Content uploaded by Samuel J L Gascoigne
Author content
All content in this area was uploaded by Samuel J L Gascoigne on Nov 17, 2023
Content may be subject to copyright.
iScience
Article
Seasonal weather impacts wine quality in Bordeaux
Andrew Wood,
Samuel J.L.
Gascoigne,
Gregory A.
Gambetta,
Elizabeth S.
Jeffers, Tim
Coulson
wood_and@hotmail.com
Highlights
Wine quality measured
through critics scores varies
across space and time
Weather at multiple points
during the year impacts
wine quality in Bordeaux
Climate shifts suggest wine
quality may improve with
future predicted climate
Wood et al., iScience 26,
107954
October 20, 2023 ª2023 The
Author(s).
https://doi.org/10.1016/
j.isci.2023.107954
ll
OPEN ACCESS
iScience
Article
Seasonal weather impacts wine quality in Bordeaux
Andrew Wood,
1,3,
*Samuel J.L. Gascoigne,
1
Gregory A. Gambetta,
2
Elizabeth S. Jeffers,
1
and Tim Coulson
1
SUMMARY
Critics judge quality based upon subjective characteristics of wine. These judgments are converted by
critics into quantitative scores, which allow for comparison of vintages. This paper uses high resolution
discrete and continuous time-based weather estimates at both a local and regional level to determine
the role of weather conditions on producing high quality Bordeaux vintages, as determined by critics
scores. By using discrete-time weather variables across local AOCs, this study reveals climate-quality
relationships across the whole year, including previously ignored season effects. By using continuous
time weather variables, we reinforce the evidence for these local effects by finding higher quality wine
is made in years with higher rainfall, warmer temperatures; and earlier, shorter seasons. We propose
management impacts of our results and suggest that as the climate continues to change, the quality of
Bordeaux wines may continue to improve.
INTRODUCTION
Climate change is globally impacting agricultural produce, both in terms of yield and quality.
1,2
Despite these expected effects, the link
between climate change and agricultural produce quality has not been widely explored. Wine (Vitis vinifera) presents the ideal system to study
this relationship as wine price is governed primarily by quality,
3
which is dependent on weather during the vine’s growing season.
4
Addition-
ally, wine quality in Bordeaux (France) has been measured by many independent experts over time, meaning that there exists a multi-critic
regional and local longitudinal dataset for quality.
5–11
With the availability of high-resolution weather data we can now use this information to
examine how weather influences quality on both a regional and local scale.
Local variation in the quality of wine was first acknowledged with the introduction of wine rating systems. The Bordeaux Grand Cru system
was created in 1855 to classify individual vineyards into one of five categories based on price and perceptions of quality. This Grand Cru
classification system has been expanded such that there now exists 14 defined categories of wines in Bordeaux, with other wines simply being
categorized as unclassified via this method.
11
A series of geographical protections were introduced in 1936, referred to as appellations d’ori-
gine contro
ˆle
´e, or AOCs. Acting on the local scale, they create individualities for wines, with each AOC having distinct viticultural character-
istics and vinicultural identities.
12
Such identities can link an AOC to perceived quality, with some becoming more famous than others.
Regional and local disparities can be explored by comparing scores for the whole of Bordeaux to individual wine scores linked to an
AOC. Consequently, each individual wine is wrapped in its own historic quality ratings which have the potential to shift perceptions of the
current and future wines. Such perceptions must be considered in any attempt to understand quality.
Some studies have directly examined quality using tasting scores.
13–15
In Bordeaux these tasting scores traditionally take the form of a
primeur score. These scores are bestowed by wine critics at tastings approximately 10 months after harvest and just after blending. While
these wines are not mature and often highly tannic,
11
this scoring system provides a direct standardized measurement of quality and allows
for an ascertainment of the quality of the wine before it fully ages. Other critics, mainly wine merchants, rate Bordeaux as a whole region,
giving an overall classification as to whether or not a year is good or not. Due to this two-scale rating system, there exists the potential to
compare regional tasting scores to local tasting scores.
Weather conditions have also been demonstrated to have an impact on the wine quality. Most famously, Ashenfelter’s (1995) Bordeaux
equation
6–8
suggested that the average price of wine in Bordeaux is a linear function of winter precipitation and summer temperature. Other
models have used monthly weather, demonstrating that finer resolution weather data and local chateau characteristics
6,16
can contribute to
explaining price variation in Bordeaux wines. These local effects have been examined using this same price-based approach by Lecocq and
Visser (2006) using local weather stations.
17
The models in Lecoq and Visser (2006) found similar results at both regional and local scales and
thus suggested that in most cases regional and local weather records are interchangeable.
Tasting scores have been correlated with single-year metrics of weather such as annual mean temperature and precipitation
7,9,18
in wine-
growing regions from Australia
14
to California.
13
Consensus from the Bordeaux equation, price modeling and current quality scores suggest
that higher temperature and lower precipitation leads to higher quality grapes.
5,7,11
Ashenfelter and Jones (2013) suggest that: critics scores
‘‘reflect qualitatively the same weather factors that have been documented to be determinants of wine quality.’’
6
Multiple studies have shown
1
Department of Biology, University of Oxford, 11a Mansfield Road, Oxford OX1 3SZ, UK
2
EGFV, Bordeaux Sciences Agro, INRAE, Universite
´de Bordeaux, ISVV, Villenave d’Ornon, France
3
Lead contact
*Correspondence: wood_and@hotmail.com
https://doi.org/10.1016/j.isci.2023.107954
iScience 26, 107954, October 20, 2023 ª2023 The Author(s).
