Introduction
Viticulture currently faces major challenges, being among the land uses most prone to erosion (Rodrigo-Comino, 2018) and one of the most reliant on pesticide applications (Chayre et al., 2024; Leal & Gramaje, 2024; Pertot et al., 2017). The practice of maintaining bare soils, along with frequent tractor use, results in soil compaction, the reduction of water infiltration, and leads to runoff and erosion. These phenomena are responsible for significant soil loss, but also at a broader scale for landslides, floods, and water pollution through the leaching of nutrients and pesticides (Rodrigo-Comino, 2018). Mediterranean vineyards are particularly affected because of a climate characterised by severe summer droughts combined with brief and intense rainfall events (Gaume et al., 2016), all the more so in a context of climate change, with shifts in the frequency and intensity of extreme precipitation events and intensified drought conditions (Dinis et al., 2022). Moreover, the decline in soil organic matter (Visconti et al., 2024) limits the formation of aggregates and increases soil erodibility (Le Bissonnais & Arrouays, 1997). There is an urgent need to improve the resilience of winegrowing systems by reducing pesticide use (Fouillet et al., 2022) and addressing the degradation of vineyard soils and water contamination associated with intensive management practices (Winter et al., 2018).
One of the answers to these challenges is to establish service crops, which are non-commercial crops grown to provide ecosystem services (Garcia et al., 2018; Lamichhane & Alletto, 2022). Indeed, service crops in viticulture improve aggregate stability, reduce soil erosion, and improve water infiltration (Garcia et al., 2018; Giffard et al., 2022). They also increase belowground and aboveground biodiversity, including natural enemies for vineyard pests that can help reduce the use of pesticides. In addition, service crops have a positive effect on increasing Soil Organic Carbon (SOC), which means they sequester carbon in the soil and help mitigate CO
2 emissions in the atmosphere (Abad et al., 2021a).
However, service crops are also prone to compete with grapevines for water and nitrogen (N) (Celette & Gary, 2013). Water stress induces vine stomatal closure, which reduces photosynthesis and grapevine growth (De Souza et al., 2005; Pellegrino et al., 2005). Water stress is intensified by the mismatch between the water demands of service crops, which start absorbing water before budburst, and those of grapevines, whose primary water requirements occur after budburst. Although the vine can concentrate its root system more in depth below the row instead of in the inter-row in the presence of a service crop in the inter-row, this is not likely to totally compensate for transpiration by the service crop (Celette et al., 2008). While in wet climates, the competition between service crops and vines can reduce excessive vigour and yield and improve wine quality, it is not the case in Mediterranean vineyards where water availability is a major limiting factor (Novara et al., 2021). Regarding soil N, legume service crops usually increase soil N content by fixing N
2 from the atmosphere (Ovalle et al., 2010; Ovalle et al., 2007), but grass service crops may compete for N with the grapevine (Celette et al., 2009). In the case of low soil moisture, service crops can also result in a decrease in N due to reduced N mineralisation (Celette & Gary, 2013). Moreover, an increase in soil N content in the presence of a legume service crop does not always result in an increase in Yeast Assimilable Nitrogen (YAN) (Abad et al., 2021a; Abad et al., 2021b). Finally, the risk of competition for water and N between grapevine and service crops is an obstacle to the adoption of service crops by winegrowers (Fernández-Mena et al., 2021; Sharifi & Zolfaghari, 2025), and may explain the relatively low proportion of vineyard associated with service crops in Languedoc (43 %) compared with the rest of France (64 %) (Agreste, 2025).
Service crop management strategies allow for achieving a balance between ecosystem services and disservices (Garcia et al., 2018; Griesser et al., 2022; Liebhard et al., 2024; Winter et al., 2018). In particular, it has been demonstrated that the choice of service crop species influences their impact on soil structure and the extent to which service crops compete with the vine for water (Capri et al., 2023; Novara et al., 2021; Garcia et al., 2020) and N (Curtright et al., 2025; Fernandez Pulido et al., 2025). A survey conducted in 2016 among 334 winegrowers in the Languedoc-Roussillon region of France highlighted the considerable spatial and temporal diversity in service crop management strategies employed by farmers (Fernández-Mena et al., 2021). Despite this diversity, most studies evaluating the ecosystem services provided by service crops in viticulture or their impacts on vineyards in dry climates do not reflect this complexity. Instead, they typically adopt a binary perspective, contrasting soil tillage practices with permanent cover using either sown or natural vegetation (Gaudin et al., 2010; Lopes, 2016; Pou et al., 2011; Ramírez-Pérez et al., 2024). The effect of a third treatment involving partial spatial or temporal disturbance in the inter-row was assessed by Griesser et al. (2022) in temperate wine-growing regions. Similarly, the importance of service crop termination methods has been demonstrated in other crops, such as vegetable production (Eivazi et al., 2024) and irrigated corn (Alonso-Ayuso et al., 2020). All the same, Delpuech and Metay (2018) demonstrated a decrease in grapevine yield with the percentage of soil covered by a service crop and recommended partial service crops for Mediterranean vineyards. Recently, new service crop termination methods have also been studied, such as interrow rolling and sub-row mulching compared with green manuring and alternate tilled-grassed interrow in Northern Italy (Cunial et al., 2025; Pelusi et al., 2025). However, there is still a lack of knowledge on how the diversity of service crop management strategies affects both vineyards and ecosystem services in the Mediterranean context.
