Primary and secondary growth of Pinus halepensis are more sensitive to inter-annual drought variability than to 14 years of rainfall exclusion in a Mediterranean forest

Abstract

Background and Aims

With ongoing climate change, the impact of droughts of increasing intensity on forest functioning is of critical concern. While the adverse effects of drought on tree secondary growth have been largely documented both at the tree and stand scales, our understanding of how primary growth morphological traits, which control crown development, respond to drought remains limited, especially in the long term.

Methods

Based on 14 years of monitoring of four primary growth morphological traits (e.g. shoot elongation, polycyclism rate, branching and needle length) and stem secondary growth in a rainfall exclusion experiment, we investigated (1) the climatic drivers of above-ground growth and (2) the effect of long-term exacerbated drought conditions on the growth response to drought in a mature Pinus halepensis stand.

Key Results

Above-ground growth was strongly and negatively impacted by drought duration during the current year (stem secondary growth), the previous year (polycyclism) and both years (branching, shoot length), and by drought during spring (needle length). While excluding 30 % of the incoming rainfall did not significantly affect the number of ramifications, polycyclism rate or stem secondary growth, it reduced needle and shoot lengths by 14.3 and 7.7 % over the entire study period, respectively. However, this effect was significant only in the first years after the treatment was established. Such acclimation to exacerbated drought conditions is also reported in the drought–growth relationships which are similar among treatments, except for needles that were slightly shorter under a similar level of drought stress in the exclusion.

Conclusions

Our study highlights the key acclimation capacity in the primary and secondary growth response of P. halepensis to drought. In addition to tree structural adjustments, the relatively limited effect of the 30 % rainfall exclusion may also be caused by (1) the substantial inter-annual rainfall variability typical of Mediterranean climates, which modulates the exclusion effect on drought duration, and (2) the inherent inter-individual variability in drought sensitivity.

INTRODUCTION

Climate change has led to an increase in drought frequency, duration and intensity in many regions of the globe in recent decades, and this trend is forecasted to strengthen in the future (IPCC, 2021; Vogel et al., 2021). During a drought, the water and carbon balance of trees is altered by the reduction in soil water content and the increased atmospheric evaporative demand (Bréda et al., 2006; Martin-StPaul et al., 2017; Grossiord et al., 2020). One of the first processes to be impacted is cell growth through the loss of turgor pressure, followed by stomatal closure (Klein et al., 2014a; Lempereur et al., 2015; Peters et al., 2021; Cabon et al., 2022). If water stress continues to increase and sap tension becomes too high, xylem embolism occurs, potentially leading to the mortality of leaves, branches, roots and the whole tree (Cruiziat et al., 2002; Bréda et al., 2006; Choat et al., 2018).

The negative impact of drought on tree growth has been largely documented through the study of tree rings (e.g. Fritts et al., 1965; Anderegg et al., 2015; Babst et al., 2018). Reduced ring-widths are notably associated with high drought stress experienced by the tree (e.g. Vitasse et al., 2019), as well as high mortality risk (Cailleret et al., 2017; DeSoto et al., 2020). The reduction in tree leaf area, a key integrative trait that directly controls transpiration and carbon assimilation, is also well-documented at both tree and stand scales (e.g. Ogaya and Penuelas, 2006; Limousin et al., 2009; Martin-StPaul et al., 2013; Maggard et al., 2016; Bose et al., 2022; Liu et al., 2023). On a finer scale, tree leaf area can be reduced by shifts (1) in leaf lifespan (i.e. delayed leaf emergence and/or leaf shedding), and/or (2) in morphological traits (reduced leaf size, shoot elongation, branching rate; Oleksyn et al., 1997; Borghetti et al., 1998; Ogaya and Penuelas, 2006; Limousin et al., 2012; Grossiord et al., 2017; Liu et al., 2023). These primary growth traits are key because they control the development and architecture of the crown and therefore leaf area in the long term (Bréda et al., 2006; Barthélémy and Caraglio, 2007; Martin-StPaul et al., 2013; Adams et al., 2015). Along with other traits (e.g. sapwood area, carbon storage capacity) primary growth traits govern the so-called ‘legacy effects’ of past climatic conditions on stem radial growth (Zweifel and Sterck, 2018; Kannenberg et al., 2020; Zweifel et al., 2020) and on mortality risk (Sterck et al., 2024). While the relationship between radial growth and climate has received considerable attention (e.g. De Luis et al., 2013; Babst et al., 2019), research on climate effects on the primary growth traits of mature trees is relatively scarce (but see Oleksyn et al., 1997; Limousin et al., 2012). However, the climatic factors that drive primary and secondary growth might show discrepancies due to differences in the phenology of these growth processes (Girard et al., 2012; Magnin et al., 2014). Besides, the primary growth response to drought is usually deducted from measurements performed on a limited number of axes often only sampled in the upper part of the canopy (but see Thabeet et al., 2009; Bose et al., 2022), while intra-canopy variability in response to drought can be considerable and should be taken into account (Borghetti et al., 1998; Girard et al., 2012; Adams et al., 2015; Sperlich et al., 2015).

The impact of drought on tree growth can be investigated by (1) comparing several sites along bioclimatic gradients, (2) monitoring within-site inter-annual variability in drought conditions or (3) implementing rainfall manipulation experiments. The combination of the latter two approaches allows us to assess both short- and long-term drought effects on trees compared to a control established under exactly the same environmental conditions (e.g. climate, topography, soil, stand structure and composition), which are confounding factors when studied along spatial gradients. Even though rainfall exclusion experiments are valuable to assess how drought affects the functioning of forest ecosystems and the plasticity of their response (e.g. Ogaya and Penuelas, 2006; Limousin et al., 2012; Grossiord et al., 2017; Pretzsch et al., 2020), they are often not monitored for long enough to provide long-term data that are essential for identifying the climatic conditions that drive growth (Girard et al., 2012; Sala et al., 2012; Estiarte et al., 2016; Peltier et al., 2018). Indeed, most rainfall exclusion experiments studies in mature forest ecosystems report results after less than 6–7 years of treatment (but see Gavinet et al., 2019; Rowland et al., 2015; Peltier et al., 2023; or Silvestre-Carbonell et al., 2023). Such experiments are also a precious tool to assess the time-lag and persistence of exacerbated drought effects as effects may dampen over time or on the contrary be delayed, and they can be over- or underestimated if studied over short periods (Leuzinger et al., 2011; Saunier et al., 2018; Bose et al., 2022; Liu et al., 2023).

