Grape Harvest Dates and Temperature Variations in Eastern France since 1370

I. Chuine, P. Yiou, N. Viovy, B. Seguin, V. Daux and E. Le Roy Ladurie

Supplementary Methods

Data sources

Our grape harvest date series (Ht) is a corrected and updated harvest date series from Le Roy Ladurie 1 covering 1370 to 1977 and completed until 2003 for the Burgundy region (Fig. S1).

Figure S1 Grape harvest date series of Burgundy from 1370 to 2003 (Ht). Yearly harvest dates (in days since Sept. 1st) in black and the averages of non-overlapping 50-year intervals in red. Gregorian calendar corrections are taken into account.

We collected the grape harvest dates from up to 18 cities or villages in Burgundy (the actual number depends on the year) since 1370. To avoid possible biases due to the variability of data availability each year, we chose Dijon — the longest series overall and the only one available for some periods — as the reference series. However, the 17 other series were used to take into account the regional variability. For that purpose, we first standardized all series such that they present the same average date as Dijon over their common recorded period. We then computed for each year t the harvest date Ht as the median date among all available standardized dates including Dijon.

To fit the phenological model parameters for the vine variety Pinot Noir we used observed flowering dates, véraison dates and harvest dates of Pinot Noir grown in Colmar (France), with corresponding daily temperatures for the 1964-2001 period (incomplete series, n= 28 years for flowering and n= 26 years for véraison) 2.

Phenological model development

Woody species phenology can be accurately simulated using well validated process-based models and daily mean temperatures 3. Such models are species dependent—or in the case of vine, variety dependent — and their parameters are estimated using long time series of phenological observations. We used a set of recent observations of the timing of development of Pinot Noir made between 1964 and 2001 in Colmar, France to develop such a model for this variety (E. Duchêne and C. Schneider (INRA Colmar), pers. comm.). Grape maturation date depends primarily on the temperature conditions of the preceding months 4. Before maturation, two important development phases occur, flowering and véraison (changing in colour from green to either white or red), about two months and one month respectively before maturation. The correlation of flowering date or véraison date with harvest date is extremely high (r=0.90, p<0.0001 and r=0.92, p<0.0001 respectively), meaning that most of the climate effect on harvest date takes place before véraison and flowering. The delay between véraison and harvest is also very much constant (standard error of this delay is 0.74 days in the Colmar dataset, n=26) and does not depend on temperature, but rather on the interaction between the phytosanitary state of the vineyard after véraison and the occurrence and amount of rain. For these reasons the inversion of the models was based on an estimated véraison date, which occurs on average 33.5 day before harvest in Burgundy. The véraison date was simulated with the model described in here after. The response to temperature differs between the period leading to flowering and that leading to véraison, this is why two different models, one simulating the flowering date and another simulating the veraison date were developed. For flowering dates we adopted the methodology of Chuine 5.

Flowering date (tf) model:

tf such that=F*,(1)

Rf(xt) = ,(2)

with xt the mean temperature of day t and F*, Tb, t0 fitted parameters. Rf is classically referred to as the rate of forcing or the degree-day function and Tb as the threshold temperature above which degrees have an effect on the organs development.

For the veraison date model we adopted the model developed by Chuine & Beaubien 6 for fruit maturation dates prediction. This model considers two phases of development of the fruit which differ in the way organs development is affected by temperature. The first phase (tf to t1) is a short phase of cell elongation, followed by a phase (t1 to tv) of photosynthate accumulation.

Véraison date (tv) model:

tv such that = P*,(3)

t1 such that =C*,(4)

Rv(xt) = ,(5)

Rp(xt) = mxt3 + nxt2 + oxt + p,(6)

m = ,(7)

n = ,(8)

p = ,(9)

o = ,(10)

with xt the mean temperature of day t and P*, C*, To, a, b fitted parameters. Rp(xt) is the photosynthetic activity curve as described by Pisek 7, 3rd degree polynomial of temperature, varying between 0 when temperature is superior or equal to 40°C or inferior or equal to -5°C, and 100 when temperature reaches the optimal temperature To which is comprised between 10°C and 25°C. Rv, similarly to Rf, is a rate of forcing corresponding to the phase of elongation of the fruit cells, it is a sigmoid function of the temperature and varies between 0 and 1 and equals 0.5 when the daily temperature xt equals b.

Parameters were fitted using the Metropolis algorithm according to Chuine et al. 8 with the first continuous part of the time series (17 years for flowering and 16 years for veraison), and the estimates were cross-validated (holdout cross-validation method and not leave-one-out cross-validation) with the end of the time series (11 years for flowering and 10 years for veraison). Their values as well as the percentage on variance explained by each model are provided in Table S1.

We determined the temperature anomalies with respect to a reference temperature (April to August mean temperature of Dijon between 1960 and 1989) by inverting the model. The Dijon temperature data was provided by J.M. Moisselin (Météo-France). The inversion consisted in fitting the average anomaly temperature that would provide the observed harvest date. Reconstructed anomalies correspond to the April to August period because this is the period during which temperature affects the vine development according to the flowering and veraison models.

Comparison with other temperature data sets

We performed moving correlations with the data sets of Table 1 to show the stability of the correlations between the reconstructed temperature and the other long data. Those correlation figures are available from P. Yiou ().

Table S1 Parameter estimates, goodness of fit and validation of the phenological models developed for Pinot Noir

flowering / véraison
Model fit
Tb (°C) / 8.63 / a / -0.504
F* (°C) / 412.97 / b (°C) / 17.541
t0 (DOY)† / 92 / C* / 15.96
To (°C) / 20.34
P* / 3605.9
n / 17 / n / 16
R2 / 0.846 / R2 / 0.897

Cross-validation

n / 11 / n / 10
R2 / 0.818 / R2 / 0.913

Parameter estimates and percentage of variance explained (R2) by the flowering model and the véraison model. n, number of years of observations used to fit and cross-validate the model. † DOY, day of the year.

Reference List

1.Le Roy Ladurie, E. Histoire du Climat depuis l'An Mil (Champs Flammarion, Paris, 1983).

2.Duchêne, E. & Schneider, C. Agronomie In Press (2004).

3.Chuine, I., Kramer, K. & Hänninen, H. in Phenology: An Integrative Environmental Science (ed. Schwarz, M. D.) 217-235 (Kluwer, Netherlands, 2003).

4.Jones, G. V. in Phenology: An integrative environmental science (ed. Schwartz, M. D.) 523-539 (Kluwer Academic Publishers, London, 2003).

5.Chuine, I. J. Theoretical Biol. 207, 337-347 (2000).

6.Chuine, I. & Beaubien, E. Ecology Lett. 4, 500-510 (2001).

7.Pisek, A., Larcher, W., Vegis, A. & Napp-Zin, K. in Temperature and Life (eds. Precht, M., Christophersen, J., Hensel, H. & Larcher, W.) 102-126 (Springer Verlag, Berlin, 1973).

8.Chuine, I., Cour, P. & Rousseau, D. D. Plant, Cell & Environment 21, 455-466 (1998).

1