E. Costantini