In this study, we explore the response of four primary growth morphological traits and of secondary growth (here at stem level) of mature Pinus halepensis trees subjected to a 30 % rainfall exclusion during 14 years at the Font-Blanche experimental monitoring site (Moreno et al., 2021). This Mediterranean species exhibits high phenotypic plasticity (see de Luis et al., 2013; Voltas et al., 2015) and is capable of producing several growth units in the same growing season (i.e. polycyclic species; Girard et al., 2012; Hover et al., 2017). Primary growth morphological traits (shoot elongation, branching, polycyclism rate and needle length) were monitored from 2009 to 2022 on 20–25 axes at different vertical positions within the canopy per tree on 13 different trees, five of them being subjected to a reduction of rainfall of 30 % resulting from an exclusion treatment. These trees were also cored in 2023 to assess their stem secondary growth at an annual resolution.

We aimed to evaluate the drought sensitivity of P. halepensis above-ground growth and to elucidate the mechanisms involved in its adjustment under long-term drought conditions, based on 14 years of annual growth data. More precisely, our objectives were: (1) to identify the main climatic drivers of primary and secondary growth, which are not necessarily the same; and (2) to assess whether the exclusion treatment would induce a shift in those growth traits (and if the effects were immediate or lagged, and if they persisted over time) and in drought–growth relationships, thus revealing acclimation processes under exacerbated drought conditions (Estiarte et al., 2016).

MATERIAL AND METHODS

Study site of Font-Blanche

The study was conducted at the Font-Blanche long-term experimental site located in southeast France near Marseille (43°14′27″N, 5°40′45″E, alt. 420 m). It is part of the ICOS (Integrated Carbon Observation System) and AnaEE France (Analysis and Experimentations on Ecosystems) networks. It was established in a mixed natural forest of ~70 years, with an upper stratum of P. halepensis (mean height of 13.0 m in 2020), a middle stratum of Quercus ilex (mean height of 5.3 m), and a shrub understorey dominated by Quercus coccifera and Phillyrea latifolia. Stand density and basal area average 1152 stems ha^–1 and 25.0 m² ha^–1, respectively; P. halepensis accounts for less than a third of the number of stems but almost 75 % of the basal area. The soil is rocky and shallow, and the bedrock is a karstified limestone with many fractures and clay inclusions (more details are available in Moreno et al., 2021). Climatic conditions were continuously recorded by a weather station installed in 2007 at the experimental site. The climate is typical Mediterranean, with hot and dry summers contrasting with cool and mild winters. Over the 2008–2022 study period, mean temperature and mean annual rainfall were 14.2 °C and 649 mm, respectively, with high inter-annual variations (e.g. 371 and 1069 mm for the years 2017 and 2014, respectively; Supplementary Data Fig. S1).

A rainfall exclusion experiment was established in winter 2008 on a 25 × 25-m plot, with PVC gutters set at 2 m high covering 30 % of the ground area to exclude ~30 % of the rainfall. This level of exclusion was designed to align with climate models projections from the IPCC (2007) report, which predicted a reduction of ~30 % in rainfall for this region. We acknowledge that lateral water flows can occur from outside to inside the plot, which can dampen the exclusion effect. Instead of trenching at the plot borders, which could damage the trees’ root system, the water excluded by the gutters is collected in a water tank outside the exclusion plot; furthermore, given the extensive root systems that mature Aleppo pine can develop, which probably enables them to access water resources beyond the area covered by the gutters, the monitored trees were selected to be as far as possible from the plot edges to minimize edge effects (mean distance between the monitored trees and the nearest plot edge: 6.6 m). Two control plots of the same area were simultaneously set up, one without gutters and the other with upside-down gutters also covering 30 % of the plot to account for gutter effects on microclimate and albedo (infrastructure control; Fig. 1A–C). Treatments were effective in January 2009. The effectiveness of the rainfall exclusion treatment was controlled by the daily monitoring of soil water content with automatic soil moisture probes (Decagon EC-5 Volumetric Water Content sensors) settled within the first 50 cm of soil of each plot. Since there were no differences in terms of soil water content, sap velocity, water potentials and stem diameter increments in the two control plots (Supplementary Data Fig. S2; see also Moreno et al., 2021), they were grouped together as a single control treatment in the analyses.

Fig. 1.

Experimental design: (A) control plot, (B) control plot with upside-down gutters, and (C) rainfall exclusion plot. (D) Example of one of the monitored Pinus halepensis axes. (E) Schematic representation of the measured morphological primary growth traits on an axis of architectural unit type ‘branch’: shoot length, number of ramifications, polycyclism rate and needle length. See illustrative examples in Supplementary Data Fig. S5.

Primary growth

Since 2009, primary growth morphological traits have been measured on a monthly basis on two to four dominant trees per plot, and 20–25 leafy axes per tree (an axis being defined as a cylindrical organ capped by an apical meristem, encompassing all the elements derived from the activity of this meristem). These axes were labelled, and a permanent scaffolding was set up in each plot to facilitate the monitoring. Because all monitored trees in each plot are accessed by the same scaffolding, the monitored trees were necessarily selected close to each other (average distance between two monitored trees in each plot: 4.7 m, with a minimum at 2.2 m and a maximum at 8.3 m). The selected tree dimensions, age and experienced competition were similar across plots and their canopies did not overlap in such low-density stands (using the Hegyi competition index to quantify the competition experienced by each tree; Supplementary Data Figs S3 and S4). In total, 502 different axes have been monitored on 13 different trees (in 2022, control plot: four trees; control plot with reversed gutters: two trees; exclusion plot: four trees).

For each year and each axis, we measured: (1) the total length of the annual shoot (shoot length, which is a good proxy of the number of needles; Girard et al., 2012; Vennetier et al., 2013; in mm); (2) the number of growth units [polycyclism rate; P. halepensis having the ability to produce multiple growth units per year, a growth unit being defined according to Hallé and Martin (1968) as the segment of an axis developed during an uninterrupted period of extension]; (3) the number of ramifications (lateral branches or buds produced considering all growth units, i.e. branching rate); and (4) the mean length of the needles (needle length; in case of a polycyclic shoot it is calculated by weighting the needle length of each growth unit by the length of the respective unit; in mm; Fig. 1D, E; Supplementary Data Fig. S5).