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
1
ll
OPEN ACCESS
that a higher number of warm days during flowering and at the onset of berry ripening (spring and summer) and lower precipitation during
berry maturation (autumn) leads to higher quality.
5,19,20
But other conflicting studies have shown impacts outside of this time frame, with
weather affecting quality across the year. Notably, the Bordeaux equation suggests that primarily winter precipitation and summer temper-
ature is important. But Vittorio and Ginsburgh (1996) use the number of days of hail in April as well as temperature and precipitation during
June to September.
16
Jones and Storchman (2001) look at phenological stages and find that four different weather aspects (evapotranspira-
tion; total rainfall; and the number of days with temperatures more than 25C and 30C) all have an effect on the price of the final wine.
11
Baciocco et al. (2014) suggest that low rainfall and high heat accumulation over the year lead to higher ranked wines.
10
Bonada et al.
(2020) claim that rainfall during winter dormancy impacts quality.
21
Alongside such varying insights in the literature comes a finding of a reduc-
tion in quality with high temperatures.
18,20
Other evidence also links spring frosts to changes in quality.
22
Overall, these findings suggests the
potential for regional differences in climate change to potentially impact wine quality.
23
Thus, it is important to understand which features of
the weather are affecting wine quality and when, in order to determine the precise impacts of climate change on a viticulturally relevant spatial
resolution.
In this study, we explore the link between weather and critic quality scores, using weather and quality scores for Bordeaux wines, at both
regional and local levels. We use discrete time models with time steps such that the impact of temperature and precipitation on wine quality
scores across the year can be ascertained. We then use continuous time models which explore the weather across the whole year as single
functions for rainfall and temperature. In turn, we aim to give greater understanding as to when wine quality is most susceptible to changes in
temperature and precipitation, and how we can examine such changes.
RESULTS
Between 1950 and 2020 there was a general increase in quality scores for wine quality in the Bordeaux region. The maximal annual mean score
was 98.67 points (1961), and the minimum was 32.5 (recorded in 1965). A generalized linear model was fitted to determine the location annual
trends, as per model 1 in the model summary figure (Figure 1A). In this GLM, the overall critic score for Bordeaux is predicted by the year,
controlling for critic. Year fitted as a continuous variable in the model was found to be statistically significant (coef = 0.0195, c(1) =
4.4528, p < 0.05, R
2
= 0.27), meaning there is a general increase in critics’ quality score over time (Table S3).
Quality scores were also examined on a local scale. The maximal mean critics score was 99 points (recorded once in 2019) and the minimum
was 28 (recorded once in 2006). Critics showed high correlation between ratings (Figure S2). A binomial GLM was fitted, again as per model 1
AB
C
Figure 1. Methods panel plot
(A) Summary of the generalized linear models run to analyze the relationship between wine critics scores and the weather, controlling for location annual trends
(location and year interactions). In each row the number on the bottle refer to the model number, and tilde means ‘‘is a function of’’.
(B) Depiction of sine wave with depiction of parameters fitted. mm is the mean temperature, athe amplitude, sthe wavelength, and 4the phase shift.
(C) Mean critics scores (scaled from 0 to 100) over time for Bordeaux as a whole with colored lines showing the GLM fitted.
ll
OPEN ACCESS
2iScience 26, 107954, October 20, 2023
iScienc
e
Article
on the model summary figure (Figure 1A). It reported no significant coefficients for Year nor Year:AOC interactions. This suggests that, for the
time period, there is no local increase in quality over time (for test statistics see Table S4).
Discrete time model
Generalized linear models were also built to explore the relationship between grouped monthly weather variables and the mean overall
Bordeaux general score, controlling for the yearly trend of improvements (R
2
= 0.61; Table S5), as per model 2 in the model summary figure
(Figure 1A). As weather has been normalized, only the sign (positive or negative) and the relative size of the coefficients are important. The
largest significant (p < 0.05) coefficient in the model is the positive coefficient for summer temperatures (coef = 4.48, F (1,59) = 122.1716, p =
0.0004), followed by the negative coefficient for summer precipitation (coef = 7.14, F (1,59) = 114.0373, p < 0.05), and then the positive
coefficient for winter precipitation (coef = 3.95, F (1,59) = 164.4203, df = 3, p < 0.05), as shown in Figure 2A. Models were found to fit well
from visual inspection of residual plots (see Figure S3). According to this model, changes in temperature and precipitation at other times
of year would not change the overall Bordeaux scores, excluding stochastic extreme weather events.
A generalized linear model was also run on the local (AOC) scale, as per model 3 on the wine model summary figure (Figure 1A). Model 3
examines the relationship between grouped monthly weather variables and individual wine scores, controlling for year increases, AOC, and
Grand Cru status (n = 4521, R
2
= 0.35; for full details see Table S6). All weather terms were found to be significant, with coefficients shown
graphically in Figure 3. Again, temperature and rainfall have been normalized for comparison purposes, and so exact coefficients are without
real-world meaning. The largest coefficient is the negative term for temperature in autumn, followed by the positive coefficient for summer
temperature, the positive term and the third largest impact is the negative coefficient for precipitation in summer. Models fitted well from
visual inspection of plots (Figure S4). The variation in the coefficients shows the heterozygosity of the impacts, with positive and negative
impacts occurring across the year. The coefficients appear to be in a wave formation, with both temperature and precipitation increasing
and decreasing in a cyclical manner.
