Using remote sensing to predict grape phenolics and colour at harvest in a Cabernet Sauvignon vineyard : Timing observations against vine phenology and optimising image resolution

46
Predicting grape phenolics and colour at harvestAustralian Journal of Grape and Wine Research 10, 46–54, 2004
Using remote sensing to predict grape phenolics and colour at
harvest in a Cabernet Sauvignon vineyard: Timing observations
against vine phenology and optimising image resolution
D.W. LAMB1,2,4,5, M.M. WEEDON1 and R.G.V. BRAMLEY3,4
1 National Wine and Grape Industry Centre, Charles Sturt University, Locked Bag 588,Wagga Wagga NSW 2678
2 presently: School of Biological, Biomedical and Molecular Sciences, University of New England,Armidale NSW 2351
3 CSIRO Land and Water, PMB 2, Glen Osmond SA 5064 Australia
4 Cooperative Research Centre for Viticulture, PO Box 154, Glen Osmond SA 5064 Australia
5 Corresponding author:A/Prof David Lamb, facsimile: +61 2 6773 3268, email: dlamb@pobox.une.edu.au
Abstract
Optical remote sensing can provide a synoptic view of grapevine photosynthetically-active biomass over
entire vineyards both rapidly and cost-effectively. Such output offers viticulturists and winemakers a
management tool of enormous potential with red grape varieties, especially if canopy architecture
(de?ned in this way) can be linked to production of phenolics and colour in ripe grapes. Accordingly, this
paper describes such associations for a Cabernet Sauvignon vineyard in Australia’s cool-climate
Coonawarra region. A link is established between physical descriptors of grapevine canopies (derived
from remotely-sensed images), and subsequent measurements of grape phenolics and colour. High-
resolution images were acquired on three occasions during each of two consecutive growing seasons and
post-processed to a range of on-ground resolutions. The strength of correlation between those images and
berry properties (both total phenolics, and colour levels at harvest), varied according to spatial resolution
and vine phenology at the time of imaging. An image resolution corresponding approximately to row
spacing resulted in the strongest correlations between berry constituents and image-based data on all
occasions. Referenced to grapevine phenology, correlations were initially weak (insigni?cant) at bud-
burst, reached maximum strength at veraison, then diminished somewhat as grapes ripened. Prospects for
applying such remotely-sensed imagery (at an appropriate resolution and timing), to predict berry
phenolics and colour at harvest, are discussed.
Abbreviations
dGPS differential global positioning system; EM electromagnetic; GPS global positioning system; GIS geographical
information system; NDVI normalised difference vegetation index; PAB photosynthetically-active biomass
Keywords: multispectral imaging, Precision Viticulture, remote sensing, grapevine canopy, grapevine vigour, grapevine
phenology, photosynthetically-active biomass (PAB), grape colour, grape phenolics, terroir
Introduction
polyphenolic compounds including ?avan-3-ol monomers
Flavonoids in grapes, and their subsequent presence in
such as catechin and proanthocyanidin polymers, and
extracted juice, are important determinants of wine qual-
contribute signi?cantly to mouthfeel and colour stability
ity. While present in both red and white grapes,
with anthocyanins in red wines. This large and diverse
?avonoids exist at much higher concentrations in red
group of compounds is commonly referred to collectively
grape varieties, and especially in skins, where they con-
as tannins, and is synonymous with the generic term
stitute a signi?cant group of compounds that in?uence
‘phenolics’ or ‘total phenolics’ as used here. Tannins
wine style, aroma and ?avour. Three key subgroups of
reside primarily within the skin cells of grape berries,
?avonoids include anthocyanins, ?avonols, and tannins.
although some tannins occur in seed, stems and berry
While tannins occur in both white and red varieties,
?esh (Coombe 1990, Kennedy et al. 2000, Souquet et al.
anthocyanins are speci?c to red grapes and are primarily
2000).
responsible for red wine colour. By contrast, ?avonols are
Overall, the synthesis and accumulation of ?avonoids
colourless. They probably confer UV protection in both
in grapes is greatly in?uenced by bunch exposure to sun-
white and red grapes, and act as co-pigments for antho-
light (Winkler et al. 1974, Pirie and Mullins 1980, Archer
cyanins in red grapes. The third and quantitatively largest
and Strauss 1989, Jackson 2000). Consequently, the
sub-group of flavonoids (or tannins) include mostly
location of grapes within a given canopy, as well as

