G. Due et al., MODELING GRAPEVINE PHENOLOGY AGAINST WEATHER - CONSIDERATIONS BASED ON A LARGE DATA SET, Agricultural and forest meteorology, 65(1-2), 1993, pp. 91-106
Phenological data for 11 districts, 12 varieties and four seasons were
modelled against averages of daily records of maximum and minimum tem
peratures and wet bulb depression. Weather data were taken from the ne
arest government weather station, which was commonly up to 10 km from
the vineyard. The use of so-called 'heat degree day' methods (e.g. Win
kler et al., 1974) were abandoned because the inherent correlation bet
ween elapsed time and the value of an accumulation confounds regressio
n procedures and interpretation; averages, rather than accumulations,
were used as variates in all regressions. The model of budburst involv
ed small district and year effects, and accounted for 94.5% of the var
iation in the data. This demonstrated that off-site weather data can p
rovide effective models. Models of flowering and harvest accounted for
89.2% and 92.5% of the variation in the data respectively, but involv
ed large district and year effects. Averages of weather taken from 3 w
eeks or more prior to all events were significantly associated with th
e dates of those same events. The wet bulb depression provided signifi
cant and useful variates in the model of budburst. In addition to othe
r associations, the date of flowering was significantly associated wit
h the weather conditions prior to budburst. Likewise, the date of harv
est was significantly associated with weather conditions prior to both
budburst and flowering. Budburst and harvest dates were also associat
ed with averages of weather taken over the 10 days just before the eve
nt. The variety of these associations encourages the use of a wider ra
nge of variates than is common in phenological studies. The district a
nd year effects in the case of the flowering and harvest models could
result from the complicated effect of the canopy on vine microclimate,
or the influence of short term weather events such as storms, or, in
the case of harvest, they could result from imprecise definitions of m
aturity. Time series methods are advocated as a means to cope better w
ith short term weather events and also to facilitate physiological int
erpretation of phenological models.