Results for models using parameters from continuous time weather models
Cumulative precipitation increases linearly with increasing month (Figure S5). Over time there exists a generalized trend of increasing precip-
itation with increasing year (year coef = 0.0000318, F(1,6185) = 1676.8, p < 0.05, for full results see Table S7). When mean critic scores are
modeled against cumulative precipitation controlling for AOC and year, as per model option 4 in Figure 1A, a positive correlation is found
(coef = 1779.8, F(1,6171) = 28.1, n = 4521, p < 0.05, R
2
= 0.29, Table S8). This suggests that higher whole year cumulative rainfall is beneficial for
the production of higher quality wines.
Mean monthly temperatures have minima at around 3 to 4 months after September, in December and January. Peaks occur between 9-
and 10-month past September—in June and July (Figure S6), increasing and decreasing in a wave pattern. Sine curves were fitted across each
of the mean monthly temperatures and fitted the data well (mean R
2
is 0.95 and standard deviation of R
2
is 0.02, Table S9). mm and awere
found to be the most variable terms, with means of 8.01 and 286.72 and standard errors of 0.79 and 0.65 (Table S9). sand 4were found to have
means of 0.52 and 3.07 and standard errors of 0.02 and 0.23 (Table S9). This suggests that mean temperature and temperature extremes can
vary more than the timings of when seasons change.
Mean precipitation and the sine parameters of quality were used together as explanatory variables in a GLM, as per model 4 in Figure 1A.
All factors, namely: mean precipitation (MeanPrecip), mean temperature (m; C), amplitude (a; A), periodicity (s, omega), and curve shift (4; phi)
were found to have significant (p < 0.05) positive coefficients (see Table S10 for test statistics). The marginal effects from this model can be
AB
Figure 2. Weather variable model coefficients
(A and B) Weather variable model coefficients for generalized linea r models (GLMs) fitted to explain (A) mean critic score for the whole of Bordeaux and (B) mean
critic score controlling for AOC, Grand Cru status, and year. Bar presence signifies coefficient had a p value of less than 0.05 with standard error bars.
ll
OPEN ACCESS
iScience 26, 107954, October 20, 2023 3
iScienc
e
Article
seen plotted in Figure 3. These results suggest that the greater the extremes in temperature (higher amplitude), the shorter and earlier the
season (decrease in wavelength and positive temperature phase shift), and the larger the mean temperature and mean monthly precipitation,
the higher the mean critic score of the wine.
DISCUSSION
All of the models suggest that weather is an important factor in the determination of wine quality. Taken as a whole, the models suggest that
weather impacts the quality of wine over the course of the year, with importance varying between seasons and aspects of seasonal weather.
This paper includes a new method of examining the weather-wine linkage, using continuous time rather than discrete time periods. It con-
cludes that different aspects of temperature and precipitation are important to quality throughout the year, with high quality requiring periods
of both high and low temperature and precipitation. Exploring weather as a continuous series, we find that higher quality wine is made in years
with greater temperature extremes; earlier, shorter seasons; and potentially a higher mean temperature.
At a regional scale, quality can be seen to have increased over the last several decades (Figure 1C). The cause of such a trend cannot be
distinguished statistically on such a scale. Multiple factors all act in concert to improve wine critics scores, namely: climate change, increasing
technology, increased positivity in critics reviews, and increased matching of wines consumer pref.
4,28,29
. This increase in technology and
consequential changes in wine characteristics since the 1960s have been described as the ‘‘Peynaudization’’ of Bordeaux wines.
30
Controlling
for such trends therefore allows for greater exploration of the impact of individual aspects of wine quality scores.At a local scale, no increase in
trend can be seen to exist, and thus no universal increase in quality can be detected (Table S4). This may be due to the shorter time duration,
variation in wine making techniques, or potentially even that the regional trends reflect critic regional sentiment rather than specific wines and
thus no actualized trend exists.
The overall Bordeaux generalized linear model (model 2 in Figure 1A) suggests the traditional view of high winter precipitation, high sum-
mer temperatures and low precipitation in the summer and autumn lead to high quality grapes (Figure 2A). This combination of precipitation
and sunshine has previously been termed the ‘‘Bordeaux Equation,’’
7,8
and has shaped the global understanding of grapevines.
29,31
However,
the weather location we used to examine these regional scores is in the city center of Bordeaux. This is an urban area not related to viticulture.
This suggests that general regional weather has some effect, but that this is not the whole story.
Like the Bordeaux equation, our overall models also advocate for the impact of out of growing season effects too, corroborating statis-
tically with findings from Bonada et al. (2020) that an increase water availability during the dormancy phase (in our case from precipitation and
in their case due to irrigation) leads to an increase in quality.
21
During winter the grapevines are experiencing dormancy, and a negative
temperature coefficient here suggests a cooling period is required for high quality wine. The models also concur with previous findings
that rainfall during the winter leads to higher quality,
17,21
with agricultural suggestions that this may be due to lower soil salinity.
