Abstract
Soil texture, topographical data, fruit zone light measurements, yield components, and fruit composition data were taken from 125 locations in each of four Vitis vinifera L. cv. Cabernet Sauvignon vineyards in the Lodi region of California during the 2017, 2018, and 2019 seasons. Data were compared against three sources of normalized difference vegetation index (NDVI) with different spatial resolutions: Landsat 8 (LS8NDVI; 30 m), Sentinel-2 (S2NDVI; 10 m), and manned aircraft (at high resolution, HR) with the interrow removed (HRNDVI; 20 cm). The manned aircraft also captured canopy temperature (CT) derived from infrared (thermal) wavelengths (HRCT; 40 cm) for additional comparisons. HRNDVI was inversely related to HRCT, as well as to several chemical components of fruit composition including tannins and anthocyanins. While some constituents of fruit composition such as anthocyanins may be related to NDVI, canopy temperature, and/or indirect measurements collected in the field, results presented here suggest that yield and fruit composition have a strong seasonal response and therefore environmental conditions should be considered if more accurate predictions are desired. Furthermore, freely available public satellite data sources with mixed canopy and interrow pixels, such as Sentinel-2 and Landsat 8, provided similar information related to predicting specific fruit composition parameters compared to higher resolution imagery from contracted manned aircraft, from which the interrow signal was removed. Growers and wineries interested in predicting fruit composition that accounts for spatial variability may be able to conserve resources by using publicly available imagery sources and small numbers of targeted samples to achieve this goal.
- objective measures of fruit quality
- precision viticulture
- principal components analysis
- remote sensing of vegetation
- vineyard variability
- Vitis vinifera cv. Cabernet Sauvignon
The prediction of winegrape composition in a vineyard, and the resulting wine quality, is difficult for numerous reasons. Manual sampling of fruit quality in vineyards can be complicated and time consuming (Wolpert and Vilas 1992, Meyers et al. 2011), especially for operations comprising many vineyards or many differentiated products. Additionally, comprehensive measures of fruit composition are expensive, requiring either costly commercial lab submissions, or in-house laboratory equipment and personnel. Furthermore, spatial variability exacerbates both of these issues because patterns of variability have shown to be temporally consistent but the magnitude of variability of fruit composition from a vineyard is unlikely to be consistent (Bramley 2005, Arnó et al. 2012, Sams et al. 2022).
Individual destructive vine measurements have long been the industry standard for determining fruit composition and potential wine quality. Commonly used vine measurements to evaluate vineyard performance are berry sizes or mass (Gladstones 1992), total yield estimates (Keller et al. 2005), dormant season pruning weights (Smart and Robinson 1991), and chemical analysis of berries or whole clusters (Niimi et al. 2020). More recently, to describe within-vineyard zones of potentially similar vine performance with a few targeted samples, indirect measurements like apparent electrical conductivity (ECa) mapping have been shown to be useful in describing soil variability and its relationship with vine performance (Acevedo-Opazo et al. 2008, Bramley et al. 2011). Each of these measurements has an economic cost to vineyard managers and/or wineries, and few commercial operations collect sufficient samples to adequately characterize the full range of variability in a vineyard or have dedicated staff for time consuming surveys that are susceptible to changes in the environment or management practices over time (Ferreira et al. 2020).
Remote sensing can capture information on vineyard variability rapidly and repeatedly, but must be “ground-truthed,” or validated by measurements collected at the field level, for crop metrics to be useful (Sun et al. 2017), although targeting indirect measurements or samples to vineyard zones based on spatial variability can reduce sample requirements (Meyers et al. 2011). Because all available sources of remote sensing information must be validated by ground measurements, and high-resolution imagery captured by many commercial manned aircraft platforms requires additional spectral and geometric calibrations for quantitative assessment, it would be useful to understand the capabilities of different sensors in predicting fruit composition at harvest to reduce these costs. As imagery from highly calibrated public satellite platforms like the European Space Agency’s (ESA) Sentinel and the National Aeronautics and Space Administration’s (NASA) Landsat becomes available at increasing spatial and temporal resolution, and at no cost, imagery obtained from manned and unmanned commercial aircraft flying at relatively low altitudes may not be necessary for many applications, such as discriminating zones for differential harvests or zonal vineyard management.