---

Lamb, Weedon & Bramley
Predicting grape phenolics and colour at harvest
47
canopy density and size, will in?uence the development
potential exists for generating synoptic information con-
of ?avour and colour in red winegrapes by virtue of light-
cerning spatial variations in quality attributes such as
driven variation in skin ?avonoids (Smart and Robinson
colour and total phenolics. To date, remote sensing of
1991, Downey et al. 2004).
spatial variations in vine PAB (or its equivalent) has been
In approaching this issue of canopy size from a per-
linked to spatial variations in grape yield (Baldy et al.
spective of remote sensing, Hall et al. (2002) encapsulat-
1996, Lamb et al. 2001), and fruit maturity (Johnson et
ed the combination of vine-leaf biomass and leaf chloro-
al. 2001, Bramley et al. 2003). With regard to spatial vari-
phyll content in their de?nition of photosynthetically-
ation in maturity, remotely sensed imagery has been used
active biomass (PAB ), a term that integrates canopy size,
to divide blocks into zones to facilitate segmented har-
density and vigour. A priori, high-PAB vines would be
vests. Fruit of given ‘quality’ from different zones can
likely to shade bunches from direct sunlight to a greater
then be batched for separate winemaking, but delineation
extent than would low-PAB grapevines (Mabrouk and
of those zones remains somewhat arbitrary.
Sinoquet 1998), and as a corollary, average bunch tem-
Given those developments, a quest for additional pre-
peratures would be lower and light climate within larger
dictive applications using imagery of different spatial res-
canopies would be more heavily attenuated. Accumu-
olutions seemed warranted. Accordingly, this paper now
lation of ?avonoids would be altered accordingly, so that
presents results from a detailed investigation of links
some quantitative relationship might exist between PAB
between remotely-sensed descriptors of vine canopy
at critical phenological stages and ?avonoid content at
vigour at different phenological development stages, and
harvest. The present project was undertaken to explore
subsequent measurements of total phenolics and colour
that possible link.
in harvested fruit. The studies were undertaken on
Grapevine PAB is in?uenced by numerous physical,
Cabernet Sauvignon grapes in a cool climate vineyard
biological and chemical factors, including spatial varia-
(Coonawarra region) in Australia. The potential of
tions in topography, physical and chemical characteristics
remote sensing at key phenological stages as a viticultur-
of soils and the incidence of pests and diseases. The spa-
al management tool for predicting grape quality at har-
tial variation in such factors within a vineyard will cause
vest is then discussed.
a spatial variation in canopy development and con-
sequently grape colour and phenolic content at harvest.
Recent work (Bramley 2001) has demonstrated that con-
siderable spatial variation exists with quality indicators
such as colour and total phenolics across individual vine-
yards. From a management point of view, such variation
could result in an overall reduction in wine quality.
Given the likelihood of increased differentiation in pricing
between grapes based on quality attributes (Winemakers
Federation 1996), vineyard management decisions would
then need to accommodate such spatial variability in
quality in order to produce categories of grapes with
higher unit value.
Figure 1. Black and white aerial photograph of the target vineyard,
Pricing decisions will, however, rely on accurate and
acquired February 2000, showing the three separate blocks (dotted
reliable data that describe spatial variability for indicators
line) of Cabernet Sauvignon imaged and subsequently sampled.
of grape quality; especially colour and total phenolics.
Traditional methods of generating such data are general-
ly time consuming and expensive (Lamb and Bramley
2001). As a general rule, measurement of basic fruit
Materials and methods
quality and yield variables takes between ten and thirty
Vineyard site and measurement of grape quality
minutes per sampling location. The move toward rapid in
The investigation was completed in a 7.3 ha block of
situ measurement of grape colour and total phenolics is
Cabernet Sauvignon located in the Coonawarra region of
therefore welcome, but such measurements are still
South Australia (37º 17’ S, 140º 48’ E) during the
constrained by lack of appropriate sensor technology. A
1999–2000 and 2000–2001 growing seasons. The vines,
system has now been devised to sense grape phenolic
planted on their own roots in 1974, were trained onto a
composition of harvested fruit using optical-?bre-based
single wire with 3 m row spacing and rows orientated
visible-NIR spectroscopy (Celotti et al. 2001), but pre-
approximately north-south (Figure 1). Dates of major
harvest assessment within a vineyard still remains a
phenological development stages and a summary of the
challenge.
management applied to this block over the two consecu-
As an alternative to proximal sensing, multispectral
tive growing seasons studied in this work are given in
airborne remote sensing is capable of providing a non-
Table 1. Dates corresponding to key phenological stages
intrusive, instantaneous and synoptic view of grapevine
or vine management in Table 1 differed by no more than
PAB in vineyards and vineyard blocks (Hall et al. 2002,
twelve days between the two seasons.
Lamb and Bramley 2001). Given the links between
During harvest (both seasons), grape samples were
canopy PAB and grape quality in red winegrapes, the
hand-picked from 190 vines distributed evenly through-

