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

Discussion

To support the objective of this study, we propose a workflow to quantify the effect of heat extremes (i.e., heatwaves) at the commercial scale by linking long-term weather observations (1981 to 2023) to historical records of grape yield and fruit composition of Cabernet Sauvignon at harvest. For the first time, heatwave effects on key grape production metrics were determined, based on the integration of large phenological, weather, and grape production data sets (Figure 1).

Summary features extracted from weather series (Supplemental Table 1) were used to group similar growing seasons via clustering analysis. In previous research, changes in given parameters were assessed over time to identify positive or negative effects over the period investigated (e.g., Gambetta and Kurtural 2021, Faralli et al. 2024). To address the residual year-to-year variability observed in the historical data analyzed in these studies, our approach was to classify years according to their weather patterns. Of the 220 weather-based features created, several were aimed at quantifying levels of heat, such as the number of heat days and heatwaves (base temperature of 38°C); PRISM weather observations were used. Others have addressed the benefits and limitations of spatially interpolated data (Schneider and Ford 2013, Van Wart et al. 2013, Mourtzinis et al. 2017). A certain degree of inaccuracy was accepted in this analysis (see comparison with local weather station in Supplemental Figure 4), as PRISM offered historical weather series deemed suitable to characterize year-to-year differences. HCA results (Figure 3) discriminated three groups of seasons with distinct heat intensity and temporal distribution. Variables with the largest contribution to this separation were directly related to heat indices, defined by phenological stage or time of year (Figure 4B). C3 was dominated by the lowest values for all heat-related features, with T_max well below the critical threshold of 38°C and zero heat days and heatwaves per season (Figure 5). Seasons of the remaining two clusters displayed much higher heat levels than C3, although they occurred at different times of the growing season. This indicates that the clustering algorithm was able to discriminate seasons according to both the intensity and timing of heat events. Cluster 1 described seasons with the most extreme heat (Figure 5), namely, the highest T_max and number of heat days and heatwaves, and the lowest rainfall levels (Supplemental Figure 13). In C1, heat extremes were predominantly distributed in the postveraison period (VER to H; Figure 5A to 5C). Levels of heat in C2 were still higher than the critical threshold, but heat spikes occurred preveraison (FL to VER). The robustness of this clustering solution was validated using the RF algorithm (Figure 4). Even under strict conditions such as bootstrapping and dividing the data set 50:50 for training and testing, the clustering was extremely stable, with an average error rate below 2%. Correct cluster classification was highly dependent on heat-related features, especially those calculated using phenological stages (Figure 4B). We observed highly similar trends for heat-related variables when expressed using chronological and phenological intervals (Figure 5A). These results are site- and variety-dependent, and the alignment between aggregated months and phenology may disappear or require recalibration when a similar approach is applied elsewhere. Given the link between heatwave timing and production performance of Cabernet Sauvignon reported in our analysis, phenological data should be preferred when available to align heat events with specific stages of grape reproductive development.

The objective of this study was to understand how heatwaves, defined as events where heat intensity and duration are well above plant functional and productive optimums, affect vineyard production parameters, namely harvest date, vineyard yield, and fruit composition. This was achieved through a process of attribution, summarizing each response variable by cluster, representing season types (Figure 2). Data were summarized for individual vineyards then all observations were pooled into large distribution curves for cluster comparisons (Figures 6 and 7). Given the large number of historical records, heatwave effects were visualized as clear shifts in the distribution of harvest date, yield, and several fruit quality markers. Harvest date was anticipated in seasons with both PRE-V (C2) and POST-V (C1), on average 17 and 13 days, respectively (Figure 6A). Earlier harvest under warmer conditions is explained by advancing grapevine phenology from higher temperatures (Webb et al. 2007, Fraga et al. 2016, Ramos et al. 2018). Additionally, heat stress and water deficit are known to accelerate berry dehydration (Bonada et al. 2013, Xiao et al. 2018), which increases sugar concentration. Earlier harvest dates in C1 may also reflect attempts to avoid late-season heatwaves (Figure 5C) by harvesting early, especially in highly susceptible vineyards or to attain specific wine styles. Another effect observed in heatwave seasons (C1 and C2) was a wider harvest window compared to cool seasons of the C3 cluster, as evident in more variable distributions of harvest date points (C1 and C2; Figure 6A). This may be due to vineyard blocks displaying different sensitivities to heat, or specific management practices applied to reduce heat effects on grapevines, which were not included or considered in this analysis (Valentini et al. 2021, 2024, Marigliano et al. 2022, Previtali et al. 2023, Wilson et al. 2024).