Considering that drought may have a different impact on branch development depending on its vertical position (e.g. Cano et al., 2013; Sperlich et al., 2015), the monitored axes were selected at different vertical positions within the canopy and then grouped into two classes: high vs. low areas of the living canopy using a threshold at 50 % of its length. Similarly, to account for the fact that the axes have different functions (e.g. space exploration vs. carbon assimilation; Barthélémy and Caraglio, 2007) and are potentially affected differently by climate (e.g. Adams et al., 2015; Borghetti et al., 1998), we sampled axes from different architectural units following the morphological classification developed by Caraglio et al. (2007) based on the polycyclism rate and the branching capacity of each axis. Leafy axes are classified into three elementary architectural unit types: ramlet, branches and twigs. A ramlet is a monocyclic axis that does not produce lateral branches (assimilation function), a branch has a high polycyclism rate (annual mean ≥1.4; exploration for better light conditions) and the other axes are twigs (assimilation and exploration). Polycyclism rate and the number of ramifications considered to establish this classification were determined for each axis regarding its five first measured annual shoots. Since the monitored P. halepensis trees were mature, their annual shoots were mainly monocyclic or bicyclic, and tricyclism was rarely expressed (79.0, 19.7 and 1.3 % of the observed annual shoots, respectively). Thus, we did not distinguish bicyclic from tricyclic annual shoots and grouped them as ‘polycyclic’ annual shoots (Fig. 1E; Supplementary Data Figs S5 and S16).

The monitored axes frequently abort or break due to abiotic (mainly wind gusts) or biotic causes (e.g. the lepidoptera Rhyacionia buoliana or the fungi Crumenulopsis sororia). Therefore, the average monitoring time of an axis was four consecutive years (ranging between 2 and 5 years for 50 % of the axes; min. 1 year, max. 14 years). To keep sampling as consistent as possible over time, the aborted or broken axes were regularly replaced by equivalent axes (i.e. same function and vertical position within the canopy; Supplementary Data Fig. S6). New axes were also regularly added higher up in the canopy as trees grew during this 14-year period (+~2.3 m in height on average) and self-thinning occurred in the lower part of the canopy. Nevertheless, for each year, we only retained annual shoots that were still measured after the end of August for shoot length, number of ramifications and polycyclism rate (when at least 90 % of the axis elongation is completed; Fig. S7a–c). Similarly, only annual shoots that were still monitored after the end of October were considered for the analysis of needle length (when needles have reached at least 90 % of their final size; Fig. S7d). When a new axis was measured, the primary growth traits of the preceding years were assessed retrospectively as far as possible (except for needle length, for which only the —one or two preceding years could be retrieved since needle longevity is of 2–3 years, and no data were available before 2009).

Stem secondary growth

In July 2023, one core was sampled at 1.3 m height for each living tree with a north-east orientation to avoid reaction wood due to the dominant north-west wind (control plots: eight trees, exclusion plot: four trees). For the trees that were already dead, discs were sawed at 1.3 m (control plots: two trees, exclusion plot: one tree, but this latter was unusable due to fast wood decay). Cores and discs were sanded and scanned at 800 dpi, and ring widths (RWs) were measured using the WinDENDRO software (Regent Instruments Inc.). The resulting RW series were cross-dated using the local pointer years defined in Veuillen et al. (2023b) and by calculating correlation coefficients with the reference chronology using the dplR package (Bunn, 2008) from the R software v.4.3.1 (R Core Team, 2019).

Drought characterization

We computed several water deficit indices (WDIs) for the period 2008–2022 to detect those that best explained each of the studied traits. The first set of WDIs are indices widely used in dendroecology, computed directly from climatic data, allowing us to highlight the periods when growth is most sensible to drought (climatic WDIs; e.g. Royo-Navascues et al., 2022; Veuillen et al., 2023a for P. halepensis). However, these indices do not take into account stand and soil characteristics, which are important to explain soil water demand and availability and, thus, the water stress experienced by the trees (e.g. Klein et al., 2014b; Manrique-Alba et al., 2016). Since direct measurements of soil water content have been available only since 2014 and are partly incomplete (see Supplementary Data Fig. S2), we therefore computed a second set of more biologically meaningful WDIs using the SUREAU plant hydraulic model (physiological WDIs; Cochard et al., 2021). ‘Sureau-ecos-V2’, the latest version of the model, was used (see Ruffault et al., 2022).

For the climatic WDIs, we computed climatic water deficit defined as the opposite of the daily water balance (CWD = PET-P; where PET is the daily potential evapotranspiration and P the daily precipitation; in mm) using the Penman–Monteith equation to calculate PET (Penman and Keen, 1948). A positive value of CWD thus depicts a negative water balance, i.e. a water deficit. Following Veuillen et al. (2023a), CWD was then aggregated over all possible monthly intervals considering the current (n) and previous year (n − 1) on a moving temporal window from 1 to 12 months wide. The same methodology was applied to daily temperatures to control for their specific effects on growth. In the rainfall exclusion treatment, daily rainfall was reduced by 30 % while the other parameters were kept unchanged.

For the physiological WDIs, we estimated the daily relative extractable water of the soil [REW; varying between 1 (field capacity) and 0 (residual soil water)] based on the SUREAU model, which uses a set of daily meteorological variables (rainfall, minimum, mean and maximum temperature and relative humidity, global radiation, wind speed), as well as soil, stand and species characteristics (notably, maximum extractable water, Leaf Area Index (LAI), species-specific hydraulic properties; the full description of the required variables is given in Supplementary Data Table S1 and Fig. S8).

The duration, intensity and timing of drought events are important predictors of primary and secondary growth as they can impact meristematic activity (Borghetti et al., 1998; Limousin et al., 2012; Körner, 2015; Hoffmann et al., 2018; Aldea et al., 2022). Therefore, we tested three different physiological WDIs depicting these three components of drought for each year and each treatment, assuming that tree functions are affected by drought when REW drops below a 0.4 threshold (Granier et al., 1999; Fig. S9):

• The first index DUR is the number of days with REW below 0.4, and depicts water stress duration.

• The second index INT cumulates the daily differences between REW and the 0.4 threshold when REW < 0.4. It integrates drought intensity in addition to drought duration.

• The third index EAR is defined as the number of remaining days in the year from the moment when the 0.4 threshold is crossed for the first time (EAR = number of days in the year − first day of the year with REW < 0.4). The higher the EAR index, the earlier the water stress onset.

Among the tested climatic and physiological WDIs, DUR and INT, CWDfeb-june_n, CWDnov_n-1-sept, and to a lesser extent CWDmarch, were significantly interrelated (Supplementary Data Fig. S10). Drought duration and intensity were higher, and drought occurred earlier in the exclusion treatment each year after its establishment (see Fig. S1 for more details).