---

48
Predicting grape phenolics and colour at harvestAustralian Journal of Grape and Wine Research 10, 46–54, 2004
Table 1. Dates of major phenological development stages
Table 2. Dates of image over?ights and time relative to
and key management applied to the vineyard site during
budburst calculated using principal dates from Table 1.
the 1999–2000 and 2000–2001 growing seasons.
Season
Over?ight date
Days post-budburst
Stage/Management
1999–2000 season
2000–2001 season
1999–2000
2 November
36
Pruning
By end of August
By end of August
30 November
64
16 March
170
60% woolly bud
14 September
19 September
60% green tip
20 September
28 September
2000–2001
14 December
70
Budburst
27 September
5 October
20 February
138
(60% leaf emergence)
5 April
182
5–6 leaves unfolded (80%)
11 October
23 October
75% ?owering
3 December
28 November
75% fruit set
15 December
12 December
cessing software ER Mapper (Earth Resource Mapping,
Berries pea size
22 December
18 December
San Diego, California, USA). Geometric distortion in the
Irrigation (25 mm
mid-January
mid-January
imagery results from imperfections in the lens and
via travelling irrigator)
appears as barrel or pin-cushion distortion in the imagery.
Veraison
9 February
7 February
Radiometric distortions are inherent brightness variations
Harvest
8 April
6 April
in imagery due to the combination of camera lens and
iris. Radiometric and geometric distortions were both cor-
out the vineyard block. The exact location of each vine
rected following the procedure outlined in Spackman et
was measured using a differential global positioning
al. (2000). Image recti?cation (assigning map coordinates
system which had a measurement accuracy of approxi-
to individual image pixels) was then completed using 16
mately ± 50 cm. The grapes were immediately packed
ground control points (readily identi?able features, of
in a cool-box for transhipment to the laboratory and
known dGPS coordinates, in images) which comprised
following subsequent sub-sampling, were stored frozen
the ends of selected vine rows. Due to inherent mis-
prior to further analysis. Grape colour (mg antho-
alignment of the four cameras relative to each other (a
cyanins/gram berry weight) and total phenolics (280 nm
maximum error of approximately 1 pixel), band-to-band
absorbance units per gram berry weight) were sub-
registration was also completed using ER Mapper.
sequently measured following the procedures of Iland et
al. (2000).
Linking on-ground measurements to airborne imagery and
extracting image pixel values for each vine
Airborne multispectral imaging
Ground coordinates of individual sampled vines often
Multispectral airborne images of the target vineyard site
failed to coincide exactly with the location of the vines in
were acquired using Charles Sturt University’s airborne
the geo-recti?ed imagery. This was a consequence of spa-
video system (ABVS) (Louis et al. 1995). The ABVS
tial errors in locating individual vines with the dGPS, and
comprises 4 CCD video cameras in a 2 × 2 array and an
residual errors in removing geometric distortions from
IBM-compatible computer containing a 4-channel
the imagery. In order to check the coincidence of sampled
framegrabber board. Each camera captures a static image
vines with vines in the imagery, the ground coordinates
in a separate waveband governed by an interchangeable
of every sampled vine were overlayed onto the geo-
?lter. For this research, the standard vegetation wave-
recti?ed imagery and checked. The location of those
bands of blue (450 nm), green (550 nm), red (650 nm)
points that were not coincident with the centre of the
and near-infrared (770 nm) were used. Each ?lter had a
vine rows (always on the eastern side of the row centres)
band-pass of 25 nm and each camera was ?tted with a 12
were consistent with the fact that the actual on-ground
mm focal-length lens. Images of the vineyard site were
measurements of vine coordinates were completed by a
acquired at an altitude of 1,000 m (±10 m) above ground
person, equipped with a dGPS mounted in a back-pack,
level, resulting in a spatial resolution of 60 cm and an
standing and leaning into the vine from the eastern side
image coverage of 14 ha. Three image over?ights were
of each vine row. An automated process was therefore
conducted during both seasons. However, the date of
developed where, for points that were not coincident
each over?ight was different in each season to provide a
with vine centres, the ground-coordinates were progres-
complete dataset of imagery at six different time intervals
sively shifted west in the imagery until the vine centre
following budburst. The dates of image over?ights and
was reached (de?ned by a local maximum value in near-
calculated time after budburst are given in Table 2.
infrared re?ectance).
Image over?ights were conducted at a carefully select-
Pixel values, for each of the blue, green, red and near-
ed time on each occasion to ensure that illumination con-
infrared wavebands, were then extracted from single pixels
ditions were similar, thereby minimising calibration errors
representing the centre of each sampled vine, as well as
associated with using images in digital number (rather
from groups of pixels representing larger areas centred on
than re?ectance) format. Each image was corrected for
the centre of each sampled vine. The pixel combinations
camera-induced geometric and radiometric distortions,
and their corresponding on-ground dimensions are
and recti?ed to map coordinates using the image pro-
summarised in Table 3.

