Veraison was recorded as a percent estimate of the cluster with color and softness changes

Clusters were chosen from the most basal position on the shoot arising from the most basal bud. Flowering was estimated by percent of the cluster appearing to have caps fallen and flowers showing. At the point where 50% of the caps on flowers have fallen away, called anthesis, the cluster is considered at full bloom . Flowering is recorded as a percent estimate of the cluster in bloom. These clusters were then tagged with fluorescent tape loosely and followed subsequently for veraison and maturity. Records were taken every two to three days during each of the phenological stages. At each stage of monitoring, researchers calibrate observations with each other and with photographs from previous years. The Wang and Engel model uses a maximum temperature of 40°C, which may be the biological threshold for grapevine growth . The biological optimal temperature for grapevines is likely around 25°C , and when temperatures exceed 30°C, there are impacts on anthocyanins , and temperatures exceeding 37°C decreased coloration in grape berry skin and degradation of aromatic compounds .

Therefore, we include the variable of cumulative number of days when temperature reached a maximum at or over 40°C in our models of veraison, the only phenological stage that encounters these days, round pot with variety as a random effect . This study spanned four years and is ongoing to evaluate the sensitivity of different varieties to climate change. It is important to note that timing and duration of the winter pruning was variable, which could have introduced error because the timing of pruning can impact budburst . Some varieties were discontinued from the study because of death, disease, or pest damage. Some varieties were only included in later years of the study once they reached maturation. Measurements of percent budburst, flowering, and veraison were converted to GDD using R . For each stage, a linear model was fit with phenological development as a percent as the response variable and time as the independent variable. The fitted model was used to estimate the day a cluster reached 50% budburst, flowering, or veraison. Data were cleaned by removing individuals with illogical estimates for timing. It was determined for these removed points that too few measurements were made for those vines, and the individuals removed were from the year 2015. The limits for estimates were based on observational data. If budburst was predicted earlier than day 60 of the year, this individual was removed, because this was earlier than measurements were recorded.

Flowering was limited to day 111 of the year, and veraison was limited to begin at day 175 of the year. We obtained for each of the four years the GDD’s required to reach the three phenophases for an individual plant. These GDD’s were the response variable of the hierarchical models described in the next section. The days over 40°C from January until September were also quantified for each year. To quantify the variation in growing degree days across and within grapevine varieties, we used Bayesian linear mixed effect models as implemented in the package RStanArm . The default set of priors was used for the RStanArm package. A model using GDD’s with a base temperature of 10°C were compared against a model with base 0°C temperature. Models were fit using GDD with a base temperature of 0°C on the day of a phenological event as the response variable. Separate models were fit for the three phenological stages: budburst, flowering, and veraison. For the budburst and flowering stages, GDD was the response variable with “utility” as a fixed effect and “geography/variety” as random effects. For veraison, we included “utility” as a fixed effect, “geography/variety” as random, as well as “days above 40°C” as a random effect tied to variety. The climate data were summarized in R from the raw CIMIS data, and the cumulative measurements for each stage included the weeks prior to each stage. The explanatory models of maximum daily temperature, and cumulative number of days with temperatures reaching over 40°C, and year as fixed effects were incorporated into models for each stage. Overall, we added grape utility, in terms of wine or table grape or both, geographic groups, and country of origin as nested random variables .

The advantage to using a fixed effect model to predict GDD by variety is that we fill gaps by including data from other varieties’ responses to predict individual variety response. We quantify variety-level GDD by leveraging information from all varieties together. Rather than using an average for each variety by year, we utilize the temporal redundancy to estimate consistency across years, and we can see from the average of all varieties together which ones fluctuate the most during years with more change in climate. We used the posterior predictive check of a PSIS diagnostic plot to ensure khats were all less than 0.7. The preference for rstanarm to evaluate these mixed effects models is based on the Bayesian approach using MCMC, rather than restricted maximum likelihood estimation, which tends to underestimate uncertainties . This Bayesian approach estimates uncertainty for all the model levels, including our random effect of variety which contains 137 or less estimated parameters. When ELPD-difference was compared, the differences in log probability for the five models were almost completely within their individual standard errors. For this reason, we can consider all five models substantially predictive, but we chose the top model based on lowest ELPD-difference and biological relevance of the variables in the model. For veraison, the differences in log probability , were not within their individual standard errors . The models of veraison were improved by the addition of the variable, “days above 40,” referring to the cumulative number of days with daily maximum temperatures at or above 40°C. The genetically identified geographic origins provided by Bacilieri et al. Supplemental Information added predictive information to our final models for budburst, flowering, and veraison. We see differences in sensitivity to climate across stages for each of the geographic groups, visualized by the coefficient of variation over the four years analyzed . The intercepts reported in Supplementary Table 2, in terms of growing degree days, provide a predictive range to expect phenological variability from these groups. The range of phenological timing for specific cultivars can help match varieties with ideal climates and regions. There are varieties from each of these regions with the potential to be late ripening. From the Italian Peninsula, there is Dolcetto with a relatively early veraison and Aglicanico with a relatively late and variable veraison . The timing of stages can be extremely consistent, such as with Gamay Noir from Western Central Europe, but there can also be a wider range of timing like that of Mourvedre, from the same region. Therefore, while region is predictive, analyzing the timing for specific varieties is also useful when selecting alternative varieties for planting. years. Budburst had the highest coefficient of variation, likely due to the impact of conditions during dormancy . There may also be an accumulation of climatic impacts over the season resulting in the highest variability in timing at veraison. In a previous common garden experiment, round plastic planter the timing of maturity also had the largest standard error with more predictable timing for budburst and flowering . Sensitivity across stages does not have a strong correlation, but Budburst and Flowering seem to have the strongest relationship, with the highest R2 for the Balkans geographic region at 96% . The parameter estimates of the three models reported the highest sigma for variety for all three models. From an ecological perspective, a vineyard is a system that responds to its environment. This system includes the soil, international varieties, and the climate. We modelled the response of varieties’ phenological timing to climate, and the results present unique sensitivities to climate over 4 years.