Garcia et al. (2024) studied the combined effects of service crop termination tool (mower, mower + superficial tillage, roller-crimper) and period (early termination
vs budburst termination) with a three-year experiment. They observed that the termination through superficial tillage was more effective at killing service crops and thus at reducing competition for water and N with the grapevine compared to other methods. If the budburst termination period combined with no-till termination methods showed higher soil microbial biomass, the termination period was not shown to influence soil water stock and grapevine water status. However, the termination at budburst analysed in this study is not very late, and one wonders about the impact of termination at grapevine flowering when there is no study on the subject.
In the present study, we conducted a three-year on-farm experiment between 2022 and 2024 on 7 commercial farms in the French departments of Hérault, Aude, and Pyrénées-Orientales. During three years, we evaluated different service crop termination strategies for service crop mixtures, combining two service crop termination periods (around budburst and one month after budburst) and two termination methods (mower + superficial tillage, roller-crimper), compared with a tilled control treatment. We monitored the service crop biomass, the surface temperature and water potential of the soil in the inter-row, and a variety of indicators on the grapevine (yield components, pruning weight, apex growth indicator, δ
13C, yeast assimilable nitrogen (YAN), sugar and pH in musts).
The following question was investigated: Which operational service crop management strategy can be adapted to limit disservices in a Mediterranean context? We hypothesised that i) terminating the service crop at grapevine budburst and with a superficial tillage suppresses competition for water and N, thus promoting grapevine yield and vigour, and that ii) roller-crimper termination has a greater soil surface temperature buffering effect than superficial tillage termination, due to the plant residues left on the soil surface, but increases competition with the grapevine for water and N.
Materials and methods
1. Site description and experimental design
The experiment took place over three years in 2022, 2023, and 2024 on seven fields from a network of organic wine farms in the departments of Hérault, Aude, and Pyrénées-Orientales (Occitanie region of France, Figure 1A). Monthly rainfall and mean monthly temperature data for the seven fields and over the three years of the experiment are presented in Figure 2. It should be noted that the years of the experiment, in particular 2022–2023 and 2023–2024, were dry years compared with the rainfall usually observed in the region. Indeed, mean annual rainfall over the experimental period was below the 30-year regional average in all three seasons, with deficits of 47 mm (2021–2022), 227 mm (2022–2023), and 195 mm (2023–2024). All fields were planted with a Syrah cultivar except fields 34.2 (Carignan cultivar) and 66.3 (Grenache cultivar), and had different topographical contexts, soil types, and management practices (Tables 1 and 2).
Each field was divided into 5 groups of 2 to 3 inter-rows (Figure 1B) to which 5 different treatments were applied. The treatments consisted of a control maintained with bare soil with tillage, and 4 sown treatments differing in the method and period of service crop termination: early tillage (
i.e., around budburst, E-T) or late tillage (
i.e., around 1 month after budburst, L-T), early roller (E-R) or late roller (L-R) (Figure 1B). In the E-T and B-T treatments, tillage was preceded by mowing to improve termination efficiency and prevent crop residues from being carried out of the inter-row. Each treatment had 4 to 6 plots of 5 vines each. The dates of sowing, early and late termination, and the species sown for each year and each field are listed in Tables S1 and S2.
The service crops sown were generally composed of a mixture of legumes (Fabaceae), grasses (Poaceae), and crucifers (Brassicaceae). The precise composition of the mixture varied from field to field and from year to year depending on farmers’ choice (Table S1). The seeds came from a local cooperative producing population varieties (SCIC Graines équitables).
Figure 1. A: location of the commercial farms. The dots correspond to the locations of the fields. The lines correspond to the administrative boundaries of the departments of Hérault, Aude, and Pyrénées-Orientales. B: experimental design for each field. E: early termination; L: late termination; T: tillage; R: roller.
![]()
Figure 2. Umbrothermal diagram for 2021–2022, 2022–2023, and 2023–2024 for the 7 fields monitored during the experiment. In blue, monthly cumulative rainfall. In red, the mean monthly temperature (solid line) and the mean monthly lower and upper temperatures (dotted lines). For each year and each field, the cumulative annual rainfall from September to August is shown. For comparison, mean rainfall calculated over the last 30 years for each field: 34.1: 794 mm, 34.2: 647 mm, 11.1: 672 mm, 11.2: 610 mm, 66.1: 575 mm, 66.2: 661 mm, 66.3: 562 mm. Data source: Météo-France via the SICLIMA platform (Maury <italic>et al.</italic>, 2024).