---

Lamb, Weedon & Bramley
Predicting grape phenolics and colour at harvest
49
Table 3. Pixel combinations extracted from imagery for
Table 4. Pearson correlation coef?cients for NDVI against
each waveband (blue, green, red and near-infrared) at
colour for extracted pixel groups, each centred on the
each sampled vine location in the vineyard and corre-
190 sample vines in the vineyard.
sponding area on the ground.
Season
Days post- NDVI-single
NDVI 3 × 3
NDVI 5 × 5
Number of pixels
Centred on
Area on ground
budburst
pixel
extracted per vine
1999–2000
36
0.01
0.04
0.06
1
Vine centre
60 cm × 60 cm
64
–0.07
–0.09
–0.11
3 × 3
Vine centre
1.8 m × 1.8 m
170
–0.48
–0.53
–0.57
5 × 5
Vine centre
3.0 m × 3.0 m
2000–2001
70
–0.13
–0.22
–0.23
138
–0.27
–0.40
–0.43
182
–0.21
–0.25
–0.25
Calculation of NDVI from extracted pixel values
Spectral vegetation indices reduce the multiple-waveband
data at each image pixel to a single numerical value
Table 5. Pearson correlation coef?cients for NDVI against
(index), and many have been developed to highlight
total phenolics for extracted pixel groups, each centred on
the 190 sample vines in the vineyard.
changes in vegetation condition. A detailed discussion of
the use of vegetation indices in remotely sensed imagery
Season
Days post- NDVI-single
NDVI 3 × 3
NDVI 5 × 5
of vineyards is given elsewhere (Hall et al. 2002). In this
budburst
pixel
present work, values for Normalised Difference Vegetation
Index (NDVI), the most widely used indicator of plant
1999–2000
36
–0.01
0.04
0.12
vigour or relative biomass (Lamb 2000, Hall et al. 2002),
64
–0.07
–0.10
–0.08
were calculated from the extracted single-pixel or
170
–0.49
–0.53
–0.56
multiple-pixel values using the relationship below (Rouse
2000–2001
70
–0.30
–0.39
–0.38
et al. 1973):
138
–0.36
–0.53
–0.59
182
–0.29
–0.26
–0.24
(near infrared) – (red)
NDVI = (near infrared) + (red)
where ‘near infrared’ and ‘red’ were the pixel values in
This is consistent with the results of Lamb et al. (2001)
the near-infrared and red waveband respectively. Taking
who used imagery of 20 cm resolution and increased the
the 3 × 3, and 5 × 5 pixel groups, average NDVI was cal-
sample or pixel size of the imagery (by averaging groups
culated. Even though the imaging campaign was
of pixels) to produce larger pixels which were a mix of
designed to reduce differences in illumination conditions
vine and inter-row space. Such mixed pixels are an inte-
(namely, selection of appropriate clear sky-conditions,
gration of both vine size and vine density information.
time of day for similar solar elevations and azimuth),
Where the 20 cm imagery was sub-sampled to 3 m reso-
some small variations in the range of NDVI values were
lution, similar to the vine-row spacing, the complex vine
observed in imagery. Ultimately, and to ensure that such
structure information evident in the high resolution
variations are canopy-related rather than environmental,
imagery (size and shape, density) was reduced to a simple
imagery would be collected in re?ectance mode rather
relative-mix of vine and non-vine signature, and were
than digital numbers. Other researchers, in an attempt to
found to provide a clearer indication of the location of
address the issue of variable environmental conditions
high and low PAB vines. In this current work, sampling
when acquiring multi-temporal imagery, either normalise
groups of 5 × 5 pixels is equivalent to a ground footprint
NDVI (or similar index) values, or alternatively bin them
of 3 × 3 m, which coincides with the 3 m spacing
into high, medium and low values for each image (for
between adjacent rows.
example Dobrowski et al. 2003). However in this present
The Pearson correlation coef?cients for the NDVI
analysis, rather than collecting all data into a single
calculated from the 5 × 5 pixel groupings (NDVI 5 × 5)
dataset, relationships between NDVI and quality indices
are plotted against days post-budburst in Figure 2. Super-
were assessed on an individual image basis, thereby
imposed on the scatter-plot of Figure 2 are the key
allowing use of ‘raw’ rather than binned NDVI values.
phenological stages listed in Table 1.
In Figure 2, correlations between extracted NDVI val-
Results and discussion
ues and either total phenolics or colour grow stronger
Results from simple linear correlation analyses between
(more negative) after budburst. Maximum strength is
extracted NDVI and colour, as well as between NDVI
reached around veraison, and then decreases from
and total phenolics, are summarised in Tables 4 and 5,
approximately 130 days post budburst. There is clearly an
respectively.
association between the spatial distribution of canopy
Data presented in Tables 4 and 5 suggest that the
PAB and the distribution of total phenolics and colour in
strength of correlations between extracted NDVI and
the grapes from an early stage in fruit development. This
measured total phenolics and colour increases (becomes
is consistent with the hypothesis that microclimate, taken
more negative) when larger areas of pixels are averaged
here to imply exposure to sunlight and average bunch
around the centre of ?eld-sampled vines in the imagery.
temperature will influence both total phenolics and