32
It has
been suggested that more rain in the winter could lead to a better water balance during the growing season, however, it has previously
Figure 3. Model effects plots
Model effects plot showing the vintage score against the sine wave parameters and mean monthly precipitation, with the slope of the line being the coefficients of
each parameter.
ll
OPEN ACCESS
4iScience 26, 107954, October 20, 2023
iScienc
e
Article
been shown that in Bordeaux only 3 of the last 70 years have started the growing season not at full water soil capacity.
2
Flowering, fruit set, and
potentially the onset of berry ripening (depending on the year) all occur in summer, with hotter and drier weather again being suggested to
make high quality wine in this time period, potentially due to lower promotion of major grape diseases.
7,19
The individual AOC model (model 3 in Figure 1A) suggests a more complex view of the relationship and quality. While the same relation-
ships between weather and quality are there during the summer and autumn for temperature and precipitation, additional effects are also
present (Figure 2B). These are the negative effects of temperature in winter and autumn and precipitation in autumn, as well as the positive
effect in temperature and precipitation in spring. The higher number of significant time periods suggests that weather impacts occur right
across the year, with impacts of weather on quality score potentially varying due to the phenological stage of the grapevine.
5
As well as
the dormancy effects, higher precipitation and higher temperatures in spring advocate for wetter and warmer weather for bud and leaf burst.
Finally, cooler and drier weather is best for ripening in autumn and the optimal harvest to make a high-quality vintage. There is also an element
of susceptibility suggested, with the impacts of water deficit on wine quality having greater impact on wine quality during the winter and sum-
mer months, and wine quality being more susceptible to temperatures in summer and autumn.
Combined, the two normalized weather models (models 2 and 3) suggest a difference between the regional and local levels. They suggest
that heterogeneity at the local (AOC) level is being masked when only examining the regional level. They therefore suggest that, in order to
improve the viticultural understanding and hence relevance of such modeling approaches, more local scale weather effects should be
considered.
While these discrete-time suggestions are useful independently, they do not inform about time sequences of weather, which is exactly how
it occurs. To explore this more fully, weather was treated as a continuous time variable. When examined in continuous time, monthly precip-
itation was found to be erratic and thus suited a cumulative approach, with increases being added and forming a linear accumulation. The fact
that such an accumulation is well approximated by a linear model suggests an almost constant aseasonal pattern of precipitation, with the
slope of this linear accumulation being the mean monthly precipitation. Conceptually, examining this mean monthly precipitation allows us to
consider whether a whole year is wet or dry, rather than just the segments set out in the discrete GLMs. The significant precipitation term
suggests that, even controlling for temperature, year, location, and class, a positive relationship exists between mean monthly precipitation
and wine quality (Table S10). Wetter years appear to lead to higher quality wines. Coupled with the discrete time models (Figure 2), this model
therefore suggests that this high rainfall should optimally occur post-harvest and pre-growth, during the dormancy period.
Temperature is not linear, rather it fluctuates according to seasonality and thus can be well approximated with a sine curve. Each of the
parameters of the sine curve informs aspects of a temperature regime over the course of a year. In the GLM exploring the impact of temper-
ature and precipitation parameters on mean quality score (again controlling for year, class, and AOC), the coefficients of each of the terms are
found to be positive (Table S10). This suggests that, aside from being wetter, years that make higher quality wine are characterized by greater
temperature extremes, with a higher mean temperature, and earlier, shorter seasons. While the increase in mean temperature concurs with
previous research,
7,20
more extreme weather suggests colder winters and hotter summers give higher quality. Earlier seasons suggests that
consistently warmer weather during early phenological stages is also beneficial. Warmer weather means lower risk of frost, suggesting that
damage to crops extends beyond losses and into quality.
22,33
Warmer weather also potentially suggests that higher metabolic rates and
higher photosynthetic rates lead to higher quality grapes. Shorter seasons suggest that the cooling of temperatures toward the end of
the growing season may positively impact the ripening of grapes. With increases in both mean and extreme temperatures predicted across
France,
34
and changes in timing and length of growing season also predicted across growing regions,
35
this leads to the potential suggestion
that wine in Bordeaux may continue to improve over time.
Among these trends there still exists the question of the local versus regional disparities. The differences between the local and regional
models in both their model coefficients and statistical significances suggest the impacts from the local scale are being masked when exam-
ined at the regional scale. This may be because of the individual differences in weather, or due to disparities in the wine making in each of the
AOCs. For each of the models that are built, the Grand Cru classification system suggests significant differences between the classification
levels, and similarly AOC level differences appear to exist (see Tables S6–S10). However, one potential source of variation in these data may be
bias in expert opinion.
36–40
Statistically, it is impossible with these data to fully disentangle wine bias from perceived quality. Future studies
where the data for wine quality is both linked to local weather and also rated in a double-blind fashion, will be necessary to capture the degree
to which expert bias informs or weakens our predictions.
We suggest that such variation in quality classification between regions has masked local variation in impacts of weather on quality across
the year. While we accept that biases exist within wine, both for a specific locale or classification, we have also demonstrated a significant
shared understanding of quality.
There are clear management implications of these findings, which result from trying to optimize the environment for vine growth and fruit
composition. Our models suggest that water regime is critical in determining higher quality wines. Specifically, to increase wine quality, ideal
conditions include high water abundance during the winter months and low water abundance in the summer, coupled with high tempera-
tures. Climate change in Bordeaux will likely lead to more extreme weather, with variation depending on the location.
41
While some places
will be in drought, others will encounter less total rainfall punctuated with short heavy rain events.
