This study was carried out using data from vineyards in the Cognac wine-growing region. The area covered by the Cognac Protected Designation of Origin (PDO) is located in western France along the Atlantic coast and covers more than 80,000 ha. Vineyards grow under a cold Köppen Cfb-type semi-continental climate with no dry season and an extended warm summer (Beck et al., 2018). The average annual rainfall is 860 mm. Prevailing winds are from the south-west and north-west. Around 98 % of the wine produced in this area is made from the Ugni Blanc grape variety, and the average planting density is 3,000 to 4,000 vines per hectare. Diversity in the rootstocks was observed (Fercal, Ruggieri 140, 333 EM, 41 B, 161 49 C, RSB 1, SO4, 1103 Paulsen, Gravesac). The vineyards are extremely vigorous and production targets are quite high, ranging from 120 to 150 hl/ha. The PDO area is characterised by five soil types varying in compaction risk, water retention capacity and mineral content (Table 1).

The data were collected cover a 30-year period (1988–2018). Each year, 55 plots were monitored. The plots were not all the same in all the years, as some plots were removed from the network, having been grubbed up, and were replaced in the database by another plot. In total, 107 plots were monitored. Each year in each plot, four subplots of five vines (or vine locations when the vines were dead or missing) were monitored. These locations remained the same throughout the plot lifetime. Data for the monitored plots and locations were pulled from two databases: a plot database and a subplot database.

The plot database initially contained 116 raw variables per plot: 36 (32 %) static descriptive variables describing the plot (e.g., ‘soil type’ or ‘plot location’) and 80 (68 %) dynamic variables, comprising one yield variable (referred to as ‘annual yield’ from this point on) and 79 variables describing the agricultural practices implemented on each plot (e.g., number of phytosanitary treatments). The dynamic variables were collected each year by extension agents alongside the winegrowers. The subplot database contained data on the yearly status of the 20 vine locations monitored in each plot, which was used to quantify mortality.

The raw database was cleaned and the vineyard plots and variables were sorted. The 107 plots were classified in the first part of the analysis according to decline intensity (see Step 2). Then, for the purposes of the cultural practices analysis (Steps 3 to 6), we selected plots based on the following two criteria: i) at least 10 years of information on the plot was available in the database, and ii) the period of time for which there were data fell within the time frame of when decline is generally observed (which has been shown to occur from 15 to 35 years after planting, Merot et al., 2023). Based on these criteria, 43 of the 107 plots (40 %) were selected for analysis.We then only kept the variables that had less than 50 % of their data missing and that had been monitored for more than 15 years. Variables related to dynamics, specifically those related to intercropping and rootstock, were modified to make them consistent or simpler. This was done when redundant information was reported or when the number of factors was too high for further statistical analysis compared to other nominal variables. We chose to simplify the rootstock variable by considering the rootstock attributes in terms of vigour, sensitivity to total limestone, moisture sensitivity, drought sensitivity and adaptation to the soil in the plot (source French Vine and Wine Institute). The rootstock’s adaptation to the plot’s soil was established by comparing soil type and the genetic soil-related characteristics. Finally, we created a variable representing annual mortality at plot scale for the 20 locations observed every year, which we refer to here as new_dead. This variable was calculated by comparing the status of each location in year n-1 with its status in year n. The new_dead variable for year n corresponds to the number of new dead vines in year n in the 20 locations.

The study thus focused on the ‘cleaned’ database containing the 62 remaining variables (59 %): 44 dynamic variables related to agricultural practices, 18 static variables describing the plots, 3 yield variables and 3 vine mortality variables (Supplementary data). Missing values accounted for less than 2 % of the data.

We considered two processes related to mortality in the plots: i) a high level of mortality observed for all years or during specific years, and ii) a situation where mortality increased over time. We characterised each of these processes using a static indicator built from the new_dead dynamic variable. The first indicator calculated for each plot was the average number of new dead vines per year (av_new_dead). Av_new_dead is the total new_dead in all 20 vine locations during the plot observation period divided by the number of years of observation. The second indicator, new_dead_slope, characterises changes in new dead vines over time.

We also calculated two indicators of yield dynamics that described i) the change in yield over time, and ii) yield loss, in the same way as for mortality.