Remote sensing analysis of vineyards began in earnest two decades ago as Dobrowski et al. (2002) used imagery acquired using manned aircraft, and Johnson (2003) used imagery from satellites as tools for understanding grapevine canopies. Lamb et al. (2004), Bramley et al. (2011), Hall et al. (2011), Trought and Bramley (2011), and Song et al. (2014) found relationships between canopy vigor and fruit composition using normalized difference vegetation index (NDVI) and plant cell density derived from imagery acquired from manned aircraft, while Sun et al. (2017) used Landsat-derived products (NDVI and leaf area index) to predict winegrape yield. Sozzi et al. (2020) showed significant relationships in vineyard canopy variability (NDVI) captured by Sentinel-2 versus high-resolution imagery from an unmanned aerial vehicle (UAV), with comparisons of interrow removed and included, but they did not relate these to measurements of yield or fruit composition. Given these findings, and the fact that several compounds found in Cabernet Sauvignon grapes known to influence wine chemistry are related to environmental conditions (Bergqvist et al. 2001, Kliewer and Dokoozlian 2005, Lee et al. 2007, Li et al. 2013, Martínez-Lüscher et al. 2019), we hypothesize that it should be possible to characterize the relationships among vine productivity, seasonal growing conditions, and remote sensing. Principal component analysis (PCA), a method commonly used to reduce dimensionality of large data sets, has been used by others to infer relationships between some aspects of yield and fruit composition with soil variables in Chardonnay in Ontario, Canada (Reynolds et al. 2013), and imagery, water status, soil characteristics, basic grape chemistry, and yield of different varieties in France (Acevedo-Opazo et al. 2008).
We present results of PCA and Pearson correlations combining data from four commercial vineyards in the Lodi region of California with different management strategies, terrains, soil types, and magnitudes of variability. Our objectives were to compare variables from different remote sensing sources with a large volume of spatially distributed and ground-truthed environmental and fruit composition measurements, as well as to determine the most useful and efficient ground-truthing descriptors of vineyard characteristics correlating to fruit composition and quality. To our knowledge, no study exists incorporating yield components, fruit chemistry, soil texture, fruit zone light, and remote sensing with high density ground validation. Additionally, the inclusion of Cabernet Sauvignon grape mouthfeel and aroma precursors differentiates the present analysis from previous studies relying mainly on measures of basic chemistry.
Materials and Methods
Four Cabernet Sauvignon vineyards were selected based on their range of geographic locations and assessments of spatial variability found in aerial imagery and were sampled for yield components and fruit composition at harvest for three consecutive seasons, 2017 to 2019 (Sams et al. 2022). The vineyards were in the American Viticultural Area (AVA) of Lodi, California and were within 40 km of one another and spatially distributed across the AVA. Climate of the region is classified as dry summer subtropical with an average of 190 mm of precipitation annually. Summertime highs regularly exceed 40°C and winter lows rarely dip below 0°C. Climate data were summarized from a centrally located weather station in the Lodi region for the three years in which data were collected (Table 1). Table 2 details the physical characteristics of each vineyard. Briefly, all four vineyards—Vineyards A through D—were drip irrigated and had sprawling, spur-pruned canopies. Vineyard C was planted in 1998, while Vineyards A, B, and D were planted from 2010 to 2013.
