Precision Agriculture

, Volume 12, Issue 1, pp 103–117

Within-season temporal variation in correlations between vineyard canopy and winegrape composition and yield

Authors

    • National Wine and Grape Industry CentreCharles Sturt University
    • School of Environmental SciencesCharles Sturt University
  • David W. Lamb
    • Precision Agriculture Research Group, School of Science and TechnologyUniversity of New England
  • Bruno P. Holzapfel
    • National Wine and Grape Industry CentreCharles Sturt University
  • John P. Louis
    • National Wine and Grape Industry CentreCharles Sturt University
Article

DOI: 10.1007/s11119-010-9159-4

Cite this article as:
Hall, A., Lamb, D.W., Holzapfel, B.P. et al. Precision Agric (2011) 12: 103. doi:10.1007/s11119-010-9159-4

Abstract

Remote optical imaging can rapidly acquire information describing spatial variability in vineyard block performance. Canopy characteristics were derived from very high spatial resolution (0.25 m) optical imagery of a Cabernet Sauvignon vineyard acquired at various canopy growth stages. Within-season changes to correlation coefficients between vineyard canopy and ultimate composition and yield of harvested fruit were then investigated. Canopy area and density were observed to have significant relationships with yield and fruit quality indicators including berry size, anthocyanins and total phenolic content, but less significant relationships with total soluble solids. The strength and type of correlation varied with canopy growth stage. For anthocyanins and total phenolic content, correlations varied from non-significant before flowering to negative after flowering. For berry weight and yield, correlations varied from negative before flowering to positive after flowering. For total soluble solids, there were some significant relationships but no clear temporal pattern. The results confirm that remote sensing is a useful tool to determine spatial variability in fruit composition and yield. However, both the timing of image acquisition and the way in which canopy is quantified are important determinants of the direction and strength of correlations with fruit composition and yield.

Keywords

Precision viticultureRemote sensingFruit qualityYieldCanopyPhenology

Abbreviations

CA

Grapevine canopy area

CD

Grapevine canopy density

NDVI

Normalised difference vegetation index

PAB

Photosynthetically active biomass

r

Coefficient of correlation

r2

Coefficient of determination

TSS

Total soluble solids

Introduction

Remote sensing is a precision agriculture tool used to inform management of spatial variability in vineyards. Spatial variability in environmental conditions may lead to tenfold differences in grape yield within single vineyard management blocks (Bramley and Hamilton 2004). Canopy related factors can affect yield at different stages of reproductive development: at inflorescence initiation (May 1965; Sanchez and Dokoozlian 2005; Watt et al. 2008), during flowering (May 2004), during berry growth and maturation (Ollat et al. 2002), and close to harvest (Smithyman et al. 1998). Winegrape quality has been shown to exhibit high levels of spatial variability (Bramley 2005), which can also be related to canopy conditions (Smart and Robinson 1991).

Important indicators of quality are the levels of both total soluble solids (TSS) and phenolic compounds in the winegrapes. Assessment of TSS is a convenient way of estimating winegrape sugar levels and is the most common guide to ripeness (Hamilton and Coombe 1988). Carbohydrate production in the grapevine is dependent on the total photosynthesis taking place, which is in turn dependent on the vegetative canopy, light intensity and temperature (Howell 2001). In addition to TSS, phenolic compounds are key compositional elements of red winegrapes, and subsequently in extracted juice, and play a role in the quality and value of finished wines (Harbertson and Spayd 2006). Both exposure to sunlight and temperature are influential on the development of phenolic compounds in winegrapes (Bergqvist et al. 2001; Spayd et al. 2002). Negative correlations between fruit phenolics content and foliage density have been reported using ground based imaging systems (Mabrouk and Sinoquet 1998).