Statistical analysis

To investigate the relationship between primary and secondary growth and drought and their response to the 30 % rainfall exclusion treatment, we proceeded in two steps. First, we built mixed-effect models for each year separately to predict the value of each trait in a given year in each treatment accounting for intra- and inter-individual variability among the monitored trees (‘Primary growth’ and ‘Stem secondary growth’ below). Second, we used the predictions from those annual models to assess (1) climate–growth relationships (‘Growth response to inter-annual variations in water deficit’); (2) the treatment effect on the observed trends; and (3) the potential change in the treatment effect between years, i.e. to assess whether the treatment effect was immediate or lagged; if it persisted over time or if there was any acclimation (‘Quantification of the treatment effect’).

Primary growth

For each morphological trait (shoot length, the number of ramifications, polycyclism rate and needle length) and each year, we built generalized linear mixed-effect models at the axis level using the lme4 R package (Bates et al., 2015), assuming a normal distribution for needle length and shoot length, a binomial distribution for polycyclism rate, and a Poisson distribution for the number of ramifications:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{M}\mathrm{o}\mathrm{d}\mathrm{e}\mathrm{l}1:Y_{ijklm}=\mu +Treatmen{t}_i+A{U}_j+H_k+{\gamma }_{ij}+{\omega }_{ik}+{\delta }_l+{\epsilon }_{ijklm}}\end{array}}$$

which includes treatment (i ϵ[1, 2]; control or exclusion), architectural unit (AU, j ϵ[1:3]; branch, twig or ramlet) and vertical position within the canopy (H, k ϵ[1,2]; high or low) as fixed effects. The interactions between treatment and AU (${\gamma }_{ij}$) and between treatment and H (${\omega }_{ik}$) were also included to control for different treatment effects according to the axis function and environment. The inter-individual variability in primary growth rate among the monitored trees was accounted for by considering the tree as a random effect on the intercept (δ_l; l ϵ[1:13]) – but not on the slope, assuming that the effects of the architectural unit and vertical position of the axis were similar among trees. Here ${\epsilon }_{ijklm}$ stands for the residual error of the model. By using mixed-effects models including tree as a random effect, we could account for the pseudoreplication inherent of such a rainfall exclusion experiment (Zhang et al., 2009; Davies and Gray, 2015).

Parameters of model 1 were fitted using the restricted maximum likelihood method (Bates et al., 2015). Concerning shoot length, a natural logarithm transformation was applied to satisfy the assumption of normality of the residuals. Polycyclism rate and the number of ramifications were only analysed for branches and twigs, as ramlets never express polycyclism, and very occasionally lateral shoots. We estimated marginal and conditional R², depicting the variance explained by the fixed effect only and by both fixed and random effects, respectively, using the MuMIn R package (Barton, 2024). Then, we performed Tukey’s post-hoc tests to assess treatment effects for all axes but also for each architectural unit and vertical position within the canopy.

Finally, based on the models built for each trait and each year, marginal means and associated 95 % confidence intervals were estimated for the control and exclusion plots by averaging over the other covariates using the emmeans R package (applying a back-transformation for shoot length; Length, 2024). In this way, the intra- and inter-individual variability among the monitored trees was removed – but still considered in the parametrization of the models. Given the proximity between the monitored trees required by the experimental design (see ‘Primary growth’), Moran tests were performed on the model residuals to ensure that the models correctly accounted for potential spatial autocorrelation (Dales and Fortin, 2002). The potential effect of the competition experienced by each tree on its primary growth traits was assessed with linear relationships between the random part of the models δ_l and Hegyi competition index.

Stem secondary growth

Tree-ring widths (mm) were converted into basal area increment (BAI, mm²) as it is less dependent on the size-related geometric effects (Biondi, 1999) and better represents biomass increment (Bouriaud et al., 2005). To control for an age effect on tree radial growth, tree-ring widths were also converted into ring-width indices (RWIs) using the regional curve standardization (RCS) method (Esper et al., 2003; Nehrbass-Ahles et al., 2014). To build this regional curve, we used ring-width data from 78 sites located in south-east France that have been recently compiled by Veuillen et al. (2023a). Since the results were very similar (the monitored trees showing homogeneous ages and dimensions; see Supplementary Data Fig. S3 and S11), we only showed the results for BAI. BAI series were averaged into treatment-scale chronologies using a Tukey bi-weight robust mean (control: ten trees, exclusion: four trees). The potential effect of competition experienced by each tree on its radial growth was assessed with linear relationships between BAI and Hegyi competition index.

Growth response to inter-annual variations in water deficit

To assess the response of growth traits to inter-annual variations in water deficit, we used (1) marginal means predicted for each treatment by the mixed models described above for primary growth traits; and (2) BAI chronologies of each treatment for stem secondary growth, for each year of the 2009–2022 period.

We first explored the temporal trends of the primary and secondary growth traits and the correlations between them using a principal component analysis (PCA) approach. Then, to explore the relationship between each trait and the different climatic and physiological WDIs as well as temperature, we computed bootstrapped Pearson correlation coefficients (r; n = 1000). The climatic conditions of the previous year (n − 1) are known to strongly influence growth (e.g. Girard et al., 2012; Limousin et al., 2012; Peltier, Barber and Ogle, 2018), and were thus considered in the calculation of these correlations.

Quantification of the treatment effect

The WDIs showing the highest significant (at the 0.05 threshold) correlation coefficients with each growth trait were retained to assess whether the treatment induced a shift in the WDI–growth relationships. More precisely, we explored whether there was a shift in the slope (i.e. in sensitivity to drought) or in the intercept (i.e. growth performance under similar growth level) of the relationship following Estiarte et al. (2016). We developed linear models for each trait, focusing on the interaction between WDI and treatment:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{M}\mathrm{o}\mathrm{d}\mathrm{e}\mathrm{l}2:Y_{ik}=\mu +{\alpha }_k+(\beta +{\gamma }_k)WD{I}_{ik}+{\epsilon }_{ik}}\end{array}}$$

where i is the year (we tested both i ϵ[2009,2022] and i ϵ[2010,2022] since some primary growth traits predetermined during organogenesis processes might not be impacted on the very first year of treatment; Girard et al., 2012; Hover et al., 2017); k is the treatment (control or exclusion), ${\alpha }_k$ is the effect of the treatment, $\beta$ is the effect of the WDI, and ${\gamma }_k$ is the differentiated effect of the treatment on the relationship between the trait and WDI. If the interaction between WDI and treatment was not significant (i.e. ${\gamma }_k$ not significantly different from 0), we removed the interaction term from model 2. Then, we used the resulting model to assess if there was a treatment effect on the intercept of the WDI–growth relationship (i.e. ${\alpha }_k$ significantly different from 0).