41
For red wine production we suggest that if
irrigation were to be considered, it would be best to target the water regimes highlighted in this work: a well-replenished soil water profile
over the winter months followed by moderate to severe water deficits during the summer months (depending on yield and wine style con-
siderations). In cases where heavy rainfall could be an issue in summer, increased drainage, erosion control, or, at an extreme, rainfall covers
could be necessary. With regard to temperature management, summer management strategies which promote localized higher
ll
OPEN ACCESS
iScience 26, 107954, October 20, 2023 5
iScienc
e
Article
temperatures are suggested (especially during the ripening period). This can be achieved by increasing defoliation around the berries to
reduce shading, but caution is warranted to guard against exposing fruit to temperature extremes. Finally, we agree with common practices
of avoiding frost damage by raising temperatures around the vines during the spring months. With predicted phenological and weather
changes leading to hotter and earlier summers, our results suggest that average Bordeaux quality scores may continue to increase.
This paper has explored the impact of weather on wine, seeking to determine the optimal growing conditions for high quality Bordeaux
vintages. It explores the infamous Bordeaux equation, finding that the equation works well for explaining regional patterns, but that for in-
dividual AOCs the weather impacts occur over the course of the year. Exploring this weather sequentially, this paper finds that higher quality
wine is made in years with greater temperature extremes; earlier, shorter seasons; and a higher mean temperature. This all suggests that as
climate change increases, the wine quality may continue to get better.
Limitations of the study
We appreciate that the study was only conducted using ratings for Bordeaux chateaus, and that the corresponding limitations are therefore
that we can only control for the winery at the winery level. We cannot control for the winemaker changing, or any potential changes in the exact
plots used to make the wines. Finally, it is statistically impossible to tell the difference over time between improvements in wine due to climate
and winemaker (and hence this trend has been removed from dataset). Despite these limitations, we have shown a robust trend within the
dataset concerning the impact of seasonal weather on the quality of wines.
STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:
dKEY RESOURCES TABLE
dRESOURCE AVAILABILITY
BLead contact
BMaterials availability
BData and code availability
dMETHOD DETAILS
BWeather data
BQuality data
dQUANTIFICATION AND STATISTICAL ANALYSIS
BGeneralized linear models
SUPPLEMENTAL INFORMATION
Supplemental information can be found online at https://doi.org/10.1016/j.isci.2023.107954.
ACKNOWLEDGMENTS
The authors would like to thank the data providers and F. Lovell-Read for her help with sine curves, C. Bernard for his support and T. Taberer
for her proof-reading. This work was supported by funding from the Biotechnology and Biological Sciences Research Council (BBSRC) [grant
number BB/M011224/1].
AUTHOR CONTRIBUTIONS
A.W. and S.G. conceived the main ideas and initial methodology; all authors contributed improvements to ideas and the methodology; A.W.
collected the data and A.W. and S.G. analyzed the data; A.W. led the writing of the manuscript, with edits from all authors. All authors contrib-
uted critically to the drafts and gave final approval for publication.
DECLARATION OF INTERESTS
We declare no conflicts of interest.
INCLUSION AND DIVERSITY
We support inclusive, diverse, and equitable conduct of research.
Received: May 18, 2023
Revised: June 26, 2023
Accepted: September 13, 2023
Published: October 11, 2023
ll
OPEN ACCESS
6iScience 26, 107954, October 20, 2023
iScienc
e
Article
REFERENCES
1. Lobell, D.B., Field, C.B., Cahill, K.N., and
Bonfils, C. (2006). Impacts of future climate
change on California perennial crop yields:
Model projections with climate and crop
uncertainties. Agric. For. Meteorol. 141,
208–218. https://doi.org/10.1016/j.
agrformet.2006.10.006.
2. van Leeuwen, C., and Darriet, P. (2016). The
Impact of Climate Change on Viticulture and
Wine Quality. J. Wine Econ. 11, 150–167.
https://doi.org/10.1017/jwe.2015.21.
3. Roucher, A., Aristodemou, L., and Tietze, F.
(2022). Predicting wine prices based on the
weather: Bordeaux vineyards in a changing
climate. Front. Environ. Sci. 10, 2266. https://
doi.org/10.3389/FENVS.2022.1020867.
4. van Leeuwen, C., Friant, P., Chone
´, X.,
Tregoat, O., Koundouras, S., and
Dubourdieu, D. (2004). Influence of Climate,
Soil, and Cultivar on Terroir. Am. J. Enol. Vitic.
55, 207–217.
5. Jones, G.V., and Davis, R.E. (2000). Climate
Influences on Grapevine Phenology, Grape
Composition, and Wine Production and
Quality for Bordeaux, France. Am. J. Enol.
Vitic. 51, 249–261.
6. Ashenfelter, O., and Jones, G.v. (2013). The
Demand for Expert Opinion: Bordeaux Wine.
J. Wine Econ. 8, 285–293. https://doi.org/10.
1017/JWE.2013.22.
7. Ashenfelter, O., Ashmore, D., and Lalonde, R.
(1995). Bordeaux Wine Vintage Quality and
the Weather. Chance 8, 7–14. https://doi.org/
10.1080/09332480.1995.10542468.
8. Ashenfelter, O. (2008). Predicting the Quality
and Prices of Bordeaux Wine. Econ. J. 118,
F174–F184. https://doi.org/10.1111/J.1468-
0297.2008.02148.X.