Although yield differences between heatwave years (C1, C2) and cool seasons (C3) (Figure 6C) were not as striking as harvest date trends, these results were still highly significant and of value, considering that they were drawn from commercial data. A larger variation in commercial data over that from conventional research trials is expected, as well-defined experimental conditions are used to evaluate heat effects in conventional trials. However, the data approach used in this study relies on a much larger sample size than in research-scale trials, improving the confidence of estimated effects. Notably, negative shifts in yield distributions of C1 and C2 (indicating average yield losses of 30% under PRE-V and 22% with late-season heatwaves) confirm that commercially implemented mitigative actions are often overridden by heat extremes such as those referenced above. These values are within the range predicted in heatwave simulations, where yield drops up to 35% have been modeled in response to a single heatwave per season (5 to 9 days, different time of the year depending on the region) in European wine regions (Fraga et al. 2020). These results also support previous evidence that grapes are more susceptible to heat preveraison, here quantified as an additional average yield drop of ~10%. These findings of increased grape susceptibility to PRE-V are in general agreement with a prior study which described transient effects on Shiraz berry weight after 3-day heatwaves postflowering (Gouot et al. 2019). In a different study, 4-day heatwaves were simulated at different stages of growth on potted Sauvignon blanc (Greer and Weston 2010), and yield was reduced the most at fruit set (−62%) and mid-ripening (−53%), and less at veraison (−31%). Effects of elevated temperature on cell division and expansion processes during the first stage of berry growth are well documented (Kliewer 1977). Impediments to berry growth may be mediated by water deficit prior to veraison (Ojeda et al. 2001, Keller et al. 2008), or the effects could be combined, depending on the exact timing of the heat with respect to phenological development in each year. Other factors could be included in this analysis to further account for within-cluster variability. For example, we repeated the same analysis, splitting vineyards by age (Supplemental Figure 15), and found that yield of young vines was equally affected in C1 and C2 compared to C3, while in mature vines C2 displayed a larger yield reduction compared to C3. Similarly, application of cultural practices (e.g., canopy management, fruit thinning, irrigation, floor management) to align vine response to production targets could improve this analysis, though such practices are often not accessible and difficult to model quantitatively.

The same analysis presented for yield and harvest date was repeated on grape composition data that were obtained by analyzing key grape quality markers in commercial samples prior to harvest for over a decade. The grape analytes were selected due to high correlations with wine quality scores in Cabernet Sauvignon (Cleary et al. 2015). Heat patterns profoundly affected grape composition, as shown by the large number of analytes that were significantly different between clusters (Figure 7). Pre- and postveraison heatwaves differentially affected the aroma profile of grapes, but the effect was negative in both cases. POST-V (C1) events increased the off-flavor 1-octen-3-ol and decreased fruity ß-damascenone but also reduced the concentration of IBMP, which results in undesirable green aromas in wine. PRE-V (C2) did not change 1-octen-3-ol compared to Cool, while ß-damascenone declined and IBMP levels increased, moving the green aroma-to-fruit aroma ratio negatively toward green aromas. The decrease of ß-damascenone in heated grapes was not reported before, however, changes in the aroma profile from “fresh” to “cooked” fruit have been reported in wines made from heat-treated vines (Bonada et al. 2013, 2015). This change in aroma profile may be explained by norisoprenoid reductions and the downregulation of genes that are involved in the biosynthesis of terpenes and norisoprenoids, as previously described in heat-treated Cabernet Sauvignon (Lecourieux et al. 2017). The observed reduction of IBMP in seasons with POST-V (C1) is in agreement with prior findings in which IBMP decreased at bunch closure in Cabernet Sauvignon under higher temperatures (+1.5°C on average) that were obtained using open-top greenhouses, while differences disappeared at harvest (Wu et al. 2019). IBMP generally responds negatively to heat and sunlight exposure (Falcão et al. 2007, Koch et al. 2010) and heat treatments repress the transcription of the VviOMT3 gene, a process that is essential for the biosynthesis of IBMP (Lecourieux et al. 2017). The increase observed for seasons with PRE-V (C2) may represent a negligible effect (1 ng/L), especially considering the low perception threshold (7 ng/L; Sidhu et al. 2015).