Finally, we aimed to assess the potential change in the treatment effect over the years. For this purpose, for each primary growth trait, the results from the models (1) were used to determine if there was a significant difference among treatments for each year. The difference among treatments over the entire 2009–2022 period was also tested with a paired Wilcoxon rank-sum test on the marginal means predicted for each treatment. Concerning secondary growth, since we only have one sample per monitored tree, the sample size was too small to run parametric tests as we did for primary growth traits (see ‘Primary growth’ above). Thus, we performed Wilcoxon rank-sum tests for each year (unpaired tests), but also over the entire 2009–2022 period (paired test), to determine the existence of a significant difference in BAI among treatments.

RESULTS

Relationships between the studied primary and secondary growth traits, and time

Shoot length, the number of ramifications, polycyclism rate and BAI were all significantly and positively correlated, with the exception of needle length (Supplementary Data Figs S12 and S13). All primary and secondary growth traits except needle length decreased significantly over time (including RWI in secondary growth traits; Fig. S13), with shoot length and the number of ramifications showing a strong reduction in the last 3 years (2020–2022; Fig. 2A, B). It is noteworthy that drought duration and intensity also increased with time from 2009 to 2022 (Fig. S10).

Fig. 2.

Temporal change in (A) shoot length, (B) number of ramifications, (C) polycyclism rate, (D) needle length and (E) BAI averaged for the trees growing in the control plots (green) and exclusion plot (red). The dashed vertical line indicates 2009, the year of establishment of the rainfall manipulative experiment. For each of the primary growth traits and treatment, marginal means and their 95 % confidence intervals were estimated from the annual models (model 1) with all other covariables fixed at their means; *significant differences between treatments based on model 1 (P < 0.05; Supplementary Data Tables S2–S5). More details of the distribution of each primary growth trait over time can be found in Figs S13–S16. For secondary growth, the BAI of the monitored trees (dashed lines) and the resulting BAI chronologies by treatment (solid lines) are shown (E). *A significant difference between the two treatments according to the non-parametric Wilcoxon test (P < 0.05).

There was considerable variability in needle length among monitored trees, as suggested by the large proportion of the variance explained by the random part of the model (i.e. high R²_c compared to R²_m values, especially at the end of the study period; Supplementary Data Table S5). This inter-individual variability among the monitored trees was lower for the other primary growth traits (Tables S2–S4). We carefully checked that potential spatial autocorrelation arising from constraints imposed by the experimental design did not lead to results appearing artificially more significant than warranted by the data (see Table S6).

Competition had a significant and negative effect on BAI and on the random part of the models for each of the four primary growth traits, but these effects were slight (R² = 0.14 for BAI in Supplementary Data Fig. S19; and R² < 0.05 for the other traits in Fig. S18), meaning that competition explained only a very small part of the remaining inter-individual variability.

Growth trait response to inter-annual variations in water stress

The climatic WDIs that best explained growth traits were generally similar among treatments, in terms of index type and temporal window (Figs 3; Supplementary Data Figs S18 and S19; Table S7). The marginal differences in correlation coefficients and significance of the climate–growth relationships observed among treatments were hardly interpretable, and were probably associated with noise caused by the shortness of the time-series and the small number of trees sampled.

Fig. 3.

Bootstrapped Pearson correlation coefficients and confidence intervals (n = 1000) between each of the growth traits and water deficit indices of the current and previous year (n − 1) in the control (upper pannels) and in the exclusion (lower pannels). The CWD windows showing the strongest and most consistent correlations with the studied traits (Supplementary Data Fig. S19) were selected to be compared with the physiological water stress indices DUR, INT and EAR of the current and previous year. Colours indicate significant correlations (P < 0.05).

All growth traits were negatively correlated with climatic WDIs as well as drought duration, intensity and earliness of both the previous and current year (DUR_n-1, INT_n-1, EAR_n-1 DUR, INT and EAR) and mean temperature, except needle length, whose correlations with drought duration, intensity and earliness of the previous year (DUR_n-1, INT_n-1, EAR_n-1) as well as CWD of the previous spring–summer were unexpectedly positive (Fig. 3; Supplementary Data Figs S18 and S19).

Generally, the physiological WDIs performed better than the climatic WDIs (i.e. higher absolute correlation coefficient values; Fig. 3). Specifically, drought duration of the previous and of the current year best explained shoot length [e.g. in the control plots, Pearson correlation coefficients (r) of −0.71 and −0.55, respectively) and the number of ramifications (r = −0.77 and −0.49, respectively). Drought intensity of the previous year was also strongly correlated with shoot length and the number of ramifications (r = −0.60 and −0.69, respectively). To a lesser degree, both variables were also negatively correlated with CWDjune_n-1-june and similar windows (Supplementary Data Fig. S20), and the number of ramifications was negatively related to the mean temperature in the same period (Fig. S21).

Similarly, drought duration of the previous year was the main driver of the polycyclism rate (r around −0.58 in both treatments; Fig. 3), and drought duration of the current year best explained inter-annual variations in BAI (r = −0.67 and −0.90 in the control and exclusion plots, respectively; Fig. 3). Drought intensity of the previous and current year were also strongly correlated with polycyclism rate and BAI, respectively (but slightly less than drought duration; Fig. 3). BAI was also negatively impacted by the mean climatic conditions from November of the previous year to September (CWDnov_n-1-sept_;  r = −0.69; Supplementary Data Fig. S20) and mean temperature (especially in winter; Fig. S21). Finally, in contrast to the other variables, needle length was not best predicted by physiological WDIs, but was instead negatively associated with water deficit during the current spring, especially in March (r = −0.71 and −0.62 for the control and exclusion plots, respectively), and positively associated with CWD in late spring and early summer of the previous year, especially in June (r = of 0.75 and 0.84 in the control and exclusion plots, respectively; Fig. S20).

Treatment effect on primary and secondary growth traits

After considering the variation in sample characteristics (number, vertical position and type of axes sampled) over time and among plots, and for differences in growth rates among trees, we did not find strong effects of the rainfall exclusion experiment on primary growth. When considering the entire 2009–2022 study period, shoot and needle length marginal means estimated from the models were significantly lower in the exclusion plot (7.7 and 14.3 % lower; P < 0.05 and P < 0.001, respectively; paired Wilcoxon test) while there was no significant effect on the number of ramifications and polycyclism rate (P > 0.05).