Clearly, vine canopy affects quality and yield of winegrapes and is likely to have different levels of effect at different phenological stages. With the known variability in yield and fruit quality within vineyard blocks, remotely sensed imagery of vineyards is potentially valuable for the mapping of management zones and for forecasting yield and quality. Lamb et al. (2004) used remotely sensed imagery with a resolution similar to the vineyard row spacing (3 m) and found significant correlations between the normalized difference vegetation index (NDVI, a common indicator of photosynthetically active biomass (PAB) used in remote sensing) and measured grape phenolics. Lamb et al. (2004) indicated that the strongest correlations between NDVI and phenolics occurred at veraison and became weaker thereafter. However, the conclusion of Lamb et al. (2004) that imagery collected at veraison will best correlate with grape yield and quality is uncertain, because the reported decrease in the strength of relationship between grape phenolics and NDVI collected after veraison did not occur within any one season. Similar correlations between NDVI and yield reported by Proffitt et al. (2006) also had no obvious trend within single seasons. Imagery has since been collected at veraison in subsequent studies involving remotely sensed imagery of vineyards (Proffitt et al. 2006; Acevedo-Opazo et al. 2008), possibly based on these reports.

Since the optimal time to acquire imagery is currently unclear, a specific goal of the study presented here was to determine the levels of correlation that the canopy has with fruit yield and quality as the pattern of relative canopy vigour changes over time within a season. In contrast to previous studies, information on grapevine canopy is derived from very high spatial resolution (0.25 m) optical remotely sensed imagery at nine different levels of canopy development over 2 years. The within-season changing pattern of canopy development is presented to explain the changes in canopy vigour correlations with fruit yield and quality.

Materials and methods

The study vineyard comprised a hedge-pruned block (ca. 1 ha) of Cabernet Sauvignon on own roots subject to drip irrigation and spatially uniform management. Vines were trellised on a single wire with a row spacing of 3.6 m and an inter-vine spacing of 1.8 m. The block lay on a gentle (<10°) incline, sloping towards the east with rows planted perpendicular to the slope. Fifty-two individual locations were selected for sampling within the block using a stratified sampling scheme based on canopy foliage differences observed in multi-spectral airborne imagery obtained prior to commencement of the study. In order to avoid the difficulties in identifying the extent of each actual vine during fieldwork, a sample ‘vine’ was defined as the canopy structure between two adjacent vine trunks along the cordon, roughly encompassing two halves of adjacent vines.

The sample vines were harvested and the fruit was analysed at the conclusion of two successive growing seasons. Grape yield was measured on location (kilograms of grapes per sample vine) and a sub-sample of approximately 100 berries from each sample vine was immediately packed in a cool box for trans-shipment to the laboratory for analysis. Berry weight was determined by weighing 50 berries randomly selected from each sample. Total soluble solids (TSS) were measured with a bench refractometer (PR-101 ATAGO, Tokyo, Japan) from settled juice extracted from the grape samples. Anthocyanins (an important quality indicating sub-group of phenolics in red wines, often simply referred to as ‘colour’) and total phenolics were assessed by spectrophotometer (UV-2101PC, Shimadzu, Kyoto, Japan). Milligrams of anthocyanins per gram of berry weight and absorbance units per gram of berry weight (a measure of total phenolics) were measured following the procedure described by Iland et al. (2000). In addition, petiole nitrogen levels and season pruning weights were assessed for each vine during Year 1. Twenty petioles were collected at flowering from opposite berry clusters from each sample vine. The petioles were dried and then ground to a fine powder. Nitrogen content was analysed by the combustion technique (Howarth 1977) using a nitrogen analyser (NA 1500 series 2 Total Combustion Gas Chromatograph, Carlo Erba, Milan, Italy). All new wood was manually pruned from each sample vine at the end of the season (i.e. post leaf-fall) and weighed in situ (pruning weights). A mid-season (just after veraison) mechanical trim of the vine canopy was conducted in Year 2.