Yield and fruit composition. On the commercial harvest dates for each of the four vineyards set by the receiving winery in each season, a three-step process for documenting yield components was completed at each of the spatially distributed 125 georeferenced data vines in each vineyard. Sample distribution maps can be found in Sams et al. (2022). A combined 100-berry sample was collected by randomly selecting 10 clusters and removing 10 total berries from the top, middle, and bottom of each cluster and weighed. Twenty randomly chosen grape clusters were then removed from each data vine, weighed, and transported to the laboratory. Data vines were hand-harvested and total yield per vine and cluster number were recorded. Weights from all three steps were combined for total vine yield. Upon arrival at the laboratory, all berries from the 20 randomly chosen grape clusters were homogenized before extraction with acidified 50% ethanolic solution. Total anthocyanins were measured using a UV-vis based method described by Iland et al. (2000). Polymeric tannins and a combined total of quercetin 3-O-glucoside and 3-O-glucuronide, referred to hereafter as “quercetin glycosides,” were measured using reversed-phase high-performance liquid chromatography (HPLC; Peng et al. 2001). Analysis of free-form volatile compounds (C6 and 3-isobutyl-2-methoxypyrazine [IBMP]) was completed using headspace solid-phase microextraction coupled to a gas chromatograph and a mass spectrometer (HS-SPME-GC-MS) adapted from Kotseridis et al. (2008) and Canuti et al. (2009). Bound form β-damascenone was extracted using solid phase extraction adapted from Whiton and Zoecklein (2002), followed by fast acid hydrolysis and SPME-GC-MS as described by Kotseridis et al. (1999), Ibarz et al. (2006), and Canuti et al. (2009). Total soluble solids (Brix), pH, titratable acidity (TA), yeast assimilable nitrogen (YAN), and malic acid were measured using standard methods of Fourier-transform infrared spectroscopy and calibrated using E. & J. Gallo’s reference chemistry standards.
Vineyard physical characteristics. Soil cores were collected between seasons (Vineyards B and C in December 2018; Vineyards A and D in December 2019) from the center of the interrow space ∼1 m from each data vine using a 5.7 cm diam soil auger (AMS, Inc.). Each core constituted a single composite sample of soil depths from 0 to 100 cm taken at 20 cm increments, mixed into a 30.5 × 25.4 cm plastic bag, and submitted to a commercial soil analysis laboratory (A & L Western Laboratories) for particle size (texture) analysis. An apparent ECa survey was conducted in each vineyard prior to the initiation of vine growth in 2018 using a Dualem-1S (Dualem Inc.) and a Trimble Geo7x Global Positioning System (GPS, Trimble Inc.). ECa data were interpolated in VESPER using local block kriging (version 1.62; Australian Centre for Precision Agriculture, The University of Sydney, New South Wales, Australia). A digital elevation model developed from a range of sources and with a final resolution of ∼10 m ground spacing (1/3 arc-second) and a vertical root mean square error of 1.55 m (Gesch et al. 2014) was downloaded from the United States Geological Survey (USGS) and used for topographic analysis (USGS 2019, 3D Elevation Program Digital Elevation Model, accessed 21 March 2020 at https://elevation.national-map.gov/arcgis/rest/services/3DEPElevation/ImageServer).
Measurements of photosynthetically active radiation (PAR) in the fruit zone of each data vine were collected in the first week of June in 2018 and 2019 at bloom (modified Eichhorn-Lorenz [E-L] stage 23; Pearce and Coombe 2004), mid-June or fruit set (modified E-L stage 27), and mid-July or veraison (modified E-L stage 35), with measurements occurring only on days with relatively cloudless sky conditions. PAR measurements were taken within two hours of solar noon using an ACCUPAR LP-80 ceptometer (Meter Group, Inc.). In the two bilateral cordon trained vineyards (Vineyards A and D), the ceptometer was placed parallel to the vine cordons in the fruiting zones of the canopies and facing up and on both the north and south sides of the vines, with one above canopy (or ambient) measurement taken before and one after the fruit zone measurements. In the two horizontally divided quadrilateral-cordon trained vineyards (Vineyards B and C), one measurement was taken on the south side of each southern cordon and one measurement on the north side of each northern cordon, with ambient measurements taken before and after fruit zone measurements. The sensor console was aligned horizontally with the vine trunk in all vineyards, ensuring that the main arm from the trunk to the center of the vine was accounted for in each measurement. Measurements from each vine were averaged and fruit zone PAR (PARFZ) was calculated by dividing the average fruit zone PAR measurements by the average of the ambient PAR measurements. Additional ambient incident PAR measurements were taken in open areas with no overhead sensor obstruction every 15 min throughout each measurement period to ensure above canopy measurements reflected ambient conditions and were not obstructed by neighboring vine rows. Dormant season pruning weights were collected in the first two weeks of December in 2018 and 2019 from each data vine and the Ravaz index was calculated as the ratio of yield:pruning weight (Smart and Robinson 1991).