The rainfall exclusion treatment did not induce any shift in drought sensitivity for any of the studied traits (same slope of the drought–growth relationships; focusing on the WDIs that best predict each trait; Fig. 4). For a similar level of water stress, the number of ramifications was higher in the exclusion than in the control (significant treatment effect on the intercept in the 2009–2022 period; +0.24; Supplementary Data Fig. S22). However, when removing the very first year of treatment from the analysis (considering that organogenesis processes take place in the bud in the preceding year; see ‘Quantification of the treatment effect’the treatment effect was no longer significant (Fig. 4). On the contrary, needles were shorter in the rainfall exclusion plot for a similar level of water stress, including the year 2009 or not (−5.6 mm; Fig. 4; Fig. S22).

Fig. 4.

Relationships between each growth trait (marginal means estimated from model 1 and BAI chronologies for primary and secondary growth, respectively) and the WDI that predicts it best. Results are shown without the first year of the rainfall exclusion experiment (see ‘Quantification of the treatment effect’; Supplementary Data Fig. S20). The effect of the treatment on the intercept of the relationship was significant for needle length (P < 0.05), but not for shoot length, number of ramifications, polycyclism rate or BAI (a single regression line was thus displayed in grey).

Before treatments were effective in 2009, there was no significant difference between treatments for shoot length, number of ramifications and polycyclism rate (P > 0.05; data not available for needle length; see ‘Primary growth’ in the Methods). BAI was similar across plots before 1997; but was lower in the exclusion plot in the 1997–2008 period, although this difference was not always significant (Fig. 2).

The year-by-year analysis did not reveal significant differences in shoot length across plots, except in 2011 when shoot length was significantly lower in the exclusion than in the control plots – especially for branches located in the lower part of the canopy (Fig. 2A; Supplementary Data Table S2). Similarly, the number of ramifications and polycyclism rate were similar across plots, except in 2010 for which the number of ramifications was significantly lower in the exclusion plot (Fig. 2B & C; Table S3 and S4). Even though we have no information on needle length before the establishment of the experiment, the effect of the rainfall exclusion treatment on this trait seems to be immediate: needles were significantly shorter in the exclusion plot immediately following rainfall exclusion establishment, from 2009 to 2011. In the following years, they were still shorter although the difference was no longer significant and tended to decline over time (e.g. dropped to 0.6 % in 2022; Fig. 2D). Post-hoc analyses revealed that the difference in needle length between plots was significant for all axes (in 2009 and 2011), mainly for ramlets (in 2010; Table S5). Finally, BAI was lower in exclusion than in control plots from 2009 to 2015 (only significant in 2010); however, since it was already the case before 2009, there was no clear effect of treatment on stem secondary growth.

DISCUSSION

Based on 14 years of monitoring in an in situ forest experiment, we found that P. halepensis growth was strongly and negatively impacted by drought duration during the current year (in the case of stem secondary growth), the previous year (polycyclism rate), both years (shoot length and number of ramifications) and by climatic water deficit during spring (needle length). Excluding 30 % of the incoming rainfall did not affect the number of ramifications, polycyclism rate and stem radial growth, but reduced needle length and shoot length by 14.3 and 7.7 % on the entire study period, respectively. However, this effect was mainly significant in the first few years after the treatment was established, suggesting that the treatment effect declined over time. In line with our expectations, the treatment did not induce a shift in growth sensitivity to drought; however, under a similar level of drought stress, needles were slightly shorter in the exclusion.

Growth decrease over time

There was a decreasing trend in all primary growth morphological traits except needle length in both plots (Fig. 2; Supplementary Data Fig. S12). This can be partly explained by the ontogenetic ‘axis drift’: all axes end up with the same characteristics, i.e. short, unbranched and monocyclic annual shoots, regardless of their initial function (Barthélémy et al., 1997; Barthélémy and Caraglio, 2007). This decreasing growth trend could also be due to the increase in drought duration and intensity over the study period, which makes it difficult to disentangle ontogenetic and drought effects on primary growth. Since the branches were not cut and the exact ontogenetic age of the axes is not known, we cannot control for age effects on shoot growth in the same way as we did for secondary growth. Yet, the decline in shoot length, number of ramifications and polycyclism rate that occurred in both control and exclusion in the last 2 years of the experiment (2020–2022) seems too strong to be explained by ontogeny and matches with the 2019–2022 dry period, suggesting a drought-induced decline in primary growth (Fig. 2). A decreasing trend in secondary growth was also found both for BAI and RWI (where the age effect on growth is removed; see ‘Stem secondary growth’ and Fig. S13), which might be an early sign of drought-induced stand decline in both plots (e.g. as shown by a study of tree-ring and defoliation in three natural stands following a severe drought in Spain; Camarero et al., 2015).

Climatic drivers of primary and secondary growth

Our results show that primary growth morphological traits are strongly associated with the climatic conditions of both the current and the previous year, which can be explained by the timing of shoot and leaf genesis and development (Cochard et al., 2005; Barthélémy and Caraglio, 2007; Gonzalez et al., 2012). Indeed, P. halepensis shoot organogenesis, which determines the number of growth units and needle primordia (and thus a large part of shoot length) as well as the number of lateral shoots, occurs in the summer and autumn of the preceding year (Girard et al., 2012; Hover et al., 2017).

Of all the climatic and physiological WDIs tested, the physiologically based drought duration indicator (DUR) better explained shoot length and BAI than the indices depicting drought intensity, i.e. INT and the climatic WDIs commonly used in dendroecology (Fig. 3). This result has already been observed on P. halepensis secondary growth (Helluy et al., 2020), and can be explained by the fact that growth is mainly a sink-limited process under water limitation as cell growth ceases when the turgor loss threshold is reached, independently of the drought intensity, and before photosynthesis cessation (Körner, 2015; Lempereur et al., 2015). Drought duration also best explained polycyclism rate and the number of ramifications that were prepared in the following year, but the underlying mechanisms of how drought impacts these predetermined traits is not well understood. The fact that the number of ramifications was also significantly correlated with drought conditions in the current year (Supplementary Data Fig. S22) was already reported but not further discussed by Girard et al. (2012), who assessed P. halepensis primary growth on the site of Font-Blanche before the rainfall manipulation experiment was established. One plausible explanation is that ramifications preformed in the bud in year n − 1 can stay dormant or abort in year n in case of severe drought, as observed for several Mediterranean coniferous species including P. halepensis in south-eastern France (Vennetier et al., 2013).

