Sugar accumulation modelling was performed using previously described 3-parameter logistic function ( de Rességuier et al., 2024; Suter et al., 2021) [Equation 1]:

(Equation 1) S D O Y = S m a x 1 + 0.005 . e ( - 4 . k s u g a r . D O Y - D O Y 95 % s u g a r S m a x )

where Smax is the estimated maximum sugar concentration (°Brix), DOY is the day of the year, DOY95% _sugar is the DOY when 95 % of the maximum sugar has accumulated, and k _sugar is the estimated maximum rate of accumulation (°Brix per day). The sugar concentration (Brix) at DOY _95% (Plateau 95 %), and the number of days between DOY _95% and veraison (Dur _95%-Ver) were also calculated and used as agronomic indicators of grape ripening.

Total acidity and malic acid generally decrease rapidly after veraison before approaching a minimum residual concentration at maturity. An exponential decay model with an asymptote was therefore selected to capture this behaviour (Equation 2 and Equation 3):

(Equation 2) T A = a . e k T A . D O Y c e n t e r e d + T A a s y m p

(Equation 3) M a l _ a c = a . e k M a l _ a c . D O Y c e n t e r e d + M a l _ a c a s y m p

where a, k, and asymptotic concentration (TA _asymp and Mal_ac _asymp) are the parameters estimated for each variable, total acidity (g of tartaric acid/L) and malic acid concentration (g/L), by nonlinear regression (nls function in R). To improve the numerical stability of the exponential model fit, the day of the year (DOY) was centred on the first measurement point for each year-cultivar combination, so that DOYcentred = 0 corresponds to the start of the monitoring period for each Grape Variety/Location/Year. Parameters k _TA and k _{Mal_ac} represent the speed at which total acidity and malic acid concentration, respectively, decrease during berry ripening, and are expressed in units of inverse days (day⁻¹). TA _asymp and Mal_ac _asymp represent the minimum asymptotic value of total acidity or malic acid concentration. DOY95% _TA was calculated and corresponds to the day of the year when total acidity reaches 95 % of its final asymptotic value, estimated by the exponential model. It thus reflects the total acidity at the end of maturation.

Preliminary inspection of the data indicated approximately linear trends over the monitored ripening period. Consequently, linear models were used to describe the evolution of tartaric acid concentration and pH (Equation 4 and Equation 5):

T a r t _ a c = k T a r t a c ⋅ D O Y + b T a r t _ a c (Equation 4)

p H = k p H . D O Y + b p H (Equation 5)

where the slopes k _{Tart_ac} and k _pH and the y-intercepts (estimated initial values) b _{Tart_ac} and b _pH were estimated by linear regression (function lm in R). The slopes were used as agronomic indicators.

Anthocyanin accumulation is characterised by an increase during ripening followed by stabilisation or decline near harvest, resulting in a bell-shaped temporal pattern. A Gaussian model was therefore used to estimate the timing and magnitude of peak anthocyanin accumulation and was fitted separately for each domain–vintage combination. The model adopted follows the form (Equation 6):

(Equation 6) A n t h o c y a n i n =   M a x a n t h o . e ( - D O Y - D O Y p e a k 2 2 σ 2 )

where Max _antho represents the maximum amplitude of anthocyanins, DOY _peak the day of the year (DOY) corresponding to the peak accumulation, and σ the peak width, reflecting the rate of accumulation and degradation. The parameters were estimated by nonlinear regression (nls function in R). The number of days between DOY _peak and mid-veraison date (Dur _Peak-Ver) was also calculated and used as agronomic indicators.

The adequacy of the models was assessed using the coefficient of determination (R ²), the Pearson correlation coefficient ( r) and the root mean square error (RMSE). R ² quantifies the proportion of the variance in the response variable explained by the model, Pearson r reflects the strength of the linear association between observed and predicted values, and RMSE measures the average prediction error, calculated as the square root of the mean of the squares of the residuals (differences between observed and predicted values).