![]()
Table 1. Characteristics of the fields monitored during the experiment
|
Field
|
Topographical context
|
Soil
|
Grape cultivar
|
Rootstock
|
Density
|
|
66.1
|
Plain
|
Fine chalky silt
|
Syrah
|
110 Richter
|
4,000 vines ha
–1
|
|
11.1
|
Plateau
|
Clay-silt
|
Syrah
|
140 Ruggeri
|
4,500 vines ha
–1
|
|
11.2
|
Plain (canal side)
|
Clay-limestone, 40 % stones
|
Syrah
|
Fercal
|
5,000 vines ha
–1
|
|
34.1
|
Scrubland
|
Clay-silt
|
Syrah
|
101-14 Mgt
|
5,000 vines ha
–1
|
|
66.2
|
Hillsides
|
Sandy-limestone schist
|
Syrah
|
110 Richter
|
4,444 vines ha
–1
|
|
34.2
|
Plain
|
Alluvial deposits, clayey-silt-limestone, low stone content.
|
Carignan
|
110 Richter
|
4,000 vines ha
–1
|
|
66.3
|
Plain
|
Quaternary limestone alluvium
|
Grenache
|
110 Richter
|
4,000 vines ha
–1
|
Table 2. General management practices of the fields monitored during the experiment
|
Field
|
Training system
|
Irrigation
|
Fertilisation
|
|
66.1
|
Cordon spur pruning
|
No irrigation
|
Chicken manure pellets (5-5-5), 500 kg ha
–1
|
|
11.1
|
Cordon spur pruning
|
No irrigation
|
Humiflor
® Germi’One®, 1 t ha
–1
|
|
11.2
|
Guyot Poussard
|
100 mm per season from May to August (20 mm every 15 days)
|
Organic fertiliser (9-4-0), 600 kg ha
–1 + green waste in winter, 20 t ha
–1
|
|
34.1
|
Cordon spur pruning
|
No irrigation
|
No fertilisation
|
|
66.2
|
Cordon spur pruning
|
No irrigation
|
20 t ha
–1 of green waste compost every 3 years (application at the beginning of the trial only)
|
|
34.2
|
Goblet pruning
|
No irrigation
|
Organic fertiliser (5-3-2), 500 kg ha
–1
|
|
66.3
|
Cordon spur pruning
|
156 mm per season from May to August (12 mm every 7 days)
|
Chicken manure pellets (5-5-5), 500 kg ha
–1
|
2. Data gaps, missing data and experimental errors
In the framework of this participatory research conducted with winegrowers, some data gaps were inevitable due to the diversity of farm contexts and the autonomy of participants in managing their fields (Roques et al., 2022). In particular, precise information on the exact composition of the sown service crop mixtures and on the specific dates of sowing and termination was sometimes unavailable. Additional missing data may also occur when farmers do not inform the research team before destroying the service crop or pruning the vines. These variations and occasional data gaps reflect the flexibility inherent to on-farm experiments, where farmers adapt practices to local conditions, seasonal constraints, and their own decision-making, thereby enriching the study with real-world relevance despite incomplete records. All missing data are indicated in Tables S3 and S4.
Regarding the treatments, in a few cases, when the biomass of the service crops was not sufficient to be effectively destroyed by the roller, the service crops were mown instead of being rolled (Table S3).
3. Measured variables
3.1. Service crop biomass
The fresh biomass of the service crops was measured before each termination period. Three 1 m
2 quadrats were randomly taken from the modalities sharing the same termination period, and the species were sorted to record their fresh biomass (Table S5). The dry biomass was then calculated from the fresh biomass using the MERCI method (Constantin et al., 2024). This method enables the estimation of the dry biomass and nutrient content of many service crop species through a large database, by indicating their sowing and biomass sampling dates.
Some service crops were destroyed before biomass measurements could be done. At the time of early termination, plots assigned to the late termination treatment had not yet been destroyed and therefore continued to accumulate biomass. Consequently, their standing biomass at late termination was necessarily at least equal to – and likely greater than – the biomass recorded just before early termination, consistent with the well-established relationship between growing duration and service crop biomass production (Pessotto et al., 2025). For site-years where only pre-early-termination biomass data were available, this value was therefore used as a conservative lower-bound estimate of biomass at late termination. Furthermore, biomass values were not included as explanatory variables in the statistical models used to assess vine performance. The field was treated as a random effect, which accounts for the agronomic variability across the trial network – including differences in service crop development levels – without requiring each site-year to provide a complete set of biomass measurements.
3.2. Soil water potential
To measure soil water potential (0 to 199 kPa), three tensiometric probes (Watermark) per treatment were installed at a depth of 45 cm, in the middle of the inter-row in each treatment (Figure 1B). Measurements were taken once a week over the three seasons of the experiment, from around September to August for the fields in the Pyrénées Orientales and from around March to August for the fields in the Aude and Hérault.
3.3. Soil surface temperature
The soil surface temperature was measured using a probe thermometer at 5 cm depth, with 3 replicates next to each tensiometer probe, for a total of 9 replicates per treatment. The measurements were carried out once a week from April to August, in the morning, at the same time as the tensiometer probe measurements.
