Long-term Weather Observations Reveal the Impact of Heatwaves on the Yield and Fruit Composition of Cabernet Sauvignon

Supplemental Table 1  List of summary features extracted from PRISM long-term weather observations. GDD, growing degree day; T_max, maximum temperature; HWU, heatwave units; DOY, day of year; T_mean, mean temperature; T_min, minimum temperature; VPD, vapor pressure deficit.

Supplemental Table 2   Statistical outputs of harvest date, yield, and fruit composition parameters by cluster (C1, cluster 1, postveraison heat; C2, cluster 2, preveraison heat; C3, cluster 3, cool season). P values of cluster effects were computed using linear mixed models (fixed effect: cluster; random effect: vineyard block ID). Data by cluster are presented as means ± 95% confidence intervals and different letters denote significant differences between means according to Tukey’s adjusted post-hoc test at p ≤ 0.05. Significant p values (i.e., ≤0.05) are marked in bold formatting. DOY, day of year; TSS, total soluble solids; YAN, yeast assimilable nitrogen; IBMP, 3-isobutyl-2-methoxypyrazine.

Supplemental Figure 1  Map of vineyard site (S1 to S5) locations and closest American Viticultural Areas (AVA).

Supplemental Figure 2  Relationships between observed phenological stages (BB, budbreak; FL, flowering; VER, veraison; H, harvest) and cumulative growing degree day (GDD) calculated from A) 1 January or B) 1 March. Cumulative GDD were calculated using PRISM daily data. 

Supplemental Figure 3  Heatmap of scaled historical weather data submitted to clustering analysis. Columns represent combinations of a vineyard site (S1 to S5) and season (1981 to 2023), rows show normalized values (mean = 0, standard deviation = 1) for each weather feature extracted from historical series. Dendrograms group rows and columns using hierarchical clustering analysis. Gray boxes represent categories of features (generic weather and heat-related) and specify whether growth stages were based on calendar days or phenological stages.  

Supplemental Figure 4  Comparison between daily temperature values from PRISM (gridded climate data) and a local weather station at one of the study sites (April to October): average (T_avg; A and B), maximum (T_max; C and D), and minimum temperature (T_min; E and F). In A, C, and D, lines connect daily temperature values and colors differentiate between data sources. B, D, and F show daily differences between PRISM and local station data.

Supplemental Figure 5  Hierarchical clustering analysis outputs by varying number of clusters from 2 to 10 clusters. The three-cluster solution was selected as the optimal number of clusters for further analysis. 

Supplemental Figure 6   Hierarchical clustering analysis (HCA) optimization with clustering performance metrics. Three different indexes were assessed in response to varying number of clusters from 1 to 10: A) within sum of squares (WSS); B) silhouette width; C) gap statistic.  

Supplemental Figure 7  Cluster validation using iterative bootstrap random forest classification with three clusters. Each panel represents the test error rate (TER) for a given number of iterations, from 50 to 500. Green points represent TER values for each iteration. The red dashed line represents the average TER across all iterations, with the value annotated in red at the top of each graph.

Supplemental Figure 8  Growing degree day (GDD) features summarized by weather cluster. A) Distribution of total GDD (April to October) by cluster; B) monthly distribution of GDD averages by cluster; C) GDD by cluster early and late in the growing season; and D) monthly distribution of GDD values by cluster. Colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison heat; Cluster 2, preveraison heat; Cluster 3, cool seasons. When present, points and error bars represent means ± standard error by cluster. 

Supplemental Figure 9  Heatwave unit (HWU) features summarized by weather cluster. HWU was calculated as the cumulative deviation of T_max -38°C, considering positive deviations only. A) Distribution of total HWU (April to October) by cluster; B) monthly distribution of HWU by cluster; C) HWU by cluster early (May to July) and late (August to October) in the growing season; D) summary of HWU by phenological stage (BB, budbreak; FL, flowering; VER, veraison; H, harvest); and E) monthly distribution of HWU values by cluster from May to October. Colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison; Cluster 2, preveraison heat; Cluster 3, cool seasons. When present, points and error bars represent means ± standard error (SE) by cluster. Ribbons represent SE values in line charts.