In addition to drought duration, BAI of P. halepensis was also strongly and negatively correlated with CWDnov_n-1-sept, matching with the period of importance for its secondary growth as reported under similar Mediterranean bioclimates by large-scale studies of climate–radial growth relationships (de Luis et al., 2013; Veuillen et al., 2023a). It was also negatively correlated with temperature during this period (Tnov_n-1-sept), which might increase evapotranspiration and thus have a negative effect on water availability. Unexpectedly, we also found a negative effect of winter temperatures Tnov_n-1-feb on secondary growth (Supplementary Data Fig. S21), while a positive correlation was hypothesized for this evergreen species which is able to assimilate carbon during winter (Gea-Izquierdo et al., 2015; Sperlich et al., 2019). This might reflect a non-causal correlation due to the limited length of the time-series, to the correlation among mean temperatures computed over different periods, or because warm years were also dry ones (see Fig. S21). Moreover, it is rather unlikely that winter temperatures significantly limited growth at our study site, as they did not drop below 4.5 °C over the study period with a lack of severe cold spells (except in February 2012; Fig. S1b; mean winter temperatures ranging between −2 and 6 °C on the species distribution area, Rameau et al., 2008). On the contrary, temperatures were weakly and not consistently negatively correlated with primary growth traits (Fig. S21). It is interesting to note that the negative effect of temperature on primary growth was much more pronounced from 1995 to 2010 (Girard et al., 2012). Thus, it appears that primary growth is less and less responsive to temperature and increasingly more responsive to drought at our study site (as has been demonstrated globally for radial growth; Babst et al., 2019). We therefore conclude that P. halepensis above-ground growth was mainly limited by drought during the 2009–2022 period at our study site.

Needle length was also negatively associated with drought duration, but was best explained by CWDmarch, which depicts drought intensity during the early stages of leaf development. This process starts in spring with cell division, the resulting number of cells being a major determinant of final leaf length (Miyazawa et al., 2003; Pantin et al., 2011; Gonzalez et al., 2012). In the early stages of their development, young leaves are strongly dependent on carbon availability (Pantin et al., 2012, 2011) mainly accumulated by mature leaves, which exhibit high photosynthetic activity during this period for P. halepensis (Borghetti et al., 1998; Maseyk et al., 2008). This might explain why spring drought conditions are so important in explaining final needle length. Polycyclism rate is best explained by the climatic conditions during late winter–early spring of the previous year (also reported by Girard et al., 2012), which is too early to match with organogenesis processes (occurring in August–September and October–November for the first and second growth units, respectively; Hover et al., 2017). In the same way, needle length was unexpectedly positively correlated with climatic water deficit in the previous June (Fig. 3).

Therefore, primary growth processes seem to be not only dependent on the climatic conditions during organogenesis and elongation (i.e. atmospheric evaporative demand), but also on tree requirements in terms of water resources, which depend on its leaf area (i.e. water demand of the tree). Following an integrative conceptual scheme, we hypothesize that favourable conditions in year n − 1 would result in long needles in n − 1, but also in high polycyclism rate, number of ramifications and shoot length (i.e. more numerous needles) in year n. If the conditions in year n are not favourable, the water demand may be too high compared to the level of available water, and the tree may experience water stress earlier and for a longer period (‘structural overshoot’; Jump et al., 2017; Zhang et al., 2021), resulting in an earlier cessation of shoot elongation and/or shorter needles. This continuous adjustment of primary growth thus allows the balance between leaf area and available water to be restored when it is disturbed, in line with the theory of hydro-ecological equilibrium (Eagleson, 1982; Limousin et al., 2009). We therefore suggest that primary growth controlling processes are highly complex and depend on (1) climatic water deficit during organogenesis and elongation and (2) tree leaf area (as a result of past architectural development, number of needle-bearing axes, needle-bearing shoot length, needle length and lifespan; Sala et al., 2012; Bose et al., 2022), (3) together with many genetic, hormonal and other environmental factors (Gonzalez et al., 2012; Körner, 2015). These ‘carry-over’ effects have been largely reported for stem secondary growth (Zweifel and Sterck, 2018), but this study represents one of the few demonstrations of carry-over effects for primary growth.

Effects of the 30 % rainfall exclusion treatment on growth traits

The observed reduction in shoot and needle length due to rainfall exclusion is a common consequence of drought, and was also reported for P. halepensis thanks to a 1-year total rainfall exclusion (Borghetti et al., 1998) and for Pinus edulis experiencing a 45 % rainfall exclusion (Adams et al., 2015; Grossiord et al., 2017; Guérin et al., 2018). In addition, there was a weak, 1-year lagged and non-persistent negative effect of the exclusion treatment on the number of ramifications. Since we found no significant treatment effect on polycyclism, the acclimation to exacerbated drought conditions through a reduction in leaf area in the rainfall exclusion plot at the tree and stand level (described in Moreno et al., 2021) appears to be mainly the result of a decrease in the length and number of needles (shoot length being a proxy of the number of needles). Focusing on each year separately, the treatment effect was significant only for the first 1–3 years (depending on the considered trait). This is in accordance with the long-term monitoring of Pinus sylvestris in Pfynwald (Switzerland), where the positive effect of 2-fold increase in incoming rainfall through irrigation on shoot and needle length disappeared after a few years, suggesting an immediate but only temporary treatment effect on growth (Bose et al., 2022).

This decreasing effect of the treatment over time might be attributed to acclimation processes, such as increased allocation to the root system, which could improve the access to deeper water pools and thus mitigate the treatment effect, as demonstrated by Liu et al. (2023) on a 7-year 50–70 % rainfall exclusion experiment. It may also result from treatment-induced changes in leaf phenology and/or axis mortality, which were not assessed in this study (Limousin et al., 2012; Song et al., 2022). For instance, in the exclusion treatment, the reduced length and number of needles in the first years could be accompanied in the long term by a shorter leaf lifespan and/or an increased branch mortality rate compared to the control. This would result in a smaller leaf area at the tree level, thereby decreasing overall water demand. This hypothesis is supported by the relative decrease in stand-level leaf area observed in the exclusion treatment compared to the control in our study site (Moreno et al., 2021), and could be validated by integrating axis-level data into structure–function models that explicitly simulate plant architectural development (e.g. Griffon and de Coligny, 2014).