High spatial resolution (0.25 m) radiometrically-corrected multi-spectral airborne images of the target vineyard site were acquired on nine occasions spanning 2 years. The red and near infra red bands (at 650 nm and 770 nm, respectively) were converted to normalized difference vegetation index (Rouse et al. 1973) images. In Year 1, three imaging over-flights were timed to coincide with flowering, veraison and maturity (just prior to harvest). In Year 2, a further six imaging overflights were conducted: between budbreak and flowering, at flowering, between flowering and veraison, at veraison, between veraison and maturity, and at maturity (Table 1). Any vegetation growing between and under the vines was removed with herbicide spray in the weeks prior to the imaging overflights. This ensured that most photosynthetically active biomass detected in the vineyard block would be from vines alone.
Table 1

Days after budbreak of imaging overflights and designation of phenological stages

Days post-budbreak

Designation

Year 1

 60

Flowering

 107

Veraison

 156

Maturity

Year 2

 31

Post-budbreak

 48

Flowering

 67

Post-flowering

 124

Veraison

 140

Post-veraison

 165

Maturity

The image processing steps are illustrated in Fig. 1. A threshold NDVI value was calculated for each NDVI image to classify vine pixels and inter-row space pixels. For each image of the vineyard block, a histogram of the NDVI values was plotted. The histogram revealed a bimodal distribution for each set of NDVI values. The first peak of lower NDVI values represented the mode of the sparsely vegetated inter-row space. The second peak of higher NDVI values represented the grapevine canopy pixels. The threshold NDVI value for each image was calculated as the average of the NDVI minima of the trough between the peaks and the NDVI maxima of the second peak. Images were processed to remove pixels below the threshold NDVI value. The Vinecrawler algorithm (Hall et al. 2003) was then applied to collect spatially-referenced, single pixel-wide cross-sections along each vine row. In addition to the image co-ordinates of the centres of those slices, two quantitative canopy descriptors were calculated for each NDVI slice: canopy area (CA), a count of the number of pixels in the slice; and canopy density (CD), the mean NDVI value of the pixels in the slice.
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig1_HTML.gif
Fig. 1

Image processing steps illustrated using imagery from veraison, Year 1: a grey scale NDVI image, b threshold masked NDVI image (pixels with NDVI less than 0.6 removed), c locations of 1-pixel-length descriptors of canopy (VineCrawler output), d interpolated map of 1-pixel-length descriptors of canopy area, e categorized canopy area of individual grapevines, f locations of sample vines where fruit yield and quality information was collected

The geographic co-ordinates of sixteen ground control points (GCPs) were acquired with a 4000SSE Geodetic Surveyor differential GPS (Trimble Navigation Limited, Sunnyvale, California, USA), which had a maximum error of less than 0.05 m. The GCPs comprised the ends of vine rows selected to give the widest coverage of the vineyard block. In each image, the GCP locations were identified and their image co-ordinates recorded. A spatial warping function from the software package IDL (ITT Visual Information Solutions, Boulder, Colorado, USA), polywarp, was used to acquire parameters (ah) for first-order polynomial transformation equations for each image:
$$ E = a + bx + cy + dxy $$
(1)
$$ N = e + fx + gy + hxy $$
(2)
where E and N are map co-ordinate eastings and northings, and x and y are the horizontal and vertical image co-ordinates. The images were not resampled, maintaining their original spectral integrity. Instead, Eqs. 1 and 2 were used to calculate the equivalent geographic co-ordinates of the centre points of each vine row slice. The regular spacing of individual vines along rows was exploited to determine the location of each sample vine in terms of row number and distance from the northern row end. The constituent slices within a range of distances from the end of the vine row coincident with the sample vine were identified. The means of the canopy descriptors and the map co-ordinates of the identified slices were calculated to provide whole sample vine descriptors (Fig. 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig2_HTML.gif
Fig. 2

Area taken up by the canopy of a single grapevine is within the rectangle bounded by the heavy dashed line. Squares represent pixels that are either above or below the threshold NDVI

The high-spatial resolution nature of the data used in this study allowed for a detailed and precise appraisal of the relationship between canopy and grape quality and yield at the scale of individual grapevines. However, it should be noted that low spatial resolution imagery has been shown to relate to on-ground measures of grapevine characteristics (e.g. Johnson et al. 2003). In terms of broad differentiation of vineyard blocks into management zones, low spatial resolution imagery may provide similar information on spatial variability in winegrape quality and yield.