Remote sensing—manned aircraft. High-resolution imagery from manned aircraft (visible/near-infrared = 0.2 m ground resolution, wavelengths = 800, 670, 550 µm, bandwidth = 10 nm; thermal infrared/canopy temperature = 0.4 m ground resolution, wavelengths = 7.5 to 13 μm; absolute error ±1°C) was sourced from a commercial provider for each vineyard at phenological stages corresponding to the PAR measurements in 2018 and 2019 (modified E-L stages 23, 27, 35). NDVI was calculated from the near-infrared (800 nm) and red (670 nm) bandwidths (Rouse et al. 1974). Using the histogram for each HRNDVI and HRCT, a bimodal separation of pixels was used to delineate canopy pixels from pixels in the interrow and nonvine signals (Salgadoe et al. 2019). All noncanopy pixels, or those below the histogram separation values in the case of NDVI and above the histogram separation values in the case of thermal infrared, were removed, and the average pixel value of each data vine was derived from the remaining pixels. Image processing was completed using ArcGIS (v10.4, Environmental Systems Research Institute).
Remote sensing—satellites. Sentinel-2 (10-m pixel) and Landsat 8 (30-m pixel) satellite images were processed and downloaded using the Google Earth Engine (Gorelick et al. 2017). Level-1 precision- and terrain-corrected Landsat 8 images were atmospherically corrected to surface reflectance prior to retrieval (Vermote et al. 2016). Because of a lack of available surface reflectance imagery for the study area in 2018, top of atmosphere images from Sentinel-2 were used to maintain a consistent source. Images from satellite overpasses corresponded to acquisition dates closest to PAR measurements and relatively cloudless dates to compare results of each source with phenological stages. NDVI (S2NDVI and LS8NDVI) was calculated for each image using each sensor’s respective near-infrared (NIR; LS8 = 851 to 879 nm, S2 = 785 to 899 nm) and red (LS8 = 636 to 679 nm, S2 = 650 to 680 nm) bands. Data points corresponding to pixels with more than 20% of each pixel’s area outside of each vineyard, i.e., edge pixels, were not included in subsequent analyses.
PCA. PCA was conducted in R statistical software (R Core Team 2020, R Foundation for Statistical Computing, https://www.R-project.org/) using the FactoMineR package (Lê et al. 2008). Because berry weights, cluster counts, and pruning weights were not recorded in 2017, the 2017 data were not included in the further analyses, which used complete data sets for 2018 and 2019. All fruit compositional data, yield components, images, and environmental measurements were standardized by vineyard and season (mean = 0, standard deviation = 1) to eliminate site specific, vintage, and management effects (Carrillo et al. 2016). A modified t-test that accounted for spatial autocorrelation (Dutilleul 1993) was calculated using the SpatialPack package in R (Vallejos et al. 2020) to assess the strength of the relationships among the characteristics of the fruit, yield components, soil, topography, and imagery.
Results
Results from PCA when data were not standardized show that relationships between variables (descriptive statistics can be found in Table 3) were driven by site and seasonal effects (Figure 1A and 1B). However, combined standardized data in PCA from all four vineyards exhibited a typical global response as points from each vineyard and season overlapped and were distributed approximately evenly around the origin (Figure 1C and 1D). Figure 2 revealed similar relationships from the three sources of NDVI imagery (HRNDVI, S2NDVI, and LS8NDVI), and inverse relationships with HRCT. Imagery from three phenological stages were within ∼3% of one another in terms of variance explained by PCA, although no single combination of component one (PC1) and component two (PC2) explained more than 40% of the multivariate variability (Figure 2). NDVI from all three sources of imagery was related to pruning weights, and the relationship strengthened as the seasons progressed. HRCT separated with PARFZ at bloom and veraison (Figure 2). LS8NDVI showed the weakest relationships with fruit composition and vine performance variables, as evidenced by its limited distance from the center axis (Figure 2) and the weakest Pearson coefficients (Table 4). Visual comparisons of imagery captured at veraison (modified E-L stage 35) in Vineyard C highlight the similarities in the pattern of vigor captured by Landsat 8, Sentinel-2, and high-resolution NDVI, as well as the high-resolution canopy temperature (Figure 3). This similarity was apparent despite the removal of edge pixels in S2NDVI and LS8NDVI and the removal of interrow pixels in the HRCT and HRNDVI. Additionally, high values of HRCT corresponded to low values of HRNDVI, S2NDVI, and LS8NDVI (Figure 3), and this visual relationship was mirrored by PCA (Figure 2) and correlation analysis (Table 4).