Supplemental Figure 10  Average temperature (T_avg) features summarized by weather cluster. A) Distribution of seasonal Tavg values by cluster (April to October); B) monthly distribution of T_avg values by cluster; C) T_avg values by cluster early (May to July) and late (August to October) in the growing season; and D) monthly distribution of T_avg values by cluster. Colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison heat; Cluster 2, preveraison heat; Cluster 3, cool seasons. When present, points and error bars represent means ± standard error (SE) by cluster. Ribbons represent SE values in line charts. 

Supplemental Figure 11  Minimum temperature (T_min) features summarized by weather cluster. A) Distribution of seasonal T_min values by cluster (April to Oct); B) monthly distribution of T_min values by cluster; C) T_min values by cluster before the growing season (January to April) and early (May to July) and late (August to October) in the growing season; and D) monthly distribution of T_min values by cluster. Colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison heat; Cluster 2, preveraison heat; Cluster 3, cool seasons. When present, points and error bars represent means ± standard error (SE) by cluster. Ribbons represent SE values in line charts. 

Supplemental Figure 12  Changes in temperature by cluster according to phenological intervals (BB, budbreak; FL, flowering; VER, veraison; H, harvest). A) Minimum temperature (T_min), B) average temperature (T_avg), and C) maximum temperature (T_max). Points and ribbons represent means ± standard error. Colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison heat; Cluster 2, preveraison heat; Cluster 3, cool seasons.

Supplemental Figure 13  Rainfall features summarized by weather cluster. A) Distribution of seasonal rainfall by cluster (April to October); B) monthly distribution of rainfall by cluster; C) rainfall by cluster before the growing season (January to April) and early (May to July) and late (August to October) in the growing season; and D) rainfall by cluster according to phenological intervals (BB, budbreak; FL, flowering; VER, veraison; H, harvest). Colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison heat; Cluster 2, preveraison heat; Cluster 3, cool seasons. When present, points and error bars represent means ± standard error (SE) by cluster. Ribbons represent SE values in line charts. 

Supplemental Figure 14   Exploratory analysis of harvest date (A) and yield (C) and cluster classification of seasons at the five sites (S1 to S5). In A and C, points represent data available at each site, with the number of observations by site annotated above the data distributions. In B, colors differentiate between clusters, indicating seasons with different heat levels: Cluster 1, postveraison heat; Cluster 2, preveraison heat; Cluster 3, cool seasons. DOY, day of year.

Supplemental Figure 15  Effect of weather clusters on Cabernet Sauvignon yield, according to vineyard age. The same analysis was repeated after classifying vineyard blocks as young (3 to 5 yr, left graph) or mature (>5 yr, right graph). Points represent average yield for individual blocks summarized by cluster. Colors differentiate between clusters: Cluster 1 (C1; POST-V, postveraison heat), Cluster 2 (C2; PRE-V, preveraison heat), and Cluster 3 (C3; Cool, cool season). Gray lines connect averages for the same vineyard block belonging to different clusters. Density plots represent distributions by cluster for each age group. Labels report cluster means and 95% confidence intervals (CI) by group, calculated using a linear mixed model where clusters were the fixed effect and vineyard blocks were the random effect on the model intercept. Different letters denote statistical differences between clusters for each age group according to Tukey’s adjusted multiple comparison post-hoc test with α ≤ 0.05.

Supplemental Figure 16   Exploratory analysis of grape analytes measured at five different sites. A to L represent single fruit composition parameters and points represent data available for each site (S1 to S5). The number of observations by site is annotated on top of data distributions in the first plot only. IBMP, 3-isobutyl-2-methoxypyrazine; TSS, total soluble solids; YAN, yeast assimilable nitrogen.

• Supplemental Data