Results and discussion

Canopy correlations with petiole nitrogen and pruning weights

Correlation coefficients (r) for relationships of pruning weights and petiole nitrogen content with the remotely sensed canopy descriptors were calculated. The coefficients were plotted against the number of days after budbreak imagery was acquired (Fig. 3). The correlation coefficients plotted outside the hachured area are significant at p = 0.05. Availability of nitrogen is a key factor driving canopy development, and the weight of new wood produced in a season (pruning weight) directly reflects canopy density and area development. Petiole nitrogen and pruning weight correlations with canopy area and density can therefore be expected to be significant. The interesting aspects of Fig. 3 are in terms of the point in time the remotely sensed canopy descriptors begin to have significant correlations with the indicators of canopy development, and in terms of the relative correlation magnitudes. Canopy area became significantly correlated with pruning weight and petiole nitrogen earlier in each season than canopy density and had larger magnitude correlation coefficients once significant (see Fig. 3). More confident forecasts about future canopy conditions could therefore be made earlier in the season using a measure of canopy area rather than canopy density.
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig3_HTML.gif
Fig. 3

Correlation coefficients (r) for relationships of pruning weights and petiole nitrogen content with the remotely sensed canopy descriptors plotted against number of days after budbreak imagery was acquired: a canopy density vs. pruning weight; b canopy density vs. petiole nitrogen; c canopy area vs. pruning weight; d canopy area vs. petiole nitrogen. + symbol indicates data from Year 1; × symbol indicates data from Year 2, grey × symbol indicates data from Year 2 acquired after summer trimming of the grapevine canopy. Correlation coefficients plotted outside the hachured area are significant at p = 0.05

Canopy area and canopy density are both closely related measures of foliage with significant correlation coefficients returned for each image except at flowering and post-flowering in Year 2 (Table 2). The correlation coefficients between them and petiole nitrogen and pruning weights are therefore similar for both canopy measures. They had insignificant or negative correlations with petiole nitrogen and pruning weights early in the season and positive correlations later (Fig. 3). The point in time at which the correlations between the canopy descriptors and pruning weight or petiole nitrogen became positive differed for the different combinations of variables. In all cases, the correlations were consistently significant after flowering. Up until flowering, carbohydrate reserves, in addition to those manufactured by the current season’s foliage, may be utilised in grapevine canopy development (Yang and Hori 1979). Utilisation of carbohydrate reserves may have contributed to the non-significant correlations observed in this period.
Table 2

Correlation coefficients between canopy area (CA) and canopy density (CD) at each growth stage

 

Year 1

Year 2

Flowering

Veraison

Maturity

Post-budbreak

Flowering

Post-flowering

Veraison

Post-veraison

Maturity

CA, CD

0.60**

0.56**

0.57**

0.44**

0.15

0.22

0.50**

0.71**

0.59**

* Significant at p = 0.05, ** significant at p = 0.01

Within-season changes in spatial patterns of relative canopy vigour

Figure 4 indicates that a similar spatio-temporal pattern of relative canopy development occurred within each of the two seasons. Foliage in the western part of the block grew faster relative to the rest of the block early in the season (Fig. 4a, d). By mid-season (Fig. 4b, e), relative canopy density was distributed more evenly. By canopy maturity, the relatively densest foliage was in the eastern part of the block and in other areas where early season growth was relatively slower. An apparent negative relationship between early and late season (a) to (c) and (d) to (f) may be qualitatively interpreted from Fig. 4. Although, correlation coefficients between canopy area at different canopy development stages (Table 3) did not produce significantly negative values, in both years, the correlation level became weaker the further apart in time the canopy development stages were.
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig4_HTML.gif
Fig. 4

Canopy density (mean NDVI) interpolated to cover the study site for three occasions in each of the two years. Earlier to later stages of growth run from left to right, with the number of days post-budbreak being: in Year 1 a 60, b 107, c 156; and in Year 2 d 48, e 67, f 124. Although the canopy developmental stage varied in terms of growth rate (possibly due to climatic variability between the two years), the pattern of development was similar in both years. Canopy developed most rapidly in the west early in the season, but as the season progressed the densest canopies overall developed in the east