The influence of grape variety, location and vintage on agronomic indicators (DOY95% _sugar, Plateau95%, Dur _95%-Ver, k _sugar, k _TA, TA _asymp, DOY95% _TA, k _{Mal_ac}, Mal_ac _asymp, DOY95% _{Mal_ac}, k _{Tart_ac} and k _pH) was studied using a mixed linear model (lmer function from the lme4 package; Bates et al., 2015). Grape variety and location were treated as fixed effects to quantify their average effects on the response variable, while vintage was included as a random effect on the intercept in order to model the interannual variability not captured by the fixed effects. To assess model performance, we calculated both the marginal R ² (variance explained by fixed effects only) and the conditional R ² (variance explained by both fixed and random effects), following the method of Nakagawa and Schielzeth (2013). This distinction allows quantifying the specific contribution of climatic variability (vintage) to the total variance. Interactions between fixed effects (grape variety × location) were also taken into account. When the linear mixed-effects model including vintage as a random effect failed to converge, typically due to a limited number of observations per vintage or low interannual variability, a classical linear model including grape variety, location, and their interaction as fixed effects was fitted instead. For the anthocyanin, due to the low number of replicates per group, only simple linear models (lm in R) were used to assess the effect of location and vintage on each anthocyanin parameter (Max _antho, DOY _peak, σ, and Dur _Peak-Ver) separately.

For each parameter, the model coefficients, their standard errors, F-values, and p-values were calculated to assess the statistical significance of each predictor. All model coefficients were estimated relative to the reference group, Pinot noir, harvested in Leytron during the 2019 vintage.

An analysis of variance (ANOVA) was then performed on the adjusted model to test the overall significance of the fixed effects ( car package; Fox & Weisberg, 2019). The normality of the residuals was assessed using the Shapiro-Wilk test, and homoscedasticity was visually verified from the graph of the residuals as a function of the fitted values. In cases where the normality of the residuals was not met, a non-parametric approach was adopted: a Kruskal–Wallis test was applied to assess the overall differences between the levels of the fixed factors, followed by a Dunn test for pairwise comparisons, with Bonferroni correction for multiple comparisons, using the rstatix package (Kassambara, 2025).

The modelling of total acidity was performed using an exponential decay model with an asymptote, and curves are shown in Figure 2A. Analysis of k _TA values, the speed at which total acidity decreased during berry ripening, revealed a significant grape variety effect ( p < 0.01) (Table 2). The positive coefficient for Divico (+ 0.03) indicated that the decrease in acidity was significantly smaller for this grape variety than for Pinot noir. This can largely be explained by the initial lower TA of Divico (Figure 2A). Although the interaction between grape variety and location was not statistically significant at the 5 % level ( p = 0.097), post-hoc comparisons suggested that the variety effect was larger in Pully (–1.346; p < 0.01) than in Leytron (–0.571; p = 0.516) (Table S1). Moreover, for the asymptotic total acidity value (TA _asymp), a significant effect of grape variety was observed, with higher values for Divico than for Pinot noir at the end of ripening (+ 0.57 g/L, p < 0.01). The location effect and the grape variety × location interactions did not reach the significance threshold. The fixed effects explained 31.4 % of the variability of k _TA and 46.40 % of the variability of TA _asymp. Upon inclusion of the random vintage effect, the explained variance increased to over 71 %, demonstrating the significant impact of vintage on acid metabolism. This finding aligns with the sugar accumulation dynamics previously described, where specific climatic signatures dictated ripening precocity. The warm spring followed by a dry summer in 2020, and the combined heat and low precipitation in 2022, which drove the earliest sugar accumulation patterns, similarly precipitated an earlier onset of acid degradation. Consequently, the decline in total acidity for both 2020 and 2022 was temporally shifted earlier in the season compared to the cooler, wetter 2021 vintage and the 2019 season, which, despite a warm summer, experienced a cooler spring. The analysis of DOY95% _TA showed no effect of either grape variety or location (Table 3).