3.4. Apex growth index
The apex growth index was calculated using the ApeX-Vigne method and a smartphone application (Pichon et al., 2021). This method classifies shoot growth into three categories: full growth (FG), slow growth (MG), and stopped growth (SG). Scoring was carried out on 50 shoot tips per treatment in the experimental plots by observing between 10 and 15 randomly selected shoots per plot. The percentages of each category were used to calculate the index of growing apex (iG-Apex) according to the following formula:
(Equation 1)
i
G
A
p
e
x
=
P
F
G
×
1
+
P
M
G
×
0.5
+
P
S
G
×
0
where
P
F
G
represents the proportion of full-growing shoot tips,
P
M
G
the proportion of slow-growing shoot tips, and
P
S
G
the proportion of dead shoot tips. Measurements were recorded every two weeks, from the end of vine flowering to the beginning of veraison, on vines that had not been pruned or topped.
3.5. Vine yield and berry quality
Different components of vine yield were measured each year. The number of bunches per vine was counted after vine flowering, on each plant in the 4 plots of each treatment (20 replicates per treatment). This measurement was only carried out in 2023 and 2024, as the number of bunches in that year was determined by the water and N stress around flowering in the previous year (Guilpart et al., 2014). At harvest, the total harvest from each plot of 5 vines was weighed (4 replicates per treatment), and the berry weight was measured on one sample of 200 berries taken from each treatment. The 200 berries were then sent to a laboratory to analyse the pH, yeast assimilable nitrogen (YAN), sugar concentration and δ
13C of the musts. The δ
13C is an indicator of grapevine water status (Gaudillère et al., 2002).
3.6. Pruning weight
The total pruning weight of each plot was weighed when the vines were pruned as decided by the winegrowers (4 replicates per treatment). For the field 34.2 in 2024, the vines had already been pruned at the time of the measurements, and the pruned shoots were left on the ground near the plots. Under these conditions, it was not possible to ensure that all shoots originating from the targeted plot were collected. Consequently, the protocol was adapted, and pruning weight was estimated by collecting 20 vine shoots randomly selected in the immediate vicinity of each plot. To ensure consistency across fields, pruning weight data were subsequently analysed as relative differences compared with the control treatment for all fields.
4. Climatic variables
The Climatic Water Balance (CWB) was calculated over different periods of the vine cycle, with CWB = P – ETP, where P: precipitation and ETP: potential evapotranspiration. The climatic data came from Météo-France and were downloaded via the SICLIMA platform developed by AgroClim-INRAE (Maury et al., 2024). Different periods of the vine cycle were considered: the CWB from mid-May to mid-June of year n–1, the critical period for determining the number of bunches in year n being around flowering in year n–1 (Guilpart et al., 2014), the CWB from July to August of year n, the period between fruit set and veraison, which has a major influence on berry size and nutrient uptake (Ojeda & Saurin, 2014), and finally the CWB from April to August,
i.e., during the vine vegetative growth period.
5. Statistical analyses
5.1. Calculations on variables
For yield, number of bunches, 200-berry weight, and must quality variables, differences from the control treatment (ΔY) were calculated. That is, for each field and year, the mean value of the control (
Y
‾
) was subtracted from all observed values.
(Equation 2)
Δ
Y
=
Y
-
Y
‾
c
o
n
t
r
o
l
For the pruning weight, relative differences compared with the control (
Δ
Y
/
Y
‾
c
o
n
t
r
o
l
) were calculated to include data from field 34.2 in 2024:
(Equation 3)
Δ
Y
/
Y
‾
c
o
n
t
r
o
l
=
Y
-
Y
‾
c
o
n
t
r
o
l
Y
‾
c
o
n
t
r
o
l
5.2. Data analysis strategy
Linear mixed models were used to explain the variables yield, number of bunches per vine, pruning wood mass, berry weight, yeast assimilable nitrogen, sugar, and pH. For each of these variables, the Treatment, Year and CWB(s) (calculated from mid-May to mid-June of year n–1 for the number of bunches and yield, from July to August of year n for berries weight, must quality variables and yield, and from April to August of year n for pruning wood mass), as well as the interaction between Year and Treatment, were tested as explanatory variables. For the must quality variables (YAN, sugar, and pH), berry weight was also added as an explanatory variable. The assumptions of the linear model were systematically verified. For all the models, the factor Field was considered as a random effect by adding a random intercept. Random slopes linked to the field and associated with climatic variables and the berry weight variable were also tested for inclusion. The random effects, then the fixed effects, were selected using a top-down approach, starting with the complete model and then progressively eliminating certain variables on the basis of likelihood ratio tests. The assumptions of normality and heteroscedasticity were systematically verified on the residuals.
Hierarchical
generalised additive models (HGAM) were used to explain the soil water potential, soil surface temperature, and iG-Apex data, which are temporal data, represented by curves. HGAMs can be used to model complex non-linear functions between covariates, with the form of the function varying according to levels of qualitative variables (Pedersen et al., 2019). The formula for the model used for the three temporal variables analysed was as indicated in Eq. 1.