Based on the annual_yield variable, expressed in hl/ha for each plot, we first calculated an indicator for the change in yield (yield_slope). Yield_slope was calculated applying the same approach used for new_dead_slope. For each plot, yield_slope corresponds to the slope of the linear regression of the annual_yield for the observation period. We then calculated the average yield ratio (av_yield_ratio, in %) for each year in each plot to determine potential yield losses. For this purpose, we determined a yield achievement ratio for each year/plot combination, which equals the ratio between the annual_yield and a chosen reference value (in this case, this was the annual average yield of the Cognac PDO) (Personal communication, Bureau National Interprofessionnel Cognac). The av_yield_ratio of a plot is the calculated mean of the yield achievement ratio over the observation period.

The indicators for changes in mortality and yield were arranged in a decision tree to characterise and categorise the plots according to their decline intensity (Figure 2). We divided the plots into three categories of decline: non-declining, moderately declining and strongly declining. Moderately declining plots are those that showed a decline in terms of the mortality indicator but not of yield. Based on the results of Merot et al. (2023), we first grouped plots together according to mortality indicators - specifically new_dead_slope, which is useful in characterising a progressive and easily observable decline phenomenon. By doing this, we distinguished plots characterised by a positive new_dead_slope from the others (Figure 2). Among plots with a positive new_dead_slope, we were able to clearly identify strongly declining plots with a negative yield_slope. This first group of strongly declining plots was important in our analysis, because it allowed us to set the ‘Ts threshold’ that we would later use in the other branches of the decision tree. The Ts threshold was calculated as the mean value of av_new_dead for the first group of strongly declining plots with clear decline characteristics (increasing mortality and decreasing yield over time), and was found to equal 0.47 new dead vines per year in the 20 vine locations. Plots with a positive new_dead_slope but no change in yield over time were categorised as moderately declining plots.

For the plots that still needed to be categorised, we considered that no change in new_dead (new_dead_slope = 0) was associated with decline phenomena when average yearly mortality (av_new_dead) reached a high level; i.e., when the value was greater than the Ts established previously. For plots with this characteristic, those with a negative yield_slope were considered as strongly declining plots. Those that did not show any change in yield (yield slope = 0) were considered as moderately declining plots.

Finally, plots characterised by a low level of mortality over time and a stable yield (yield_slope near 0) were considered to be in non-declining situations. The other combinations of yield and mortality characteristics were not considered in this study, because the decline phenomena were too ambiguous or not represented in the database.

Figure 2. Decision tree of the different plot categories.

The black downward arrow indicates a decrease in the variable over time (negative slope) and the black upward arrow indicates an increase in the variable over time (positive slope). The absence of a slope is indicated by the equal sign and “No slope”. Green boxes = non-declining plots; orange boxes = moderately declining plots; red boxes = strongly declining plots; grey boxes = situations with no plot in the dataset; and white boxes = plots with ambiguous situations of decline. Ts = average value of av_new_dead for the first group of strongly declining plots that showed clear decline characteristics (increasing mortality and decreasing yield over time).

To better highlight the critical periods of grapevine lifespan in terms of decline, we built a timeline of decline status as a function of plot age (defined as number of years after planting). This timeline showed i) the year young vines became fully productive vines, ii) the first year any dead vines were observed, iii) the year in which mortality dynamics began, mortality dynamics being defined as a three-year period with at least two years of new dead vines, and iv) the three consecutive years marking the potential start of a yield decrease, defined as a three-year period during which the annual yield was equal to or lower than the median yield of the plot during the observation period. These timelines were cross-referenced between plots for the three decline categories and qualitatively analysed.

Three categories of decline were identified in the studied plots: non-declining (44 plots), moderately declining (47 plots) and strongly declining (30 plots). These categories, the strongly declining plots (Figure 3A), non-declining plots (Figure 3B) and moderately declining plots (Figure 3C), showed clear differences in terms of yield and mortality rate. The starting point of the mortality increase and yield decrease was identified in most plots. When a mortality increase and a yield decrease were observed, we also noticed a time delay between the two processes: yield decrease occurred a few years after the mortality increase (2 to 5 years depending on the plot category).

In non-declining plots, both vine death and mortality increase tended to occur later than in other plot categories. The first vine death appeared at the average age of 24.6 years, and the mortality increase was observed after 26 years (4 years after the strongly declining plots; see Figures 4 and 5). The strongly declining and non-declining plots had considerably different mortality rates. In the non-declining plots, the av_new_dead value was nearly half that in strongly declining plots, with 0.28 new dead vine per year (Figure 5). The yield indicators of the non-declining and declining plots also tended to differ: as can be seen in Figure 3, yield decrease in the non-declining plots started 6 years later than in the strongly declining plots, at an average age of 29 years. The yield_ratio was 102.2 % for non-declining (Figure 5).