9. Almaraz, P. (2015). Bordeaux wine quality and
climate fluctuations during the last century.
Clim. Res. 64, 187–199. https://doi.org/10.
2307/24897566.
10. Baciocco, K.A., Davis, R.E., and Jones, G.V.
(2014). Climate and Bordeaux wine quality:
identifying the key factors that differentiate
vintages based on consensus rankings.
J. Wine Res. 25, 75–90. https://doi.org/10.
1080/09571264.2014.888649.
11. Jones, G., and Storchmann, K.H. (2001). Wine
market prices and investment under
uncertainty: an econometric model for
Bordeaux Crus Classe
´s. Agric. Econ. 26,
115–133. https://doi.org/10.1016/S0169-
5150(00)00102-X.
12. Martin, G.J., Mazure, M., Jouitteau, C.,
Martin, Y.-L., Aguile, L., and Allain, P. (1999).
Characterization of the Geographic Origin of
Bordeaux Wines by a Combined Use of
Isotopic and Trace Element Measurements.
Am. J. Enol. Vitic. 50, 409–417.
13. Ramirez, C.D. (2008). Wine Quality, Wine
Prices, and the Weather: Is Napa ‘‘Different’’?
J. Wine Econ. 3, 114–131. https://doi.org/10.
1017/S1931436100001164.
14. Soar, C.J., Sadras, V.O., and Petrie, P.R.
(2008). Climate drivers of red wine quality in
four contrasting Australian wine regions.
Aust. J. Grape Wine Res. 14, 78–90. https://
doi.org/10.1111/J.1755-0238.2008.00011.X.
15. Corsi, A., and Ashenfelter, O. (2019).
Predicting Italian Wine Quality from Weather
Data and Expert Ratings. J. Wine Econ. 14,
234–251. https://doi.org/10.1017/JWE.
2019.41.
16. di Vittorio, A., and Ginsburgh, V. (1996). Red
wines of Medoc vintages from 1949 to 1989 at
Christie’s auctions. J. Soc. Stat. Paris
137, 19–49.
17. Lecocq, S., and Visser, M. (2006). Spatial
Variations in Weather Conditions and Wine
Prices in Bordeaux. J. Wine Econ. 1, 114–124.
https://doi.org/10.1017/S1931436100000158.
18. Webb, L., Whetton, P., and Barlow, E. (2008).
Modelling the relationship between climate,
winegrape price and winegrape quality in
Australia. Clim. Res. 36, 89–98. https://doi.
org/10.3354/cr00739.
19. Byron, R.P., and Ashenfelter, O. (1995).
Predicting the Quality of an Unborn Grange.
Econ. Rec. 71, 40–53. https://doi.org/10.
1111/J.1475-4932.1995.TB01870.X.
20. Gambetta, G.A., and Kurtural, S.K. (2021).
Global warming and wine quality: are we
close to the tipping point? OENO One 55,
353–361. https://doi.org/10.20870/OENO-
ONE.2021.55.3.4774.
21. Bonada, M., Edwards, E.J., McCarthy, M.G.,
Sepu
´lveda, G., and Petrie, P.R. (2020). Impact
of low rainfall during dormancy on vine
productivity and development. Aust. J.
Grape Wine Res. 26, 325–342. https://doi.
org/10.1111/AJGW.12445.
22. Nemani, R.R., White, M.A., Cayan, D.R.,
Jones, G.v., Running, S.W., Coughlan, J.C.,
and Peterson, D.L. (2001). Asymmetric
warming over coastal California and its
impact on the premium wine industry. Clim.
Res. 19, 25–34. https://doi.org/10.3354/
CR019025.
23. Jones, G.V., White, M.A., Cooper, O.R., and
Storchmann, K. (2005). Climate change and
global wine quality. Clim. Change 73,
319–343. https://doi.org/10.1007/s10584-
005-4704-2.
24. Wickham, H., Averick, M., Bryan, J., Chang,
W., McGowan, L., Franc¸ois, R., Grolemund,
G., Hayes, A., Henry, L., Hester, J., et al.
(2019). Welcome to the {tidyverse. J. Open
Source Softw. 4, 1686. https://doi.org/10.
21105/joss.01686.
25. Mun
˜oz Sabater, J. (2019). ERA5-Land hourly
data from 1981 to present. Copernicus
Climate Change Service (C3S) Climate Data
Store (CDS). https://doi.org/10.24381/cds.
e2161bac.
26. Kusch, E., and Davy, R. (2022). KrigR—a tool
for downloading and statistically downscaling
climate reanalysis data. Environ. Res. Lett. 17,
024005. https://doi.org/10.1088/1748-9326/
ac48b3.
27. Wijnimport, B. (2022). BordOverview. https://
www.bordoverview.com/.
28. Rodo
´, X., and Comı
´n, F.A. (2000). Links
between large-scale anomalies, rainfall and
wine quality in the Iberian Peninsula during
the last three decades. Global Change Biol. 6,
267–273. https://doi.org/10.1046/J.1365-
2486.2000.00299.X.
29. Dalu, J.D., Baldi, M., Marta, A.D., Orlandini,
S., Maracchi, G., Dalu, G., Grifoni, D., and
Mancini, M. (2013). Mediterranean climate
patterns and wine quality in North and
Central Italy. Int. J. Biometeorol. 57, 729–742.
https://doi.org/10.1007/S00484-012-0600-4/
FIGURES/15.