---

50
Predicting grape phenolics and colour at harvestAustralian Journal of Grape and Wine Research 10, 46–54, 2004
fruit-set
veraison
Table 6. Summary of simple linear regression analyses
0.3
(n = 190) between extracted NDVI (5 × 5) and total
0.2
phenolics and colour measurements for images acquired
icient
0.1
99-00
on 16 March 1999 and 20 February 2000 (closest to
0
veraison for each season).
99-00
-0.1
99-00
-0.2
Season/Parameter
Regression equation
R2
P stat
-0.3
00-01
(95%)
-0.4
1999–2000
-0.5
00-01
00-01
Total phenolics, TP
TP = –3.02 (NDVI) + 3.02
0.32
4.5× 10-17
-0.6
Pearson correlation coeff
irrigation
Colour, C
C = –4.14 (NDVI) + 3.11
0.33
1.0 × 10-17
-0.7
0
50
100
150
200
2000–2001
Days (post-budburst)
Total phenolics, TP
TP = –3.17 (NDVI) + 2.25
0.35
6.9× 10-16
Colour, C
C = –2.51 (NDVI) + 2.07
0.19
1.9 × 10-8
Figure 2. Trend line for Pearson correlation coef?cients for NDVI (5
× 5) against colour (G) and total phenolics (L) as a function of
days post-budburst; 99–00 and 00–01 indicate the growing season
phenolics and colour (approximately 150 days after bud-
of the corresponding data.
burst), can be attributed to two sets of factors. Firstly, the
absolute content of anthocyanins in fruit may have
colour. However, light exposure and average bunch tem-
decreased as previously observed with numerous red
perature, while both likely to be negatively correlated to
winegrape varieties (Somers 1976, Keller and Hrazdina
PAB (Mabrouk and Sinoquet 1998), can have an opposite
1998, Hasselgrove et al. 2000). Indeed, Hasselgrove et al.
effect on the synthesis of phenolics and colour. Too high a
observed anthocyanins per berry to actually decrease in
bunch temperature may inhibit anthocyanin accumula-
‘exposed berries’ of Shiraz 46 days post veraison, close to
tion, even in the situation of adequate light interception
the time at which the PAB-phenolics/colour association
(for example, Hasselgrove et al. 2000). The fact that PAB
was observed to weaken in this work. The second set of
and total phenolics/colour exhibit signi?cant negative
factors could relate to the onset of water stress, and asso-
correlations in this work implies that the accumulation of
ciated leaf fall at different times in different regions of the
such compounds in the grapes is predominantly light-
vineyard. This vineyard site is typical of vineyards in the
limited. The trend in Figure 2 is also consistent with the
Coonawarra Region, where the terra rossa soils are char-
fact that phenol synthesis in red winegrapes begins early
acterised by a layer of light-red clay topsoil and under-
during berry development (in Figure 2, fruit set was
lying limestone parent material. Figure 5 is a map of soil
approximately 80 days post-budburst). Similarly, antho-
depth in this particular vineyard (Bramley et al. 2000)
cyanin production becomes pronounced only after verai-
which has been shown to control plant-available water
son, (in Figure 2, approximately 130 days post-budburst)
(PAW; Bramley and Lanyon 2002). This map indicates the
(Hasselgrove et al. 2000, Jackson 2000). Synthesis of phe-
depth from surface to underlying parent limestone.
nolics tends to decline and may cease following veraison
Comparison of this map with Figures 3 and 4 reveals that
(Hasselgrove et al. 2000, Jackson 2000). For this particu-
the lower levels of measured total phenolics and colour,
lar ?eld site, the spatial pro?le of total phenolics and
and hence higher-PAB vines (higher NDVI), are coinci-
anthocyanins would have been ‘imprinted in the grapes’
dent with areas of deeper soil. Generally, for a given sea-
between 50 and 150 days post-budburst and would be
son, deeper soils support greater root growth, which in
closely linked to the spatial pro?le of canopy PAB (NDVI)
turn supports increased vine vigour (Smart and Robinson
as detected via remote sensing. Furthermore, berry size
1991) and retention of leaves.
generally increases with fruit or leaf shading (Morrison
In both the 1999–2000 and 2000–2001 seasons, hot-
1988, Hasselgrove et al. 2000), a phenomenon most
dry conditions prevailed following veraison (Anon 2002)
likely to be associated with high levels of canopy PAB
and the water-holding ability of the soil, as determined by
(NDVI). Larger berries would also strengthen the negative
the depth of topsoil (Bramley and Lanyon 2002), would
correlation between NDVI, total phenolics and colour due
have played a signi?cant role in determining water avail-
to increased dilution of skin constituents by expressed
ability. At the time of irrigation, the initial contrast
juice during sample preparation.
between regions of apparently high-PAB versus low PAB
Results of simple linear regression analyses between
match canopy development in varying depths of soil. This
extracted NDVI (5 × 5) and both total phenolics and
is shown schematically in Figure 6(a). Following irriga-
colour, for images acquired on 16 March 1999 and 20
tion, it is likely that the vines in the regions of shallow
February 2000 (images closest to veraison for each of the
topsoil would have succumbed to water stress sooner
two sampling seasons) are summarised in Table 6. Based
than those in the deeper topsoil regions. The loss of PAB
on these regression equations, maps showing total
in the vines located in the shallow topsoil, a combination
phenolics and colour (created from the NDVI imagery)
of loss of leaf chlorophyll and leaf-fall, would have acted
are given in Figures 3 and 4.
to increase the difference in PAB between the regions
Following veraison (indicated in Figure 2), a change
already associated with differences in total phenolics or
in the strength of correlation between NDVI, total
colour, and hence increase the strength of the correlation

---

Lamb, Weedon & Bramley
Predicting grape phenolics and colour at harvest
51
2.7
a
b
2.2
/g)
AU
1.7
1999–2000
1.2
0.7
Total phenolics (
0.2
NDVI predicted total phenolics
0.30
0.35
0.40
0.45
0.50
0.55
0.60
concentration (AU/g)
NDVI
1.6
1.4
/g) 1.2
AU
1.0
0.8
2000–2001
0.6
Total phenolics ( 0.4
0.2
0.30
0.35
0.40
0.45
0.50
0.55
NDVI
Figure 3. (a) Scatter-plots of NDVI (5 × 5) values versus total phenolics (absorbance units per gram berry weight ) and (b) total phenolics
maps created from NDVI (5 × 5) imagery, acquired close to veraison for 1999–2000 and 2000–2001 seasons. The tilted red lines through the
total phenolics maps correspond to the wide corridors of exposed, scenesced covercrop/bare soil between adjacent blocks. Differences in
NDVI range between seasons correspond to differences in range of canopy PAB.
3.0
a
b
2.5
2.0
1.5
1.0
Colour (mg/g)
0.5
1999–2000
0.0
0.30
0.35
0.40
0.45
0.50
0.55
0.60
NDVI
NDVI predicted total colour
concentration (mg anthocyanin/g)
2.0
1.5
1.0
2000–2001
Colour (mg/g)
0.5
0.0
0.30
0.35
0.40
0.45
0.50
0.55
NDVI
Figure 4. (a) Scatter-plots of NDVI (5 × 5) values versus colour (mg anthocyanins/gram berry weight) and (b) Colour maps created from NDVI
(5 × 5) imagery, acquired close to veraison for 1999–2000 and 2000–2001 seasons. The tilted red lines through the total colour maps
correspond to the wide tracts of exposed, scenesced covercrop/bare soil between adjacent blocks. Differences in NDVI range between
seasons correspond to differences in range of canopy PAB.