The yield indicators for the moderately declining plots were in between the strongly declining and non-declining plots, and the mortality indicators of the moderately declining plots were similar to strongly declining plots. The plot age at which the first new dead vine was observed in the moderately declining plots was not significantly different (p > 0.05) to the strongly declining plots, but it tended to be earlier than for the non-declining plots (Figure 5). The av_new_dead and new_dead_slope values for this category were also similar to those for the strongly declining plots (Figure 4), but significantly higher than for the non-declining plots. In terms of yield, the moderately declining and strongly declining plots differed (Figures 4 and 5). The yield_ratio value tended to be similar to that of the strongly declining plots but lower than that of the non-declining plots (Figure 5). The av_yield value was in between the strongly and non-declining plots (Figure 5). Finally, just as in the non-declining plots, a yield decrease (yield_slope close to 0) was not clearly observed in the moderately declining plot category (Figure 4).

When analysing decline, the fifteen agricultural practice variables fell into two distinct classes (C1 and C2) for each five-year age period. For each five-year age period, the 11 significant variables in the two classes are presented in Table 2. Regarding the 16–20-year period, C1_16-20 and C2_16-20 differed in terms of six variables: vegetation volume in the inter-row, amount of potash applied (in majority mineral applications), total number of all treatments against pests and diseases, volume of pesticide sprayed, number of treatments for powdery mildew and number for treatments for downy mildew. Plots in C2_16-20 were more often intercropped, with vegetation in the inter-row, and competition due to vegetation development was higher in the inter-row. The same results were observed under the row with nearly double the vegetation in C2_16-20 than in C1_16-20. Vegetation in class C2_16-20 plots was also trimmed less frequently (2.9 times compared to 3.6 times in the C1_16-20 plots); however, we did not find any significant differences with respect to observed vigour. Pre-pruning, which consists in roughly and mechanically cutting vine shoots before the manual winter pruning, was not carried out in the C2_16-20 plots. Potash application was higher in the C2_16-20 plots. Finally, in the C2_16-20 plots, the number of treatments for powdery and downy mildew was lower.

The findings from the 21–25-year period were similar to those from the 16–20-year period. We observed that the C2_21-25 plots were more often intercropped with vegetation in the inter-row, but the reported competition was higher in the C1_21-25 plots. Potash application was also higher in the C2_21-25 plots, as was the volume of pesticide sprayed. Conversely, the C1_21-25 plots showed a higher number of treatment applications against powdery and downy mildew, which is consistent with the 16–20-year period. With regard to sodium arsenite applications, C2_21-25 contained plots that were treated, while the C1_21-25 plots were not treated at all.

Finally, in the 26–30-year period, the two classes were differentiated by the quantities of potash applied (the C2_26-30 plots received nearly double those of the C1_26-30 plots) and the total number of treatments against pests and diseases, with the C2_26-30 plots receiving two more treatments on average than the C1_26-30 plots. The volume of pesticides that was sprayed was also higher in C2_26-30 and sodium arsenite was not used in the C1_26-30 plots. The same trends were observed for vegetation volume and competition in the inter-row and row.

The differences observed in each five-year period showed a link throughout the 16–30-year period between the three classes C1_16-20, C1_21-25 and C1_26-30, as well as between the C2_16-20, C2_21-25 and C2_26-30. In fact, C1_16-20, C1_21-25 and C1_26-30 could be considered as a part of a single management strategy and the same applies toC2_16-20, C2_21-25 and C2_26-30.

* Only the 11/15 variables showing significant differences between C1 and C2 for each five-year period are presented in detail in the table. No differences were found for pruning date, pruning mode, magnesium application or observed vigour. Values in bold indicate ta significant difference based on the confidence interval. Values highlighted in yellow are significantly different according to the principal component analysis.