Table 3

Correlation coefficients between canopy area (CA) data derived from imagery acquired at each stage in (a) Year 1 and (b) Year 2 (imagery acquired after mid-season trimming in Year 2 excluded)

 

Flowering

Veraison

Maturity

(a) Year 1

 Flowering

1

  

 Veraison

0.52**

1

 

 Maturity

0.11

0.50**

1

 

Post-budbreak

Flowering

Post-flowering

Veraison

(b) Year 2

 Post-budbreak

1

   

 Flowering

0.64**

1

  

 Post-flowering

0.09

0.55**

1

 

 Veraison

−0.17

0.16

0.52**

1

* Significant at p = 0.05, ** significant at p = 0.01

Correlation coefficients between the canopy descriptors and grape yield and composition variables across both years are presented in Table 4 (canopy density) and Table 5 (canopy area). In addition to the within-year correlations, Tables 4 and 5 present the correlation coefficients between canopy characteristics in Year 1 and fruit composition and yield in Year 2, and between canopy characteristics in Year 2 and fruit composition and yield in Year 1. The pattern of correlation was similar across the years, suggesting temporal change in relative spatial patterns in canopy and fruit were similar each year. Temporal stability in vineyard yield and fruit composition has been described by Bramley and Hamilton (2004), Bramley (2005), and over a 7-year period by Tisseyre et al. (2008). However, these studies described inter-annual temporal stability; within-season changes to spatial variability were not reported. Table 3 and Fig. 4 show that, in this study, changes to relative canopy growth rates within the season were considerable.
Table 4

Correlation coefficients between mean canopy density (CD) and fruit composition and yield

CD

Year 1

Year 2

Flowering

Veraison

Maturity

Post-budbreak

Flowering

Post-flowering

Veraison

Post-veraison

Maturity

Year 1

 Total soluble solids

−0.12

−0.11

0.22

−0.41**

−0.08

−0.02

−0.19

−0.18

−0.28*

 Berry weight

−0.55**

0.37**

0.57**

−0.35**

−0.27*

−0.35**

0.41**

0.09

0.08

 Anthocyanin

0.23*

−0.58**

−0.44**

0.04

0.20

0.28*

−0.58**

−0.25*

−0.30*

 Total phenolics

0.35**

−0.58**

−0.57**

0.09

0.35**

0.28*

−0.65**

−0.29*

−0.32**

 Yield

−0.24*

0.33**

0.26*

0.12

−0.28*

−0.13

0.55**

0.26*

0.34**

Year 2

 Total soluble solids

0.41**

−0.45**

−0.31*

0.13

0.40**

0.34**

−0.43**

−0.01

−0.05

 Berry weight

−0.30*

0.37**

0.33**

−0.19

−0.46**

−0.30*

0.47**

0.00

0.06

 Anthocyanin

0.23*

−0.35**

−0.36**

−0.17

0.20

0.06

−0.36**

−0.41**

−0.41**

 Total phenolics

0.32*

−0.34**

−0.28*

−0.10

0.29*

0.08

−0.35**

−0.25*

−0.27*

 Yield

−0.67**

0.38**

0.60**

−0.36**

−0.52**

−0.37**

0.46**

−0.06

−0.03

* Significant at p = 0.05, ** significant at p = 0.01

Table 5

Correlation coefficients between canopy area (CA) and fruit composition and yield