---

52
Predicting grape phenolics and colour at harvestAustralian Journal of Grape and Wine Research 10, 46–54, 2004
between the high-PAB and low-PAB (NDVI) vines and
these quality attributes (Figure 6(b)). Delayed leaf-fall
from the high-PAB vines on deeper-soil, most likely
commencing after production of phenolics in the grapes
had ceased, would then reduce the difference in PAB
between the regions (Figure 6(c)) and therefore reduce
the strength of the spatial correlation between the NDVI
and the spatial phenolics and colour pro?les as measured
on the ground. In a management scenario where soil
water is non-limiting, we anticipate that the strength of
the correlations between NDVI and total phenolics and
colour would have remained high until harvest time.
Validation of that scenario is the subject of ongoing
work.
Remote sensing in vineyards is currently limited to
Figure 5. Soil depth (depth to underlying limestone parent material)
detecting and mapping variations in canopy attributes
map for the Coonawarra vineyard site. Data of Bramley et al. (2000).
such as vine PAB (Hall et al. 2002 and references there-
in). By implication, spatial variation in other quantities
such as yield or berry properties are then inferred from
that canopy index. A question thus arises as to optimal
timing for remote sensing which is intended for yield
the use of remote sensing as a means of discriminating
and/or quality prediction. In this present work, a cor-
between regions of differing fruit quality.
relation between remotely-sensed vine PAB and total
Regarding the cost-effectiveness of remote sensing,
phenolics and colour in a Cabernet Sauvignon vineyard
higher-resolution imagery could be used to extract infor-
was demonstrated. Moreover, the strength of that
mation from vine-only pixels, thereby providing infor-
correlation was shown to vary in a systematic way with
mation such as leaf-only spectra. However, extraction of
grapevine phenology, with predictive value strongest
such information may introduce an unnecessary and
around veraison.
additional level of complexity to data analysis (and at an
From an operational point of view, the appropriate
added cost to the user). Furthermore, a descriptor of the
imaging window necessary to gain an accurate insight
overall vine size may be omitted from the output and this
into spatial variations of grape quality, as described for
may signi?cantly reduce the value of generated data for
example by colour or total phenolics, would have been
the user. On balance, and mindful of operational needs in
centred on veraison. Vine PAB maps generated earlier
remote sensing, coarser-resolution data can generally be
than this would have yielded little in the way of man-
acquired more easily from airborne sensors. Moreover,
agement data related to grape quality. Maps generated
increased coverage can then enable a reduction in the
later than this would have meant that either degradation
unit cost of acquiring such data (e.g. Lamb 2000). Indeed,
of accumulated phenolics/colour compounds, or the
and as a future alternative to air-borne sensors, satellite-
impact of water stress on the vine canopy, would have
borne instruments such as those on board Quickbird or
changed the spatial vine-PAB pro?le from that at verai-
Ikonos satellite systems are capable of providing multi-
son when the canopy was likely to have exerted its great-
spectral data of similar resolution (Lamb et al. 2001b).
est in?uence on phenolic content of grapes at harvest.
Signi?cantly, the per-hectare cost of acquiring such data
The strongest correlation between PAB and total
is likely to decrease as satellite-borne instruments become
phenolics/colour required imagery at a resolution com-
more widely used for resource monitoring, so that
parable to the vine-row spacing. That coincidence implies
satellite imagery may well become a viable alternative to
an underlying importance for vine size as a component of
present analyses that are derived from on air-borne
PAB. This same coincidence has practical implications for
instrument platforms.
a
b
c
Figure 6. The impact of soil depth on changing the contrast between high- and low-PAB vines. (a) Deeper topsoil promotes more vigorous vine
development and higher PAB during the growing season. There is a large difference in PAB observed between vines growing in the two
regions. (b) Following irrigation, the deeper topsoil retains moisture and vines in shallow topsoil lose PAB as a result of water-de?cit. The
difference between PAB of vines in the regions of deep and shallow topsoil is increased. (c) The vines in the deep topsoil now respond to the
delayed water de?cit. The difference between PAB of vines in regions of deeper and shallow topsoil is now reduced.