Malic acid evolution (Figure 2B) followed the expected catabolic trend, driven by reduced synthesis and accelerated respiration during ripening, a process known to intensify under hot conditions ( Bindon et al., 2014; Conde et al., 2007). Statistical analysis of asymptotic values (Mal_ac _asymp) revealed a significant effect of grape variety ( p < 0.001) and grape variety × location interaction ( p < 0.05), demonstrating that the effect of grape variety depended on location (Table 2). Post-hoc comparisons showed that this difference was significant only in Pully ( p < 0.01), where Divico had 1.04 g/L less malic acid than Pinot noir, while no significant difference was observed in Leytron ( p = 0.707) (Table S2). With regard to the rate of decrease in malic acid, k _{mal_ac}, the model revealed a trend towards a significant effect of grape variety ( p = 0.065), while the effect of location and the interaction between grape variety and location were not significant. Divico tended to show a slower decrease rate in malic acid during ripening, likely a consequence of its lower initial concentration (Figure 2B). Moreover, as for total acidity, the grape variety and the location did not significantly affect the DOY95% _{Mal_ac} (Table 3). Again, according to Figure 2B, the vintage effect can be seen as malic acid decreased in 2020 and 2022, both hot vintages, earlier in the season than in 2019 and 2021.

The study of tartaric acid evolution, particularly the rate of decay, k _{Tart_ac}, showed a significant influence of grape variety, with higher coefficient values for Divico (+0.05, p < 0.01), indicating a slower decay of tartaric acid than for Pinot noir (Table 2 and Figure 3A). The location and the grape variety × location interaction did not show a significant influence on k _{Tart_ac}. The inclusion of the vintage effect drastically improved the model fit, raising the explained variance from 39.5 % (fixed effects only) to 87.3 %. This demonstrates the impact of the vintage on the rate of tartaric acid reduction. Rainfall patterns and vine water stress levels could also play a key role in annual climatic fluctuations, as higher water stress induced higher tartaric acid decrease ( Rösti et al., 2018).

With regard to pH, the slope value, k _pH, which indicates the rate of pH increase, was also significantly affected by the grape variety (Table 2 and Figure 3B). As expected, given the total acidity results, the coefficient for Divico was lower than that for Pinot noir, indicating a slower increase in pH in the berries during ripening (–0.01, p < 0.001). The vintage effect could not be included in the full model due to a lack of convergence; a simple linear model was retained, which explained 74.9 % of the variance, suggesting that the model was well-fitted to the available data, although other unaccounted variables may contribute to explaining the residual variability. Interestingly, at the end of maturation, the pH values for Divico were similar for all considered vintages, between 3.2 and 3.4 (Figure 3B), whereas Pinot noir showed higher final pH and greater variability, with values ranging from 3.3 to 3.6.

During the maturation of Divico berries, the malic acid concentration stabilised between 1.5 g/L and 2 g/L at the end of the ripening phase. This condition, together with the higher tartaric acid concentrations, particularly in the berries from Pully, ensures balanced total acidity at the end of the ripening period. This was also reflected in the pH values of the berries. At the end of ripening, the pH was relatively low and similar across all vintages (Figure 3B). This was not the case for Pinot noir, particularly with the berries from Pully, whose pH values rose sharply during a warm vintage, reaching a value over 3.5 (2020 and 2022). Rising temperatures as a result of global variations in climatic conditions are leading to an increase in pH levels in certain grape varieties and the related wines (van Leeuwen & Darriet, 2016). As wine pH is a critical factor in determining its microbiological and physicochemical stability, this trend is of particular concern. Indeed, pH exerts a significant influence on the selection of microorganisms and specific chemical reactions, in particular those impacting the efficiency of sulfur dioxide ( Comuzzo & Battistutta, 2019). Consequently, the slower increase in pH in the grape berries of Divico, irrespective of the conditions present during vintage, appears to be a beneficial aspect of this variety for wine production in a changing climate ( Plantevin et al., 2024).