Quercetin glycosides, polymeric tannins, anthocyanins, and β-damascenone were grouped together and were positively related to HRCT and inversely related to NDVI from all three sources (Figure 2 and Table 4). Malic acid, YAN, and C6 were typically related with low PARFZ and were closely related to pruning weights and yield, as well as NDVI in PC1 (Figure 2). Fruit yield per square meter and berry weight were relatively well spread along PC1 when compared with phenolic compounds, which were tightly grouped (Figure 2). While the results from PCA showed many of these variables to be closely related, results from the modified t-test showed that although many of these were significantly correlated (p < 0.05), the correlations for many were weak (Table 4). Correlations of fruit composition variables with yield components and pruning weights were generally strong, whereas correlations among most imagery, PARFZ, and topographic variables produced coefficients <0.2 (Table 4).
Figure 4 shows the relative positions among fruit composition, PARFZ, and yield components in each of the four vineyards. Vineyards had different mean values and magnitudes of variation in some variables (Table 3). Relationships among fruit compositional variables were similar in all vineyards, specifically along the first principal component (Figure 4). Groups of phenolic compounds (polymeric tannins, quercetin glycosides) segregated on the opposite side of several compounds related to shaded or unripe fruit (C6, malic acid). Larger yields, berry weights, and pruning weights also aligned more closely with characteristics of unripe fruit in all four vineyards (Figure 4). Similar to the relationship when all four vineyards were combined (Table 4 and Figure 2), PARFZ was positively related to phenolic compounds and inversely related to pruning weights, yield per meter, berry weight, malic acid, and YAN in all four vineyards (Figure 4). The strength of correlation coefficients was variable when assessed among individual vineyards, because some produced strong relationships among some variables and others produced very weak relationships (not shown).
The differences in canopy light environment (PARFZ) from fruit set (modified E-L stage 23) to veraison (modified E-L stage 35) shows that fruit in Vineyard A received less than 1% of ambient PAR compared with up to nearly 10% during this period in the other vineyards (Table 3). Vineyard D was only slightly more exposed than Vineyard A in the first two measured modified E-L stages (stage 23–bloom, and stage 27–fruit set), but because of canopy management practices, this percentage increased by veraison to levels similar to those observed in Vineyards B and C where fruit color was much higher. PARFZ was more variable in Vineyards B and C compared with Vineyards A and D, and generally matched the results of anthocyanins and several other variables (Table 3). Soil variables (sand, silt, clay, ECa) were sometimes statistically significant and correlated with compositional variables (p < 0.05; Table 4), but primarily varied along the less explanatory PC2 (Figure 5). Areas of higher elevation produced fruit with higher concentrations of desirable components, including polymeric tannins and quercetin glycosides (Table 4).