---

Lamb, Weedon & Bramley
Predicting grape phenolics and colour at harvest
53
Acknowledgements
Downey, M.O., Harvey, J.S. and Robinson, S.P. (2004) The effect of
This paper has been prepared under the auspices of the
bunch shading on berry development and ?avonoid accumulation
CRC for Viticulture, an organisation supported by
in Shiraz grapes. Australian Journal of Grape and Wine Research
10, 55–73
Australia’s grapegrowers and winemakers through their
Hall, A., Lamb, D.W., Holzapfel, B. and Louis, J. (2002). Optical
investment body the Grape and Wine Research and
remote sensing applications for viticulture – a review. Australian
Development Corporation, with matching funds from the
Journal of Grape and Wine Research 8, 36–47.
Federal government, and Commonwealth Cooperative
Hasselgrove, L., Botting, D., van Heeswijck, R., Hoj, P.B., Dry, P.R.,
Research Centres Program. We acknowledge the contri-
Ford, C. and Iland, P.G. (2000) Canopy microclimate and berry
composition; The effect of bunch exposure on the phenolic compo-
bution of Dr Tony Prof?tt (formerly of Southcorp Wines
sition of Vitis Vinifera L cv. Shiraz grape berries. Australian Journal of
and now with Albert Haak & Associates), helpful discus-
Grape and Wine Research 6, 141–149.
sions with Dr Richard Smart (Smart Viticultural Services)
Iland, P.G., Gawel, R., Coombe, B.G. and Henschke, P.M. (1993)
during interpretation of aspects of these data, the assis-
Viticultural parameters for sustaining wine style. In: Proceedings of
tance of Susie Williams (CSIRO Land & Water) during
the Eighth Australian Wine Industry Technology Conference
(Australian Wine Industry Technical Conference Inc.: Adelaide, SA)
acquisition of grape quality data, and Diane Stewart
pp. 167–169.
(Southcorp Wines) for provision of the vineyard man-
Iland, P., Ewart, A., Sitters, J., Markides, A. and Bruer, N. (2000)
agement and vine development data. The generous sup-
Techniques for chemical analysis and quality monitoring during
port of Southcorp Wines in providing access to the ?eld
winemaking. (Patrick Iland Promotions: Campbelltown, SA)
site and support staff is also gratefully acknowledged. The
Jackson, D.I. (1986) Factors affecting soluble solids, acid, pH and
color in grapes. American Journal of Enology and Viticulture 37,
authors are particularly indebted to Paul Kriedemann
179–183.
(Editor) for his constructive contributions in the prepara-
Jackson, R.S. (2000) Wine Science: Principles, Practice, Perception.
tion of this manuscript.
(Academic Press: San Diego, California).
Keller, M. and Hrazdina, G. (1998) Interaction of nitrogen availabil-
ity during bloom and light intensity during veraison. II. Effects on
anthocyanin and phenolic development during grape ripening.
References
American Journal of Enology and Viticulture 49, 341–349.
Anon. (2003) Australian Wine Online – Vintage Reports 2000 and
Kennedy, J.A., Troup, G.J., Pilbrow, J.R., Hutton, D.R., Hewitt, D.,
2001 http://www.winetitles.com.au/awol/vintage/index.asp (accessed
Hunter, C.R., Ristic, R. Iland, P.G. and Jones, G.P. (2000)
October 2002).
Development of seed polyphenols in berries from Vitis Vinifera L. cv.
Archer, E. and Strauss, H.C. (1989) Effect of shading on the perfor-
Shiraz. Australian Journal of Grape and Wine Research 6, 244–254.
mance of Vitis vinifera L. cv. Cabernet Sauvignon. South African
Lamb, D. W. (2000) The use of qualitative airborne multispectral
Journal of Enology and Viticulture 10, 74–77.
imaging for managing agricultural crops – A case study in south
Baldy, R., DeBenedictis, J., Johnson, L., Weber E., Baldy M., Osborn,
eastern Australia. Australian Journal of Experimental Agriculture
B. and Burleigh, J. (1996) Leaf colour and vine size are related to
40, 725–738.
yield in a phylloxera-infested vineyard. Vitis 35, 201–205.
Lamb, D.W. and Bramley, R.G.V. (2001) Managing and monitoring
Bramley, R.G.V. (2001) Progress in the development of Precision
spatial variability in vineyard productivity. Natural Resource
Viticulture – Variation in Yield, Quality and Soil Properties in
Management 4, 25–30.
Contrasting Australian Vineyards. In: Precision tools for improving
Lamb, D.W., Hall, A. and Louis, J. (2001) Airborne remote sensing of
land management. Occasional report No. 14. Fertilizer and Lime
vines for canopy variability and productivity. Australian
Research Centre, Massey University, Palmerston North, New
Grapegrower and Winemaker 449a, 89–92.
Zealand, pp. 25–43.
Lamb, D.W., Hall, A. and Louis, J.P. (2001b). Airborne/spaceborne
Bramley, R.G.V. and Lanyon, D.M. (2002) Evidence in support of the
remote sensing for the grape and wine industry. In: Proceedings of
view that vineyards are leaky – Indirect evidence and food for
the National Conference on Geospatial Information and
thought from precision viticulture research. In: Vineyard ‘leakiness’.
Agriculture, Incorporating Precision Agriculture in Australasia,
Proceedings of a workshop held at the Waite Campus, Adelaide,
Fifth Annual Symposium, Sydney (Causal Productions: Sydney,
January 24–25, 2002. Final Report on GWRDC Project No.
NSW) pp. 600–608.
GWR01/04 (CSIRO Land and Water / Grape and Wine Research
Louis, J., Lamb, D. W., McKenzie, G., Chapman, G., Edirisinghe, A.,
and Development Corporation, Adelaide SA,) pp. 31–38.
McCloud I. and Pratley, J. (1995). Operational use and calibration
Bramley, R.G.V., Prof?tt, A.P.B., Corner, R.J. and Evans, T.D. (2000)
of airborne video imagery for agricultural and environmental land
Variation in grape yield and soil depth in two contrasting Australian
management. In: Proceedings 15th Biennial American Workshop
vineyards. In: Soil 2000: New Horizons For A New Century.
on Colour Photography and Videography in Resource Assessment.
Australian and New Zealand Second Joint Soils Conference,
(ASPRS: Bethesda, Maryland) pp. 326–333.
Volume 2: Oral Papers (New Zealand Society of Soil Science,
Mabrouk, H. and Sinoquet, H. (1998) Indices of light microclimate
Lincoln University, Canterbury, New Zealand), pp. 29–30.
and canopy structure of grapevines determined by 3D digitising and
Bramley, R., Pearse, B. and Chamberlain, P. (2003) Being pro?table
image analysis, and their relationship to grape quality. Australian
precisely – A case study of precision viticulture from Margaret
Journal of Grape and Wine Research 4, 2–13.
River. Australian and New Zealand Grapegrower and Winemaker
Morrison, J.C. (1988) The effects of shading on the composition of
461a, 84–87.
Cabernet Sauvignon grape berries. In Proceedings of the Second
Celotti, E., De Prati, G.C. and Cantoni, S. (2001) Rapid evaluation of
International Symposium on Cool Climate Viticulture and
the phenolic potential of red grapes at winery delivery: application
Oenology (New Zealand Society of Viticulture and Oenology,
to mechanical harvesting. Australian Grapegrower and Winemaker
Auckland, New Zealand) pp. 144–146.
449a, 151–159.
Pirie, A.J.G. and Mullins, M.G. (1980) Concentration of phenolics in
Coombe, B.G. (1990) Research on berry composition and ripening.
the skin of grape berries during fruit ripening and development.
In: Proceedings of the Seventh Australian Wine Industry
American Journal of Enology and Viticulture 31, 34–36.
Technology Conference (Australian Wine Industry Technology
Price, J.C. and Bausch, W.C. (1995) Leaf area index estimation from
Conference Inc.: Adelaide, SA), pp. 150–152.
visible and near-infrared re?ectance data. Remote Sensing of
Dobrowski, S.J., Ustin, S.L. and Wolpert, J.A. (2003) Grapevine dor-
Environment 52, 55–65.
mant pruning weight prediction using remotely-sensed data.
Rojas-Lara, B.A. and Morrison, J.C. (1989) Differential effects of
Australian Journal of Grape and Wine Research 9, 177–182.
shading fruit or foliage on the development and composition of
grape berries. Vitis 28, 199–208.