(Equation 4)
Y
i
j
k
l
(
t
)
=
s
t
+
s
1
i
t
+
s
2
j
t
+
s
3
k
t
+
ϵ
i
j
k
l
(
t
)
Y is the variable to be explained, i the index for treatment, j for the year, k for the field, and l for the replicate. t denotes the time of observation. Hence, Y
ijkl(t) is the measure of Y for the i
th treatment, during the j
th year, in the k
th field, for the l
th replicate, and at time t. s(t) represents a temporal curve representing the overall, global non-linear effect of time (
i.e., days of the experiment) over the season, s1
i(t) represents the time effect specific to the modality of treatment, s2
j(t) represents the time effect specific to the year, s3
k(t) the time effect specific to the field, and the ϵ
ijkl(t) are i.i.d. N(0,σ
2) random variables. Diagnostics were used to check the obtained models. First, convergence of the fitting algorithm was checked. Then, residual plots were used to verify normality and homoscedasticity of the residuals. The number of basis functions used for smooth terms was also studied and compared to the effective degrees of freedom, to ensure that they are not too small, in which case the terms could fail to capture the relationships in the data. Plots to compare fitted values and responses were also plotted to visualise the adequacy of models to the data.
All analyses were carried out using R version 4.4.0. The package nlme (Pinheiro et al., 1999) was used for the linear mixed models. For the HGAM models, the analyses were carried out with the mgcv package (Pedersen et al., 2019; Wood, 2000).
Results
1. Service crops biomass
The biomass of the service crops for the different fields and years of the experiment varied between around 0.5 t ha
–1 and 7.5 t ha
–1 (Figure 3). In general, biomass differences between late and early termination were low, except for field 66.2 in 2022, where the biomass increased by a factor of ~1.5 between the two periods, and field 11.2 in 2023, where the biomass doubled between the two periods (Figure 3).
Figure 3. Dry biomass of the service crop before early termination and before late termination for the three years and for each of the fields monitored during the experiment. For fields 11.2 and 34.1 in 2022, the data represent the average dry biomass value for all families combined, due to missing data (Table S5). For some missing data, we assumed that the biomass before late termination was at least equal to the biomass measured before early termination, represented by grey question marks on the graph. Error bars represent standard deviations around the mean total biomass per field.
![]()
2. Grapevine growth and yield
2.1. Yield components
All fields and years combined, the number of bunches per vine varied from 0 to 52 bunches, with an average of 16.08. Only the factor "Treatment" significantly explained the number of bunches (Table 3). Treatments E-R and L-R showed significantly lower numbers of bunches of –1.46 and –1.78 bunches per vine, respectively, compared with the control treatment (Figure 4A, Table 3). There was no significant difference in the number of bunches between E-T, L-T, and the control treatment.
Table 3. Results of the linear mixed models explaining berry weight, number of bunches, yield, pruning weight, YAN, pH, sugar and δ
<sup>13</sup>C in musts. The table gives mean differences between the different levels of the factors and the control, associated standard errors, and a
<italic>p</italic>-value for the comparison between the levels and the control (intercept). For berry weight and pruning weight, the intercept corresponds to the control treatment in 2022. The letters in the Group column correspond to the result of a post-hoc multiple comparison test; uppercase letters indicate significant differences among years, whereas lowercase letters indicate significant differences among treatments. Two levels sharing the same letter do not differ significantly at a first-order risk of α = 0.05. E: early termination; L: late termination; T: tillage; R: roller.
|
Term
|
Estimate
|
Standard error
|
p-value
|
Group
*
|
|
Difference in berry weight relative to control (g)
|
|
(Intercept)
|
15.89
|
8.74
|
0.07279
|
|
|
2023
|
–24.84
|
10.23
|
0.01739
|
A
|
|
2024
|
–3.04
|
10.50
|
0.77326
|
B
|
|
E-T
|
–5.18
|
9.51
|
0.58720
|
cd
|
|
L-T
|
–21.05
|
9.51
|
0.02965
|
bc
|
|
E-R
|
–45.47
|
9.51
|
<0.001
|
a
|
|
L-R
|
–30.27
|
9.51
|
0.00206
|
ab
|
|
Difference in number of bunches relative to control
|
|
(Intercept)
|
–0.11
|
0.81
|
0.89521
|
|
|
E-T
|
0.76
|
0.59
|
0.19939
|
c
|
|
L-T
|
0.17
|
0.59
|
0.77248
|
c
|
|
E-R
|
–1.46
|
0.59
|
0.01353
|
ab
|
|
L-R
|
–1.78
|
0.59
|
0.00263
|
a
|
|
Difference in yield relative to control (kg)
|
|
(Intercept)
|
–0.23
|
0.14
|
0.09245
|
|
|
E-T
|
0.14
|
0.13
|
0.27835
|
c
|
|