Discussion
Remote sensing. The most significant finding of this study was the relative similarity in the relationships between fruit composition and different remote sensing platforms across multiple seasons and multiple vineyards, although relationships among many variables were not strong. Images captured from satellites, Sentinel-2 and Landsat 8 at 10-m and 30-m ground resolutions, respectively, aligned closely with those from manned aircraft at 20-cm resolution when used as predictors of fruit composition; that is, there was little, if any, benefit offered by the high-resolution airborne imagery compared to the lower resolution satellite sources. Similar studies in other regions with more diverse soil and climate variability, or in applications where higher resolution imagery is required, may be necessary to broadly extrapolate these findings. Similarly, Sozzi et al. (2020) found that Sentinel-2 images were nearly as useful as UAV imagery captured at <10 cm in describing spatial variability of canopy vigor in vineyards with no grass in the interrow, although no comparisons with fruit chemistry were reported. However, in the PCA with images captured at bloom (Figure 2A) the HRNDVI was somewhat more closely related to pruning weights, unripe and green characteristics (YAN, malic acid, C6), and yield components. The results presented here expand this finding from the spatial variability noted in several fruit compositional parameters (Sams et al. 2022). There may be applications or time sensitive phenological stages where canopy separation from the interrow is imperative for determining vine water status, where an absolute value is necessary to calibrate with ground data, or in vineyards with native, nontilled vegetation occurring in irregular patches, but our results show that this separation may be unnecessary to describe spatial variability of vineyard canopies. Possible explanations for this scenario are that areas comprised of low vigor vines will also cause low vigor of the interrow cover crop, or that by the time of image acquisition at these phenological stages, the interrow cover crop is dead or dormant and provides a mostly bare background. It could be possible that a combination of these two explanations may actually enhance visual patterns of variability because low vigor zones contain more interrow space with soil or dead vegetation as a percentage of ground cover. The separation of nonvine signal from vine canopy is not trivial because image processing represents significant economic cost to imagery providers who then pass this on to customers. Furthermore, a major benefit of the satellite data examined here is the high-level calibration made by those providing the imagery (NASA and ESA), although given the weak correlations between imagery values and fruit composition, viticulturists may be more interested in spatial patterns found in imagery because absolute values from vegetation indices may vary over multiple images. When compared with maps of fruit composition in these same vineyards shown by Sams et al. (2022), this finding of pattern importance may be even more relevant. With this in mind, growers interested in assessing variability of vineyards may still be better suited to use lower-resolution, publicly available imagery at no cost, rather than purchase higher resolution from vendors because the relationships between imagery and fruit composition described here were not sufficiently strong to give absolute predictive value. For more precise prediction of fruit composition from remote sensing, it is likely that alternatives such as hyper-spectral imagery at high resolution may be needed, although neither the precision nor cost-effectiveness of such an approach is known at present.
Similar to our results, Carrillo et al. (2016) found that berry weight, cluster number, and NDVI were similarly aligned along the first principal component in a similar use of PCA. Ballesteros et al. (2020) used NDVI (with other indices) from an unmanned aerial sensor and machine learning techniques to model grapevine yield in Vitis vinifera cv. Bobal, but found contrasting results to those presented here. The authors stated that the soil background was the primary cause for a negative relationship between yield and NDVI when interrow and shadow pixels were included in the analysis, but when those pixels were removed, the relationship was positive immediately prior to harvest. These authors made a similar recommendation to ours, in that seasonal calibration is likely necessary for the most accurate results.
Environmental conditions and fruit composition. Differences in growing degree days, incoming solar radiation, precipitation, and reference evapotranspiration were likely partly responsible for differences in fruit composition among the three years of study (Table 1). In 2018, dormant season precipitation (leaf fall to budbreak) was much lower than in the other dormant seasons. This difference, coupled with the slower accumulation of growing degree days in 2018 and higher incoming radiation (Table 1), may explain the large increase in anthocyanin concentration in all four vineyards from 2017 to 2018 (Van Leeuwen et al. 2009; Table 3). However, quercetin glycosides and polymeric tannins were lowest in 2018 in all four vineyards, marking a separation in development between anthocyanins and these phenolic compounds during these seasons, despite the consistent grouping apparent in Figures 2, 4, and 5. Despite the relatively cooler season, a lack of precipitation in the early season (Table 1), and only small differences in irrigation volume (Table 2), yield was also highest in 2018 in all four vineyards (Table 3).
While a case can be made that lower yield equates to an increase in some positive aspects of fruit composition, the reasons for this relationship are somewhat misunderstood (Smart and Robinson 1991). In many cases, low-yielding vines often have small canopies and leaf areas but higher PARFZ, allowing more light into the canopy interior than a vine with identical yield, larger canopies, and lower PARFZ. Previous studies have linked light interception with fruit composition in California (e.g., Bergquist et al. 2001, Kliewer and Dokoozlian 2005), and their results complement those presented in this study. Low PARFZ was likely at least partly responsible for the relatively low values of anthocyanins, polymeric tannins, and anthocyanins in Vineyard A. Extremely high (or low) vigor may affect measurements of light interception through excessive sensor occlusion (or through direct sensor exposure in the case of low vigor). The values of IBMP and C6 are likely elevated in this vineyard due to excessive shading in the fruit zone. Given the low percentage of ambient PAR reaching the fruit zone, large clusters, and relatively high crop load on a single bilateral cordon (Tables 2 and 3), cluster occlusion may also be responsible for these results in Vineyard A (Dunn and Martin 2004). Leaf area can vary greatly between different canopy architectures, and many cultural manipulations are difficult to detect from remote sensing. Trellis design, pruning, leaf removal, irrigation, and crop load management are all parts of this complex system and contribute greatly to the decision-making process as well as to the cost of vineyard management (Jackson and Lombard 1993 and references within).