CA

Year 1

Year 2

Flowering

Veraison

Maturity

Post-budbreak

Flowering

Post-flowering

Veraison

Post-veraison

Maturity

Year 1

 Total soluble solids

−0.28*

−0.31*

0.07

−0.13

−0.27*

−0.27*

0.08

−0.05

−0.19

 Berry weight

−0.31*

0.23*

0.42**

−0.59**

−0.35**

0.19

0.50**

0.21

0.05

 Anthocyanin

−0.30*

−0.52**

−0.62**

0.38**

−0.13

−0.47**

−0.51**

−0.41**

−0.48**

 Total phenolics

−0.15

−0.54**

−0.66**

0.48**

−0.02

−0.48**

−0.61**

−0.45**

−0.39**

 Yield

0.32**

0.37**

0.39**

−0.29*

0.10

0.37**

0.38**

0.23

0.28*

Year 2

 Total soluble solids

0.09

−0.34**

−0.39**

0.33**

0.05

−0.17

−0.52**

−0.20

−0.32**

 Berry weight

0.03

0.30*

0.46**

−0.27*

−0.07

0.19

0.49**

0.28*

0.24*

 Anthocyanin

−0.09

−0.47**

−0.54**

0.13

−0.31*

−0.52**

−0.61**

−0.45**

−0.42**

 Total phenolics

−0.04

−0.43**

−0.49**

0.18

−0.26*

−0.43**

−0.54**

−0.37**

−0.42**

 Yield

−0.29*

0.25*

0.46**

−0.69**

−0.33**

0.20

0.58**

0.20

0.18

* significant at p = 0.05, ** significant at p = 0.01

Canopy correlations with anthocyanins and total phenolics

Figure 5 presents within-year changes to canopy area and density correlations with fruit composition and yield. Both canopy descriptors were significantly negatively correlated with anthocyanins and total phenolics after veraison (~80 days after budbreak). Negative correlations occurring at veraison were observed in a different study by Lamb et al. (2004) but were then shown to decrease in magnitude. However, here, the correlation strength continued to be consistently high and even increase after veraison (Fig. 5). In addition, canopy area provided significant negative correlations as soon as 40 days after budbreak, well before veraison.
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig5a_HTML.gif
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig5b_HTML.gif
Fig. 5

Correlation coefficients (r) for relationships of fruit composition and yield variables with the remotely sensed canopy descriptors plotted against number of days after budbreak imagery was acquired. Fruit composition and yield variables are plotted horizontally: a, f total soluble solids; b, g anthocyanins; c, h total phenolics; d, i berry weight; e, j yield. Remotely sensed canopy descriptors are plotted vertically: a, b, c, d, e CD; f, g, h, i, j CA. + symbol indicates data from Year 1; × symbol indicates data from Year 2, grey × symbol indicates data from Year 2 acquired after summer trimming of the grapevine canopy. Correlation coefficients plotted outside the hatched area are significant at p = 0.05

Mid-season trimming of the vines in Year 2 reduced the magnitude of the correlations for canopy data derived from the last two imaging missions (grey ‘cross’ symbols, Fig. 5). Measures of canopy area (Table 5), rather than canopy density (Table 4), correlated earlier and more strongly with anthocyanin and total phenolic content of harvested fruit. Other than this treatment-response, the level of correlation changed little after veraison (Fig. 5). Since canopy area acquired only 40 days after budbreak correlated significantly with anthocyanins and total phenolics, the timing of image acquisition may not be as critical as previously suggested for irrigated vineyards.

Canopy correlations with total soluble solids

In contrast to the anthocyanin and total phenolic fruit composition variables, TSS exhibited few significant relationships with the canopy descriptors (Tables 4, 5). Those that were significant were not consistent across the 2 years (Fig. 5a, f). Canopy area correlated significantly with TSS on five occasions (positively at the earliest point in the season, and then negatively). However, the results were confounded by no correlation between canopy and TSS being observed close to harvest in Year 1. The lack of consistency in correlations and the generally low magnitude of those significant correlations (Tables 4, 5) suggest that TSS concentration in harvested fruit has a more complex relationship with canopy levels than phenolics concentration. For example, carbohydrate reserves may be utilised for fruit maturation (to raise TSS levels) in those grapevines that have relatively low photosynthetic capacity to fruit weight ratios (see Candolfi-Vasconcelos et al. 1994). On the strength of these results, confident forecasts of spatial variability in TSS across a vineyard could not be made based on canopy alone.