L-T
|
–0.24
|
0.13
|
0.06681
|
ab
|
|
E-R
|
–0.35
|
0.13
|
0.00905
|
a
|
|
L-R
|
–0.35
|
0.13
|
0.00812
|
a
|
|
Difference in pruning weight relative to control (%)
|
|
(Intercept)
|
–93.38
|
33.00
|
0.00534
|
|
|
2023
|
9.38
|
8.63
|
0.27872
|
A
|
|
2024
|
22.49
|
10.94
|
0.04157
|
A
|
|
E-T
|
–3.00
|
5.19
|
0.56497
|
c
|
|
L-T
|
–20.49
|
5.19
|
<0.001
|
b
|
|
E-R
|
–34.48
|
5.06
|
<0.001
|
a
|
|
L-R
|
–40.11
|
5.06
|
<0.001
|
a
|
|
CWB April-August
|
–0.12
|
0.04
|
0.00513
|
|
|
Difference in YAN relative to control (mg L
–1)
|
|
(Intercept)
|
–43.39
|
17.63
|
0.01588
|
|
|
E-T
|
–27.43
|
9.39
|
0.00449
|
a
|
|
L-T
|
–42.22
|
9.43
|
<0.001
|
a
|
|
E-R
|
–38.41
|
9.59
|
<0.001
|
a
|
|
L-R
|
–41.68
|
9.48
|
<0.001
|
a
|
|
Weight 200 berries
|
0.15
|
0.04
|
<0.001
|
|
|
Difference in pH relative to control
|
|
(Intercept)
|
0.01
|
0.02
|
0.44954
|
|
|
E-T
|
–0.05
|
0.02
|
0.00650
|
a
|
|
L-T
|
–0.05
|
0.02
|
0.00920
|
ab
|
|
E-R
|
–0.05
|
0.02
|
0.00926
|
ab
|
|
L-R
|
–0.07
|
0.02
|
<0.001
|
a
|
|
Difference in sugar relative to control (g L
–1)
|
|
(Intercept)
|
0.80
|
2.19
|
0.71609
|
|
|
E-T
|
1.52
|
2.36
|
0.52162
|
ab
|
|
L-T
|
6.10
|
2.36
|
0.01132
|
ab
|
|
E-R
|
7.28
|
2.36
|
0.00271
|
b
|
|
L-R
|
5.97
|
2.36
|
0.01313
|
ab
|
|
Difference in δ
13C relative to control
|
|
(Intercept)
|
–0.18
|
0.25
|
0.4677
|
|
|
2023
|
–0.23
|
0.17
|
0.1875
|
A
|
|
2024
|
0.47
|
0.21
|
0.0246
|
B
|
Figure 4. Effect of termination date and method on the number of bunches (A), berry weight (B), yield (C), pruning weight (D), and apex growth indicator (E). E: early termination; L: late termination; T: tillage; R: roller. (A, B, C, D) For the number of bunches and yield, only the Treatment factor was retained after model selection. For berry weight and pruning weight, only the factors Treatment and Year were retained after model selection. The stars indicate the significant differences resulting from tests comparing each of the termination methods with the control (*** :
<italic>p</italic>-value ≤ 0.001; ** :
<italic>p</italic>-value ≤ 0.01; * :
<italic>p</italic>-value ≤ 0.05; . :
<italic>p</italic>-value ≤ 0.1; :
<italic>p</italic>-value > 0.1). The points surrounded in black correspond to the outliers calculated using the Tukey criterion. (E) Global function s(t) and treatment-specific deviations s
<sub>1i</sub>(t) for the HGAM model on the apex growth indicator. The actual data for all years, treatments and fields combined are represented by the grey dots.
![]()
The berry weight ranged from 83.3 to 560 g, with an average of 317.1 g. The factors retained after model selection were Treatment and Year, but not their interaction (Table 3), meaning that the Treatment effect was the same each year. Berry weight was significantly lower at 24.84 g in 2023 compared to 2022. In 2024, however, the berry weight was not significantly different from 2022. Treatments L-T, E-R, and L-R showed significant differences of –21.06, –45.47, and –30.27 g, respectively, compared with the control treatment, whereas no significant difference was observed between the E-T and the control treatment (Table 3, Figure 4B).
All fields and years combined, the average yield was 2.309 kg vine
–1, with a minimum of 0.095 kg vine
–1 and a maximum of 8.054 kg vine
–1. Only the factor Treatment significantly explained the difference in yield relative to the control (Table 3). Treatments E-R and L-R showed significantly lower yields of 0.35 kg vine
–1 compared with the control (Figure 4C, Table 3). There was no significant difference in yield between treatments E-T, L-T, and the control, but we observed a yield difference of 0.24 kg vine
–1 between L-T and the control.
To sum up, roller terminations (E-R and L-R) significantly impacted grapevine yield and number of bunches, contrary to tillage termination strategies (E-T and L-T). The late tillage termination (L-T) and the roller termination strategies (E-R and L-R) had lower berry weight compared to the control. Finally, the early tillage termination (E-T) did not affect any of the yield components.
2.2. Pruning weight
Over the experiment, the pruning weight per vine varied from 0.014 to 0.937 kg vine
–1, with an average of 0.378 kg vine
–1. The factors Treatment, Year, and CWB
[April-August] were retained after model selection but not the interaction Treatment:Year. Pruning weight in 2024 was significantly higher by 22.49 % compared to 2022, whereas 2023 did not differ significantly from 2022 (Table 3). Treatments L-T, E-R and L-R had significantly lower pruning weights than the control by –20.49 %, –34.48 % and –40.11 %, respectively (Table 3, Figure 4D). On the other hand, the pruning weight of the E-T treatment was not significantly different from the control (Figure 4D).