To position Divico within the framework proposed by Plantevin et al. (2024), αlogmal and αTar coefficients were calculated from our dataset following their methodology (Figures S2 and S3). The slope of the relationship between the tartaric acid content (µmol/berry) as a function of growing degree days (summation of mean daily temperature starting from January 1 ^st) corresponds to the αTar coefficient. More negative αTar values indicate a greater rate of tartaric acid decrease during grape ripening. The αTar values obtained in our study were within the range reported by Plantevin et al. (2024), who reported a value of approximately −0.025 for Pinot noir. In our dataset, αTar values for Pinot noir were −0.0314 in Leytron and −0.0408 in Pully, while Divico exhibited mean αTar values of −0.0325 and −0.0294 in Leytron and Pully, respectively. These results indicate that Divico exhibits tartaric acid degradation rates comparable to those observed in Pinot noir. Although the distributions largely overlapped, Divico tended to show slightly less negative αTar values and lower variability, suggesting a relatively stable tartaric acid pool during ripening. This might explain why a significant effect of the grape variety on tartaric acid evolution was found. The slope coefficient of the relationship between ln(malic acid content (µmol berry⁻¹)) and growing degree days corresponds to the αlogmal coefficient. More negative αlogmal values indicate greater sensitivity to malic acid degradation for a given heat summation. In the study of Plantevin et al. (2024), αlogmal values for Pinot noir ranged from −0.0022 to −0.0007. In our study, αlogmal values for Pinot noir were −0.0023 in Leytron and −0.00215 in Pully. In contrast, Divico exhibited αlogmal values of −0.00105 and −0.00115 in Leytron and Pully, respectively. These less negative values indicate a slower rate of malic acid degradation and therefore a greater capacity to maintain acidity during ripening. Furthermore, the αlogmal values obtained for Divico fall within the range reported by Plantevin et al. (2024) for cultivars exhibiting relatively slow malic acid degradation. Although direct quantitative comparisons should be interpreted cautiously because the coefficients were derived from different vineyards and environmental conditions, these results suggest that Divico may possess physiological characteristics associated with improved acidity retention under warm ripening conditions.

The TPI results are presented in Figure 4A. The TPI did not change much as the Divico berries ripened. Means across the maturation period were calculated, and the influence of growing location and vintage was analysed. Although the model explained 63.6 % of the variance of the TPI variable, no significant effect was found for the location and vintage effects (Table 4).

The evolution of anthocyanins was modelled to calculate several parameters, including Max _antho, the peak date (DOY _peak), the sigma parameter (σ) and the duration between veraison and the anthocyanin peak (Dur _peak-Ver) (Figure 4B). Among these parameters, only DOY _peak was significantly influenced by the vintage ( p < 0.05), while the location showed no significant effect (Table 4). The linear model estimates for DOY _peak revealed that, in comparison to the reference vintage (2019), the anthocyanin peak occurred 19 days earlier in 2020 (-19, p = 0.021) and 22 days earlier in 2022 (-22, p = 0.016) (Table S3). Under the experimental conditions, the average DOY _peak is 265 days (± 2 days), which is approximately 22 September.

While direct experimental comparison was not conducted in this study, contextualising these findings with existing literature highlights the distinct profile of Divico. De Montmollin and Dupraz (2003) reported anthocyanin concentrations of approximately 900 mg/kg for Pinot noir and 2100 mg/kg for Gamaret. In contrast, Divico exhibits remarkably elevated content (> 3000 mg/kg).

In the context of traditional grape varieties, total berry colour is widely regarded as a pivotal indicator of quality. Consequently, any approach that enhances the accumulation of anthocyanins in berries stands to benefit grape growers and winemakers ( Conde et al., 2007). In the case of Divico berries, this abundance of anthocyanins contributes to the production of wines with intense colour. The climate plays a role in anthocyanin accumulation. Warmer and drier vintages induced earlier accumulation of anthocyanins in the Divico berries sampled in 2020 and 2022, especially in Pully. One of the most interesting results of this study is the anthocyanin accumulation peak observed during ripening, which could serve as an indicator of the optimal phenolic maturity of grapes in the future, as was previously shown for Gamaret (De Montmollin & Dupraz, 2003). The high anthocyanin content observed in Divico berries suggests that maximising anthocyanin accumulation may not be a necessary goal for this variety. However, achieving both sufficient tannin quantity and appropriate quality remains essential to ensure the visual and sensory balance of Divico wines. Consequently, further investigation into tannin quality and the influence of harvest timing on the sensory properties of Divico wines is warranted.