30. Styles, O. (2004). Emile Peynaud Dies at 92
(Decanter).
31. Grifoni, D., Mancini, M., Maracchi, G.,
Orlandini, S., and Zipoli, G. (2006). Analysis of
Italian Wine Quality Using Freely Available
Meteorological Information. Am. J. Enol.
Vitic. 57, 339–346.
32. Isidoro, D., and Grattan, S.R. (2011).
Predicting soil salinity in response to different
irrigation practices, soil types and rainfall
scenarios. Irrigat. Sci. 29, 197–211. https://
doi.org/10.1007/S00271-010-0223-7/
FIGURES/8.
33. Jones, J.E., Wilson, S.J., Lee, G., and Smith,
A.M. (2010). Effect of frost damage and
pruning on current crop and return crop of
Pinot Noir 38, 209–216. https://doi.org/10.
1080/01140671.2010.498402.
34. Machard, A., Inard, C., Alessandrini, J.M.,
Pele
´, C., and Ribe
´ron, J. (2020). A
Methodology for Assembling Future
Weather Files Including Heatwaves for
Building Thermal Simulations from the
European Coordinated Regional
Downscaling Experiment (EURO-CORDEX)
Climate Data. Energies 13, 3424. https://doi.
org/10.3390/EN13133424.
35. Ivits, E., Cherlet, M., To
´th, G., Sommer, S.,
Mehl, W., Vogt, J., and Micale, F. (2012).
Combining satellite derived phenology with
climate data for climate change impact
assessment. Global Planet. Change 88,
85–97. https://doi.org/10.1016/J.
GLOPLACHA.2012.03.010.
36. Cliff, M.A., and King, M.C. (2007). The
evaluation of judges at wine competitions:
the application of Eggshell plots. J. Wine Res.
8, 75–80. https://doi.org/10.1080/
09571269708718104.
37. Cao, J., and Stokes, L. (2010). Evaluation of
Wine Judge Performance through Three
Characteristics: Bias, Discrimination, and
Variation. J. Wine Econ. 5, 132–142. https://
doi.org/10.1017/S1931436100001413.
38. Castriota, S., Curzi, D., and Delmastro, M.
(2013). Tasters’ bias in wine guides’ quality
evaluations 20, 1174–1177. https://doi.org/
10.1080/13504851.2013.797552.
39. Livat, F., and Vaillant, N.G. (2006). Expert
opinion and brand reputation: An analysis
from a French Cuban cigars guidebook. Appl.
Econ. Lett. 13, 97–100. https://doi.org/10.
1080/13504850500390564.
40. Reuter, J. (2009). Does Advertising Bias
Product Reviews? An Analysis of Wine
Ratings. J. Wine Econ. 4, 125–151. https://doi.
org/10.1017/S1931436100000766.
41. Drappier, J., Thibon, C., Rabot, A., and Geny-
Denis, L. (2017). Relationship between wine
composition and temperature: Impact on
Bordeaux wine typicity in the context of
global warming—Review 59, 14–30. https://
doi.org/10.1080/10408398.2017.1355776.
ll
OPEN ACCESS
iScience 26, 107954, October 20, 2023 7
iScienc
e
Article
STAR+METHODS
KEY RESOURCES TABLE
RESOURCE AVAILABILITY
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Andrew Wood
(wood_and@hotmail.com).
Materials availability
This study did not generate new unique reagents.
Data and code availability
This paper analyses existing, publicly available data. All data is publicly available at locations referenced within the text. Wine data is available
here: https://www.bordoverview.com/ and climate data from: https://cds.climate.copernicus.eu/cdsapp#!/dataset/10.24381/cds.68d2bb30?
tab=overview as documented in the reference list and the key resources table. Any additional information required to reanalyse the data re-
ported in this paper is available from the lead contact upon request.
METHOD DETAILS
This study is an analysis of the linkage between two key weather variables: temperature and precipitation, and the critic scores at a regional
and local (AOC) level for the Bordeaux region. As in previous approaches,
7,23
analysis of the relationship between weather variables and wine
quality scores are based upon the assumption that beneficial weather influences will lead to higher wine quality. The quality-weather inter-
action methodology can be split into two approaches. Both approaches fit critics scores against weather using a generalized linear model
(GLM), but each uses different quantifications of weather as variables. The first approach uses the mean temperature and precipitation during
discrete time-steps as the variables; the second approach uses the parameters of functions fitted to the temperature and precipitation data as
the variables in statistical models of quality. All data extraction and analysis were undertaken in R version 4.2, using the tidyverse
24
and baseR
packages.
Weather data
Historical weather data were extracted from the ERA-5 land reanalysis weather dataset
25
for each Bordeaux AOC region and central Bordeaux
on a monthly time-step using the Krigr package
26
(see Table S1 for AOCs and their locations). ERA-5 land is a high temporal and spatial
resolution interpolated dataset which is available on a 0.1grid at time scales varying from hourly to monthly since January 1950.
25
Temper-
ature and precipitation data were extracted on a monthly time-step for a 1km radius from the latitude and longitude point given per AOC for a
period of January 1950 to December 2020. Temperature was measured in Kelvin (K), and precipitation in meters (m), both SI units for their
respective measures. A growing season was defined as running from 1
st
November to 31
st
October, with harvest occurring at the end of
the year. This aligns with standard growing season measurements (May-October) but extends them to include the overwintering effects
(November-May).