2.3. Apex growth index
For the apex growth index (iG-Apex), the results of the hierarchical generalised additive model (HGAM) showed a global curve decreasing from approximately 1 to 0.3 from May 1st to the beginning of August, meaning that iG-Apex decreased overall during the season (Figure 4E). The factors Treatment, Year, and Field significantly influenced the variation of iG-Apex over time (Table 4). Analysis of the model-predicted data showed that the E-T treatment exhibited a similar apex growth dynamic to the control, with a slower growth termination rate than the other treatments (
i.e., a less negative slope due to a positive deviation from the overall mean slope across treatments). The L-T treatment also displayed a positive deviation from the mean slope, although to a lesser extent, indicating a somewhat faster growth termination (Figure 4E). In contrast, both roller treatments showed negative deviations from the overall mean slope, reflecting an earlier and faster termination of shoot growth compared with the other treatments.
Table 4. Results of the HGAM models for the variable Apex growth index. The column "edf" gives the effective degrees of freedom for each term: this value represents the complexity of the smooth. An edf of 1 is equivalent to a straight line. An edf of 2 is equivalent to a quadratic curve, and so on, with higher edfs describing more wiggly curves. The
<italic>p</italic>-value columns show the results of significance tests for each smooth term: a significant smooth term is one for which one cannot draw a horizontal line through 0 within the 95 % confidence interval. s(t) represents a temporal curve representing the overall, global non-linear effect of time (
<italic>i.e.</italic>, days of the experiment) over the season; s
<sub>1i</sub>(t) represents the time effect specific to the modality of treatment; s
<sub>2j</sub>(t) represents the time effect specific to the year; s
<sub>3k</sub>(t) represents the time effect specific to the field
|
Term
|
edf
|
p-value
|
|
Apex growth indicator
|
|
s
t
|
1.00
|
<0.001
|
|
s
1
i
t
|
6.68
|
<0.001
|
|
s
2
j
t
|
12.42
|
<0.001
|
|
s
3
k
t
|
31.01
|
<0.001
|
|
Soil water potential
|
|
s
t
|
2.84
|
0.31
|
|
s
1
i
t
|
14.79
|
<0.001
|
|
s
2
j
t
|
23.09
|
<0.001
|
|
s
3
k
t
|
36.28
|
<0.001
|
|
Soil surface temperature
|
|
s
t
|
1.01
|
<0.001
|
|
s
1
i
t
|
3.66
|
<0.001
|
|
s
2
j
t
|
24.16
|
<0.001
|
|
s
3
k
t
|
28.20
|
<0.001
|
3. Must quality: Yeast Assimilable Nitrogen (YAN), sugar, and pH
For YAN, the fixed effects retained after model selection were the factors Treatment and Berries weight. For pH and sugar, only the factor Treatment was retained after model selection (Table 3).
Across all fields and years, YAN varied from 33 to 427 mg L
–1 with an average of 183.1 mg L
–1, pH varied from 3.21 to 4.48 with an average of 3.61, and sugar from 185 to 264 g L
–1 with an average of 233 g L
–1. All treatments showed a significantly lower YAN concentration compared with the control treatment, respectively by –27.43, –42.22, –38.41, and –41.68 mg L
–1 for treatments E-T, L-T, E-R, and L-R, respectively, although the differences between treatments E-T, L-T, E-R, and L-R were not significant (Figure 5A, Table 3). The sugar content in the musts also varied according to treatment. Treatments L-T, E-R, and L-R had a significantly higher sugar concentration of 6.10, 7.28, and 5.97 g L
–1 compared with the control treatment, while the sugar concentration of treatment E-T was not significantly different from the control (Table 3, Figure 5C). Finally, all treatments had a significantly lower pH compared with the control treatment: –0.05 units for treatments E-T, L-T, and E-R and –0.07 units for treatment L-R (Table 3). The presence of service crops, regardless of the termination method used, therefore had the effect of lowering the pH of the musts (Figure 5B). With regard to the δ
13C of the musts, only the factor Year was retained after selection of the model, meaning that Treatment did not significantly affect the δ
13C (Table 3).
Figure 5. Effect of termination date and method on A: the Yeast Assimilable Nitrogen (YAN), B: pH, and C: sugar in musts. E: early termination; L: late termination; T: tillage; R: roller. For YAN, the factors Treatment and Berries weight (co-variable) were retained after model selection. For pH and sugar, only the factor Treatment was retained after model selection. The points surrounded in black correspond to the outliers calculated using the Tukey criterion. The stars indicate the significant differences resulting from tests comparing each of the termination methods with the control (*** :
<italic>p</italic>-value ≤ 0.001 ; ** :
<italic>p</italic>-value ≤ 0.01 ; * :
<italic>p</italic>-value ≤ 0.05 ; . :
<italic>p</italic>-value ≤ 0.1 ; :
<italic>p</italic>-value > 0.1).