Canopy correlations with berry weight and yield

Both total yield and berry weight had similar patterns of correlation levels with canopy (Tables 4, 5). This may be expected because berry weight is a key factor determining total yield. Here, total yield and berry weight correlated with each other: r = 0.50 and r = 0.34, in years 1 and 2, respectively. Since the correlation magnitudes between either total yield or berry weight and canopy were similar, no inference could be made as to whether either canopy affected berry weight and hence yield, or else canopy simply affected total yield.

Total yield per vine and mean berry weight exhibited significant negative correlations with canopy area and density early in the season and then significant positive correlations late in the season (Fig. 5d, e, i, j). Vines that were relatively larger and denser at the beginning of the season were therefore generally lower yielding, whereas those that were relatively larger and denser later in the season were higher yielding. The switch from negative to positive relationships is particularly strong for canopy density, with only the earliest set of canopy data not being significantly negatively correlated with berry weight (Fig. 5d). The mid-season canopy trimming had a large effect on the magnitude of the correlations, which were not statistically significant (p > 0.05) after the summer pruning event.

For imagery acquired early in the season up until flowering, those vines with less dense canopies were likely to have experienced conditions favourable for the development of inflorescence primordia. A negative correlation between canopy at flowering in Year 1 and yield in Year 2 followed. This is a similar result to past research using on-ground measures of canopy (May 2004; Sanchez and Dokoozlian 2005). If the canopy growth characteristics are considered to be stable from year to year (as reported by Bramley and Hamilton 2004; Tisseyre et al. 2008), then consistently smaller early-season canopies may indicate perennially similar spatial patterns in primordial development and subsequent yield. Any inter-annual negative relationship between canopy and yield holds only if the relative pattern of canopy growth is similar year-to-year.

Change in correlation direction

The change in direction of the correlations, either from positive to negative or from negative to positive, between canopy and fruit composition and yield is evident in Fig. 5. Figure 6 illustrates patterns of possible canopy development to explain the temporal change in direction of the correlations between canopy and fruit composition and yield observed in this study. The point in time at which the correlation switched direction was earlier for canopy area (~40–60 days post budbreak) than for canopy density (>80 days post budbreak) (Fig. 5). Relative differences in canopy density continued to change due to further foliage development increasing the leaf area density, but with shoot growth slowing and therefore not increasing canopy area.
https://static-content.springer.com/image/art%3A10.1007%2Fs11119-010-9159-4/MediaObjects/11119_2010_9159_Fig6_HTML.gif
Fig. 6

Schematic describing levels of difference between canopy density within a vineyard block against phenological time. The greatest range in canopy density occurs early in the season (with earlier developing vines relatively more vegetated) and then late in the season (with later developing vines relatively more vegetated)

Conclusion

Canopy descriptors associated with canopy area and density were derived from decimetre resolution remotely sensed NDVI imagery of a Cabernet Sauvignon vineyard on nine occasions spanning 2 years. Correlations between the canopy descriptors of 52 grapevines and fruit composition and yield data collected from the same vines prior to harvest were examined. Both canopy area and density were consistently significantly correlated to fruit anthocyanin and phenolic content, berry size and yield. These results did not support the previously reported reduction in the correlation between NDVI with anthocyanin/total phenolics after veraison. Rather, there were generally steady increases in the magnitude of the correlations as the season progressed each year, which continued after veraison. Fruit total soluble solids content correlated the least and most inconsistently with canopy descriptors. The image-derived descriptor of canopy area produced stronger correlations with the viticultural variables than canopy density, and these correlations became consistently significant earlier in the season. The timing of imaging affected the strength and direction of the correlations. The correlations sometimes reversed in polarity, particularly those between canopy density and yield, which were negatively correlated early in the season and then positively correlated late in the season.

Acknowledgments

This work was supported by the Wine Growing Futures Program, a joint initiative of the Grape and Wine Research and Development Corporation and the National Wine & Grape Industry Centre. The authors appreciate ongoing support provided by Charles Sturt University’s Spatial Data Analysis Network.

Copyright information

© Springer Science+Business Media, LLC 2010