Geographic origin and cultivated utility of grapes explain some of the variation seen in phenological timing, which we expect is driven by physiological differences. Accumulation of daily temperature in our model is strongly correlated with phenological stage occurrence, which agrees with past modelling of growing degree days and phenology . Previous models have used individual parameters for growing degree days and base temperature based on the cultivar . Duchêne et al. used daily maximum temperature rather than GDD in their models to predict phenology, and their models included a stage specific base temperature. Our model is unique by including variety specific response to cumulative days above 40°C. We expected this to impact the timing of veraison for some varieties with higher sensitivity to heat stress. The chosen model for veraison included the variable of “days above 40°C,” which is in part due to the timing of high heat days, typically occurring later in the season, during this stage. This model outranked models for veraison that nested geographic origin, indicating that the effect of high temperature is not variety specific. The general intercepts of the models for each stage predict the mean GDD required to reach each phenological stage, and the intercepts for each variety indicate the specific GDD requirement for each variety . The general intercept for the three stages was 199 GDD for budburst, 836 GDD for flowering, and 1,699 GDD for veraison . We may expect for other regions and in California’s future that heat stress may impact flowering as we see an increase in high heat events earlier in the growing season . The dominant hypotheses indicate that budburst may be less correlated with growing season temperature changes because it is more impacted by viticultural techniques and therefore sensitive to chilling time over dormancy . In the UC Davis ampelography vineyard, all vines are experiencing the same dormancy conditions, so the difference within years in timing of budbreak is explained by the varietal differences . However, across years, the lower sensitivity of budburst timing compared to flowering and veraison may be also be explained in part by the discrepancy in the dominance of climatic versus genetic controls for vegetative versus reproductive growth, respectively . Varieties may not be sensitive to temperature in the same way across stages, as the vine switches from vegetative growth to reproductive growth with the onset of flowering . The weaker relationship between budburst and cumulative temperature than the subsequent stages may be because flowering time and maturation are more strongly controlled by genetics . Furthermore, the dissociation between vegetative and reproductive growth makes it unclear how plants will adapt to climate change . While research shows viticulture is expanding to new territories all over the world , a crucial aspect to the success of the expanding viticulture into novel territories is matching the phenology to the local climate; agriculture will fail when introduced crops cannot adjust to new seasons . Climate change will not only change the varieties suitable for a region, but also the regions suitable for planting grapes . Failure to choose appropriate varieties for novel territories can impact natural ecosystems, an unintended adverse effect of expanding viticulture . A recent study modeling changes in viticulture territories under climate change scenarios predicted that 51 % of climatically suitable for growing winegrapes would become unsuitable . The intraspecific variation in heat thresholds for grapevines impacts the adaptation capacity of each cultivar . Previous authors suggest allowing cultivar turnover to prevent these major losses, which will depend heavily on what governments allow in Europe , while we are free to plant many different varieties in California. Among many strategies of adaptation to climate change, shifting to climatically more appropriate varieties has been widely suggested . Even with our current understanding of varieties’ climate niches, only a few existing cultivars are late ripening enough to avoid the warming predicted to occur during maturation in future climate scenarios . We identified many late ripening varieties that can be tested in future studies for suitability in California . International projects such as ADVIDCLIM are currently testing phenological models of grapevine with the expectation that varieties planted will need to change in future climate conditions . Hypothetical crosses between very late ripening varieties were modelled and still struggle to be late-ripening enough to endure the predicted 23-day shift and increase of 7°C expected by the end of this century . Within existing varieties, clonal variation does not offer a wide enough plasticity for adapting to climate change, however, taking advantage of existing varieties in warm regions to grow as alternatives is a promising strategy .