By linking the succession of classes to which each plot belonged over the three five-year periods, we identified 11 practice-based trajectories that were followed by the plots in the database (Table 3). Three of these 11 practice-based trajectories were followed the most often by 43 plots: C2_15-20/C2_21-25/C1_26-30 (5 plots), C1_15-20/C1_21-25/C2_26-30 (8 plots) and C1_15-20/C1_21-25/C1_26-30 (14 plots). Three practice-based trajectories included a period during which a plot could not be assigned to C1 or C2; these plots were labelled ‘NA’ (NA-C2_21-25-C2_26-30, C1_15-20-C1_21-25-NA and NA-C2_21-25-C1_26-30). Of the 11 trajectories, five were associated with only one or two plots, which was particularly difficult to interpret; therefore, these five trajectories were not analysed. The next section focuses on the six main observed trajectories labelled with their summary names (Table 3).

* Values in bold indicate the most represented plot category in the trajectory. ND = non-declining, MD = moderately declining and SD = strongly declining. SD+MD = the sum of the moderately and strongly declining plots. NA: five-year period in which plots were not assigned to C1 or C2.

The cross-referencing of the practice-based trajectories and the categories of decline in Step 5 of the method (Table 3) resulted in the trajectories falling into two groups. Trajectories C1-C1-C1, C1-C1-C2 and C1-C1-NA were associated with a higher percentage of declining plots (strongly or moderately declining). These trajectories all contained C1 at least twice and with the following characteristics: lower competition between vines and associated vegetation in the 16-21-year period; higher competition in the 21-25-year period; and lower potash application, a lower volume of pesticides sprayed and less sodium arsenite in the 21–30-year period. By contrast, the trajectories C1-C2-C2, NA-C2-C2 and C2-C2-C1 comprised mainly non-declining plots. These three trajectories were largely associated with C2 classes during the study period. The two trajectories with the higher proportion of non-declining plots were those with C2 in the 21-25-year period.

Four variables describing plot characteristics and planting were selected for the analysis during Step 6: rootstock sensitivity to water deficit, vine vigour conferred by the rootstock, soil type and geographic location. Five soil categories (Table 1) were identified in the Cognac wine-growing region. Strongly declining plots were largely found in the silt-clay and clay-limestone soil categories; nutrient availability and aeration can be an issue in these soils. Conversely, limestone on chalk, and sandy and clay soils were found mostly in non-declining plots (Figure 6A); vines in these areas were characterised by a lower sensitivity to water deficit. Non-declining plots were also mainly located on mid-hillside areas, whereas a higher proportion of strongly declining and moderately declining plots was associated with a hillside location (Figure 6D). With regard to vine indicators, most non-declining plots were associated with strongly vigorous rootstocks (Figure 6C), whereas strongly declining plots of the C2-C2-C2 trajectory (a trajectory mainly associated with non-declining plots) were all associated with low-vigour rootstocks. It was also interesting to note that, conversely, on the strongly declining trajectory C1-C1-C1, plots also characterised by strongly vigorous rootstock at planting were identified as non-declining plots. Finally, a high level of sensitivity to water deficit conferred by the rootstock was associated with non-declining plots, whereas plots with a low level of sensitivity to water deficit conferred by the rootstock were often strongly or at least moderately declining (Figure 6B). Our data indicate that the plots in which rootstock were sensitive to water deficit were also those with lower water deficit, which could mean that winegrowers paid less attention to the water sensitivity of the rootstock in plots not highly exposed to water deficit; while showing a trend in this regard, our results were not conclusive, warranting further research

Figure 6. Distribution of decline categories for the four planting variables studied.

Distribution of strongly declining (red bars), moderately declining (yellow bars) and non-declining (green bars) plots for each of four static variables related to planting and soil–climate conditions: A) soil type, B) vigour conferred by the rootstock, C) rootstock sensitivity to water deficit, D) geographic location.

This study, which drew on more than 30 years of data on yield, mortality and agricultural practices in vineyards, highlights the complexity of the grapevine decline phenomenon and the myriad interactions between environmental characteristics, planting material and agricultural practices. We showed that grapevine trunk diseases and practices defined by water availability are key factors in triggering or promoting ongoing grapevine decline. Our findings indicate that water dynamics must be managed throughout the entire vineyard lifespan, from planting to grubbing up; this would involve, for example, making choices for the better management of soils with poor water retention capacity and choosing less drought-sensitive rootstocks and less competitive ground cover species. These results are all the more significant as vines are increasingly subject to water stress due to climate change, which affects both rainfall and temperatures.

We would like to thank the Bureau National Interprofessionnel du Cognac (BNIC) for providing the data for this study.