---

54
Predicting grape phenolics and colour at harvestAustralian Journal of Grape and Wine Research 10, 46–54, 2004
Rouse, J. W. Jr., Haas, R. H., Schell, J. A., and Deering, D. W. (1973).
Wiegend, C.L., Richardson, A.J., Escobar, D.E., and Gerbermann,
Monitoring vegetation systems in the great plains with ERTS. In:
A.H. (1991) Vegetation indices in crop assessments. Remote Sensing
Proceedings of the Third ERTS Symposium, NASA SP-351 1, (U.S.
of Environment 35, 105–119.
Government Printing Of?ce: Washington DC.) pp. 309–317.
Winemaker’s Federation of Australia (1996) Strategy 2025 – The
Smart, R.E. (1997) In?uence of light on composition and quality of
Australian Wine Industry (Magill: South Australia).
grapes. Acta Horticulture 206, 37–47.
Winkler, A.J., Cook, J.A., Kliewer, W.M. and Lider, L.A. (1974)
Smart, R. and Robinson, M. (1991) Sunlight Into Wine. A Handbook
‘General Viticulture’ (University of California Press, Berkeley,
for Winegrape Canopy Management. (Winetitles: Adelaide, SA).
California).
Smart, R.E., Smith, S.M. and Winchester, R.V. (1988) Light quality
and quantity effects on fruit ripening for Cabernet Sauvignon.
American Journal of Enology and Viticulture 39, 250–258.
Somers, T.C. (1976) Pigment development during ripening of the
grape Vitis 14, 269–277.
Souquet, J-M., Labarbe, B., Le Guerneve, C., Cheynier, V. and
Moutounet, M. (2000) Phenolic Composition of Grape Stems.
Journal of Agriculture and Food Chemistry 48, 1076–1080.
Spackman S., McKenzie G., Lamb D.W. and Louis J. (2000)
Retrieving biophysical data from airborne multispectral imagery of
rice crops. In: Proceedings of Geoinformation for all, Nineteenth
Congress of the International Society for Photogrammetry and
Remote Sensing (ISPIRS), Amsterdam, The Netherlands.
International Archives of Photogrammetry and Remote Sensing B7,
Manuscript received: 18 July 2003
pp. 1447–1451.
Revised version received: 17 December 2003

---