![]()
4. Soil variables: water potential and temperature at ground surface
Regarding the soil surface temperature, the results of the HGAM model showed an average curve increasing from approximately 20 °C to 32 °C from April 1st to the end of August. The factors Treatment, Year, and Field significantly affected the variation in soil surface temperature over time (Table 5). Analysis of the model's predicted data showed that treatments L-T, E-R, and L-R had lower soil surface temperatures from 1.5 °C at the beginning of April and 1.05 °C at the end of August compared with the control (Figures 6A and 6B). To a lesser extent, E-T had a soil surface temperature 0.5 °C lower than the control throughout the season.
Table 5. Results of the HGAM models for the variables Soil water potential and Surface Temperature. The column "edf" gives the effective degrees of freedom for each term: this value represents the complexity of the smooth. An edf of 1 is equivalent to a straight line. An edf of 2 is equivalent to a quadratic curve, and so on, with higher edfs describing more wiggly curves. The
<italic>p</italic>-value columns show the results of significance tests for each smooth term: a significant smooth term is one for which one cannot draw a horizontal line through 0 within the 95 % confidence interval. s(t) represents a temporal curve representing the overall, global non-linear effect of time (
<italic>i.e</italic>., days of the experiment) over the season, s
<sub>1i</sub>(t) represents the time effect specific to the modality of treatment, s
<sub>2j</sub>(t) represents the time effect specific to the year, s
<sub>3k</sub>(t) represents the time effect specific to the field.
|
Term
|
edf
|
p-value
|
|
Soil water potential
|
|
s
t
|
2.84
|
0.31
|
|
s
1
i
t
|
14.79
|
<0.001
|
|
s
2
j
t
|
23.09
|
<0.001
|
|
s
3
k
t
|
36.28
|
<0.001
|
|
Soil surface temperature
|
|
s
t
|
1.01
|
<0.001
|
|
s
1
i
t
|
3.66
|
<0.001
|
|
s
2
j
t
|
24.16
|
<0.001
|
|
s
3
k
t
|
28.20
|
<0.001
|
Figure 6. Effect of termination date and method on the Temperature at the ground surface (A, B) and on the soil water potential (C, D). The black line and curve on the left panels (A, C) correspond to the global function s(t) and the coloured lines and curves in the right panels (B, D) correspond to treatment-specific deviations s
<sub>1i</sub>(t) from the HGAM models. The data for all years, treatments, and fields combined are represented by the grey dots. E: early termination; L: late termination; T: tillage; R: roller.
![]()
Regarding the soil water potential, the results of the hierarchical generalised additive model (HGAM) showed a global curve rising from around 50 to 160 kPa from March 1st to the end of August, indicating that the soil dried out during this period (Figure 6C). The factors Treatment, Year, and Field significantly influenced the variation in soil water potential over time (Table 5). Analysis of the data predicted by the model (Figures 6C and 6D) showed that from the beginning of April, the control and E-T treatments had a soil water potential around 20 to 50 kPa lower than the L-T, E-R, and L-R treatments, indicating higher soil water content in the control and E-T treatments. In contrast, in the fields 66.1, 66.2, and 66.3, where water potential was monitored during the autumn–winter period (September to December), the E-R and L-R treatments exhibited soil water potential approximately 50 kPa lower than those of the control, E-T, and L-T treatments. In other words, during winter, soils were drier under the control, E-T, and L-T treatments than under the E-R and L-R treatments (Figure S1).
Conclusion
This on-farm experiment conducted in Mediterranean vineyards demonstrates that the date and method of service crop termination strongly condition their agronomic outcomes, particularly through their effects on soil water dynamics and vine water stress. However, given the variability in soil types, grapevine management, and climatic conditions across the different fields, and the participatory nature of the experiment, these results should be interpreted as context-dependent: the relative performance of termination strategies may vary with local agronomic conditions, and early tillage termination may not be equally preferable in wetter climates or on soils with higher water-holding capacity. Among the tested strategies, termination by tillage at grapevine budburst emerged as the most balanced option, as it preserved vine vigour and yield at levels comparable to the tilled control, while avoiding the increased water stress observed under late-termination or roller-crimping treatments. In contrast, late termination and roller-crimping, despite improving winter water infiltration and reducing soil surface temperature, induced higher soil drying during the growing season, greater vine water stress, and subsequent reductions in yield components. Service crops also modified must composition, notably by reducing yeast assimilable N and slightly decreasing pH, effects that should be interpreted in light of winegrowers’ production objectives and the availability of adaptive management options during vine growth and vinification. Importantly, early tillage termination did not increase must sugar concentration, suggesting that this strategy is unlikely to exacerbate rising alcohol levels under climate change. Finally, despite the inherent challenges of participatory on-farm experimentation – including protocol deviations and occasional data gaps – the trust-based relationships established between researchers, facilitators, and winegrowers, together with the co-construction of experimental objectives and protocols, contributed to the overall quality of the dataset and to the relevance of the knowledge produced. For most participating winegrowers, who were already using service crops prior to the experiment, this approach provided an opportunity to refine their management strategies through collective discussion and iterative learning. This study thus highlights the value of actively involving farmers in knowledge production and confirms the relevance of on-farm experimentation as a cornerstone of the agroecological transition of viticulture under increasing climatic constraints.