Across several premium Cabernet Sauvignon vineyards, the concentration of grape phenolic compounds at harvest was greatly affected in heatwave years and the timing of heat affected specific phenolic measures (Figure 7). Our data approach returned significant increases in anthocyanin concentrations during early heatwave years (C2), as well as significant decreases during late heatwave years (C1). Consistent with our results, preveraison heatwaves did not affect anthocyanins (Gouot et al. 2019); the expression of anthocyanin-related genes was not affected when green berries were submitted to heat, but it was reduced with heat after veraison (Lecourieux et al. 2017). When temperature was increased throughout the growing season using open-top chambers (Sadras and Moran 2012, Wu et al. 2019), anthocyanins were also found to be negatively affected as the fruit was exposed to heat until maturity.

For grape skin tannins, the present analysis returned positive effects of pre- (C2) and postveraison (C1) heatwaves on polymeric tannins, positively affecting wine quality (Cleary et al. 2015), and increases in quercetin glycosides resulted from preveraison heatwaves only (C2). In an earlier study investigating the effect of early heatwaves on grape skin tannins, it was found that the effect of the heat events (with respect to tannin biosynthesis dynamics) depended on the timing of heat (Gouot et al. 2019). The differential response to heat based on phenological stages in this study is supported by other work that showed that the modulation of genes in biosynthesis of phenolic compounds varies depending on the time of the heat treatment (Lecourieux et al. 2017). In contrast, Wu et al. (2019) reported no changes in tannin profile from increasing the temperature by 1.5°C. Overall, the findings from the present study demonstrate that elevated temperature conditions enhance grape tannin concentration, likely driven by increased biosynthesis and potentially amplified by dehydration effects.

Grape basic chemistry at harvest seemed to be solely affected by postveraison heatwaves (C1), resulting in higher TSS, pH, and malic acid (Figure 7). One likely reason for this finding is that after fruit is exposed to heat preveraison (C2), grape basic chemistry may still be improved in the remaining part of the season (Lecourieux et al. 2017). According to the cluster characterization (Figure 5), weather patterns for C2 resembled C3 more than C1 during the ripening stage, possibly indicating conditions that are close to optimum for sugar accumulation and the retention of malic acid. Such compensation cannot occur when heatwaves occur postveraison (C1), particularly when they are very close to harvest. The higher TSS levels observed for C1 could be attributed to the promotion of photosynthetic activity under moderate temperature increases (Greer and Weedon 2012), or to the concentration effects driven by berry dehydration under heat and water stress (Xiao et al. 2018). According to berry moisture data, no significant dehydration effect was observed between clusters (Figure 7). The absence of differences could be attributed to data variability and the fact that large changes in berry weight cause very little shifts in berry moisture (15 to 20% change in berry weight leads to 1 to 2% change in berry moisture [data from another study and not shown]). In addition, some level of berry shriveling may be desired to achieve specific wine style targets (Antalick et al. 2021). In the absence of berry weight data (which were not available for this analysis), it is not possible to address whether higher TSS resulted from enhanced active sugar accumulation or from higher berry dehydration. One trend that supports the hypothesis of berry dehydration is the increased level of malic acid under POST-V (C1) compared to PRE-V (C2) and Cool (C3). Malic acid is synthesized in the fruit preveraison and degraded during the ripening stage via respiration, a process that is exacerbated by higher temperatures (Sweetman et al. 2014, Rienth et al. 2016).

In summary, this study presents robust evidence that the commercial production of Cabernet Sauvignon is largely affected by heatwaves. By analyzing data from a large pool of vineyards over multiple decades, it was revealed that both pre- and postveraison heatwaves substantially reduced yield and anticipated harvest and altered the concentration of key quality markers at harvest. Further research is necessary to extend this work to other growing regions and varieties, which may lead to identification of genotypes that are less susceptible to heatwaves, or to identification of improved genotype-environment combinations that are better suited to warmer and drier conditions.