Monthly weather variables were expectedly found to be strongly autocorrelated (Appendix 6), and thus unable to be used individually for
building GLMs. Accordingly, months were grouped into autocorrelated groups, which could be roughly thought of as seasons. These groups
were determined by those consecutive months by which the inter-month temperature or precipitation Pearson correlation coefficient was
above 0.4. Winter was defined as being November and December, Spring as January to May, Summer as June to August, and Autumn as
September and October.
REAGENT or RESOURCE SOURCE IDENTIFIER
Deposited data
ERA-5 Land Hourly Weather Data https://doi.org/10.24381/cds.e2161bac
BordOverview Wine Database https://www.bordoverview.com/
Software and algorithms
RStudio
tidyverse packages
Krigr package
ll
OPEN ACCESS
8iScience 26, 107954, October 20, 2023
iScienc
e
Article
A second way of dealing with this temporal autocorrelation is to describe temporal variation in the weather data to a continuous function.
Precipitation is erratic, and hence to examine it in a continuous fashion, cumulative precipitation was used. The cumulative monthly precip-
itation was modeled linearly using a GLM, meaning that the cumulative precipitation can be approximated using the mean monthly
precipitation.
Monthly mean temperature across the year were also described using a sine curve, as in Figure 1B. Non-linear least squares used to fit the
data to the following equation:
asinðsM+4Þ+m
Where M is the number of months since October, and a;s;4;and mare parameters to be fitted. Figure 1B shows how each parameter relates
to a part of the sine curve. Starting values for parameters were chosen based upon a fixed periodicity for sand 4(s=2p
12;4=3.swas
chosen because of the annual cycle, hence division by twelve, and 4was chosen as October is 3 months after the approximate peak annual
temperature. aand mwere fitted using the maximum, minimum and mean values of the temperature data per site and year using the equa-
tions: a=ðmax ðtÞmin ðtÞÞ=2;m=mean ðtÞ.
Quality data
Annual quality scores were collected on two scales, regional and local. Regional critic scores are based on the opinions of how Bordeaux
performed as a whole, with individual variation largely ignored and general trends suggested.
20
Local scores are based on individual wines,
which are tasted en primeur and then rated based upon this premature wine.
27
For each, publicly available wine critic, scores were
transformed into a standardized 0–100 scale. Whole region scores were available for the period of 1950–2020 and were drawn from several
sources,
9,20
with additional data drawn from online vintage charts (see Table S2). Regional primeur critics scores for the period 2014–2020
were compiled by Bolomey Wijnimport
27
and consist of published ratings from major wine experts from France, UK, US, the Netherlands
and Germany. All publications were tested for correlation between their scores, and 14 wine publications were chosen based on having a
Pearson Correlation Coefficient of more than 0.4 with at least 3 other selected publications. This cut-off was chosen as a liberal threshold
for the inclusion of publications in our analysis. The chosen cut-off means that at least 16% of the variance in critics scores can be attributed
to a shared understanding of quality across 3 other publications. These scores were standardized within publications such that eachwas on a
0 to 100 scale. If wines were rated by more than one publication (78% wines) mean standardized scores were taken.
QUANTIFICATION AND STATISTICAL ANALYSIS
Generalized linear models
Generalised linear models (GLMs) were used to explore the relationship between weather and quality. A summary diagram of the models
fitted can be found in Figure 1A. In the diagram, the number in the bottle refers to the model number, and tildes mean ‘‘is a function of’’.
GLMs were chosen because they are a flexible modeling approach which does not require any particular error structure or variance.
As per model 1 in the summary diagram (Figure 1A), mean quality scores over time were investigated using a GLM with a binomial
distribution and logit link, controlling for the critic. This was due to the bounded nature of the scoring system (0–100) and decreasing variation
in quality scores over time. This allows for a comparison of longer-term regional versus local trends over time. At a regional scale, the yearly
trend interacted with region to determine the differential baseline trend over time.
For both the regional (model 2, Figure 1A) and local (model 3, Figure 1A) level scores, a GLM with a Gaussian distribution was fitted be-
tween the mean of the annual critic scores and the normalized temperature and precipitation during each weather grouping, controlling for
year and, in model 3, the AOC and Grand Cru status of the vineyard. Normality of residuals was checked visually using a qqplot and residuals
vs. fitted plot. The seasonal means of temperature and precipitation were normalized by subtracting the mean of the weather group and
dividing by the standard deviation (Standard Score normalisation). This means that the GLM examined the proportional positive or negative
impact of each of the variables on quality, rather than the absolute value. In normalizing such environmental data, comparisons could be made
between the variables and therefore the relative contributions of each can be ascertained. The year was controlled for by adding it as a contin-
uous variable, and, in the AOC model, the locality was also controlled for by adding it as a factor.
GLMs were also built to compare the mean vintage score with weather treated as a continuous function (model 4, Figure 1A). This model
utilized an additive Gaussian GLM to examine how each of these variables explained the variance within the mean critic scores, controlling for
AOC, year, and class. These models were run for both temperature and precipitation separately and then together. We also ran a GLM
without year as an explanatory variable due to covariance.
ll
OPEN ACCESS
iScience 26, 107954, October 20, 2023 9
iScienc
e
Article