The relatively weak results from analyses of soil and topographic variables may be related to the way in which they were sampled but may also stem from the adjustment of management practices such as canopy management and irrigation, based on the highly variable soils, which can directly or indirectly influence canopy light interception (Table 4). It may also be possible that the magnitude of variability in soil and topography were insufficient to cause large differences in these vineyards and may differ in other regions or conditions. Shallow soil profiles on slopes and hillsides can produce differences in water holding capacity and lead to smaller vines compared with those on foothills or flat surfaces (Van Leeuwen et al. 2004, Jasse et al. 2021). It also may be responsible for the relationships between elevation, yield, and fruit composition (Figure 5), although soil horizon depths were not measured in this study. Additionally, inherent spatial variability of soil can cause drastic differences in yield and fruit quality, even over short distances, or small changes in topography (Bramley and Hamilton 2004, Bramley 2005); results presented in this study also suggest that elevation and soil texture have a significant effect on many aspects of fruit composition (Table 4). Accordingly, a more detailed investigation using measurements made down the soil profile and using a survey-grade GPS for elevation may lead to more precise understanding of soil and terrain effects.
The combination of high-density fruit composition samples and related vineyard measurements from several large commercial vineyards presented in this study can act as a guide to the industry as to which indirect or destructive measurements are truly useful in describing and predicting fruit composition. Targeted PAR measurements, pruning weights, and/or yield estimates collected from spatially distinct “quality” zones like those shown by Sams et al. (2022), compared with a few fruit composition samples taken with these zones in mind, should provide growers and wineries with a guide to the potential quality variability in a vineyard.
Conclusion
This study illustrates that remote sensing data from low resolution satellites paired with spatially targeted seasonal calibrations of yield components, pruning weights, and/or fruit composition can describe relationships between fruit composition and vegetative vigor at relatively low cost with sufficient precision to support delineation of within-vineyard zones of different vineyard performance, although additional studies may be necessary in different regions or in different conditions. If calibrations are conducted among different growing conditions and specific cultural and management practices (e.g., trellising, canopy management, water status), it should eventually be possible to model these systems with high precision and accuracy with minimal ground validation. As satellite imagery becomes more widely available and spatial and temporal resolution increases, growers will have even more options available to tailor vineyard management to specific production targets. Data products developed from these sensors, such as estimates of crop evapotranspiration or leaf area index, will also become more useful for more applications such as yield prediction, irrigation management, and environmental damage assessment. The current study may be useful to growers for reducing the costs of acquiring information about how variability affects the magnitude of fruit quality differences in vineyards. Until handheld, proximal sensors capable of detecting fruit composition are readily available at costs suitable for commercial growers and wineries, remote sensing products provide the best platform for capturing this information.
Footnotes
Acknowledgments/Author disclosures: This work was funded by E. & J. Gallo Winery, CSIRO, and The University of Adelaide. The authors would like to thank the Winegrowing Research staff at E. & J. Gallo Winery for their help and support in the collection, processing, and analysis of the collected grape samples. The authors declare no conflicts of interest in the publication of this work.
By downloading and/or receiving this article, you agree to the Disclaimer of Warranties and Liability. The full statement of the Disclaimers is available at http://www.ajevonline.org/content/proprietaryrights-notice-ajev-online. If you do not agree to the Disclaimers, do not download and/or accept this article.
- Received September 2021.
- Revision received November 2021.
- Revision received November 2021.
- Revision received November 2021.
- Accepted December 2021.
This is an open access article distributed under the CC BY license (https://creativecommons.org/licenses/by/4.0/).