Part of these results differ from rainfall exclusion experiments carried out on species with different strategies to face drought. For instance, Quercus ilex showed a lower number of ramifications, but without change in shoot length in a 27 % rainfall exclusion (Limousin et al., 2012). They may also differ when the rainfall exclusion is stronger. Indeed, stem secondary growth was not impacted by an artificial reduction in 27–30 % rainfall for Quercus ilex (Limousin et al., 2012; Moreno et al., 2021). Yet, immediate decreases in secondary growth were reported for Fagus sylvatica and Picea abies (for the 3–5 years following the onset of the treatment, respectively; 70 % rainfall exclusion; Pretzsch et al., 2020); and a delayed decrease was reported for Quercus pubescens (after 5 years of 60 % rainfall exclusion; Saunier et al., 2018) and Quercus aliena (after 7 years of 50–70 % rainfall exclusion; Liu et al., 2023). Borghetti et al. (1998) even found a 90 % reduction in P. halepensis secondary growth after a year of total rainfall exclusion. Thus, it seems that the 30 % rainfall exclusion set up at the Font-Blanche experimental site was not sufficient to lead to a consistent and noticeable decrease in stem secondary growth, despite the trees exhibiting lower water potentials and sap velocity under the exclusion treatment (see Moreno et al., 2021).

It is worth noting that the fraction of rainfall that ran down the trunks and into the soil was not affected by the gutters. However, it was only a marginal fraction of the total rainfall (~0.9 % for P. halepensis at our study site; 2.8 % for Pinus sp., Barbier et al., 2009) and thus only slightly mitigated the treatment effect.

In addition, capturing the effect of rainfall exclusion on primary and secondary growth in in situ experiments is challenging as they may be affected by many other uncontrolled biotic and abiotic factors, and because of the high heterogeneity in the local environmental conditions, especially in mixed forests (see also Bose et al., 2022). Notably, even though we controlled for inter-individual variability by including individual trees as random effects in the mixed models, the treatment effect might have been partly occluded by the high vertical and horizontal heterogeneity in water availability in karstic soils (e.g. localized presence of cracks providing differentiated access to underground water resources; Carrière et al., 2020). Furthermore, adult P. halepensis develop large and deep root systems possibly enabling them to access resources both horizontally beyond the exclusion plot borders and vertically in deeper soil layers which may be less affected by the treatment. This could help explain the limited treatment effect. Genetic variation in growth performance and drought sensibility among the monitored trees may also explain this result (Moran et al., 2017), all the more as the number of monitored trees was low.

Finally, although the 30 % rainfall diminution was consistent over time, the resulting relative increase in drought duration differed among years, being more pronounced when the natural annual drought was shorter (ranging between +22 d, in 2019, to +95 d, in 2013; corresponding to a 16–211 % drought duration increase; Supplementary Data Fig. S23). In other words, the difference in drought duration between the rainfall exclusion and the control was smaller in drier years. Moreover, the range of the yearly relative difference in drought duration between the rainfall exclusion and the control was weaker than the inter-annual variations in drought duration in each plot (between 37 and 172 d over the 2009–2022 period; 365 % variation; Fig. S1e). Finally, even if the rainfall exclusion efficiently increases soil moisture deficit, this experimental setup is inefficient in reproducing the atmospheric effects of drought (cf. Rowland et al., 2015). These three aspects might explain why the effects of drought on primary and secondary growth traits were much lower when quantified spatially among plots than temporally among years (strength of drought–growth relationships; Figs 3 and 4). This attenuation of the treatment efficacy is probably further exacerbated by the experiment being situated within a Mediterranean climate characterized by pronounced inter-annual variability in precipitation.

Towards a better assessment and understanding of drought sensitivity

Only the relationship between drought and needle length was consistently modulated by the treatment, but only slightly (Fig. 4; everything else being equal, the needles were 5.6 mm shorter in the exclusion plot, corresponding to a 10 % decrease on the intercept). This is consistent with the lack of divergence in the relationship between drought and net primary productivity observed in most rainfall exclusion treatments (see the analysis of 11 treatments excluding 7–58 % of the rainfall by Estiarte et al., 2016; and the results from a 27 % rainfall exclusion treatment in Gavinetet al., 2019), even though this could be observed for other traits such as the net CO₂ assimilation rate (in a 45 % rainfall exclusion treatment; Limousin et al., 2013). To further explore acclimation processes, e.g. changes in drought–growth relationships with increasing site aridity, we would need to analyse them (1) in more severe drought treatments (Meir et al., 2015; Estiarte et al., 2016), and (2) over a longer period to ensure the statistical reliability of the trends observed, and to capture the dampening or increasing treatment effects over time (cf. Leuzinger et al., 2011; Bose et al., 2022; vs. Gavinet et al., 2019). We also acknowledge that these results were obtained from a specific site, in an experiment without true replicates, and with a limited number of trees, given the strong constraints associated with (1) setting up rainfall manipulation experiments (Asbjornsen et al., 2018) and (2) manually monitoring primary growth monthly over 14 years.

To improve our understanding of how drought impacts primary and secondary growth processes, it would be relevant to search for a more biologically meaningful soil drought threshold for growth limitation. Indeed, the 0.4 REW threshold used here is associated with the onset of stomatal closure and decrease in sap-flow in P. halepensis and many temperate species (Granier et al., 2007; Maseyk et al., 2008), but may not be linked to meristem activity. This could be achieved by coupling predawn water potential and microdendrometer measurements throughout the year such as in Lempereur et al. (2015). Moreover, accounting for below-ground (Klein et al., 2011; Liu et al., 2023), non-structural (Sarris et al., 2013) and reproductive (Girard et al., 2012; Vennetier et al., 2013) carbon sinks, as well as leaf phenology (Limousin et al., 2012) could modulate our results concerning growth sensitivity to drought and would improve our understanding of the whole carbon allocation processes under drought conditions.

SUPPLEMENTARY DATA

Supplementary data are available at Annals of Botany online and consist of the following.

Fig. S1: temporal changes in meteorological conditions over the 2008–2022 period. Fig. S2: Soil humidity differences between treatments over the 2014–2023 period. Fig. S3: Diameter at breast height and age of the monitored trees. Fig. S4: Competition experienced by the monitored trees. Fig. S5: Illustrative examples of annual shoots. Fig. S6: Sampling characteristics. Fig. S7: Timing of completion of annual growth. Fig. S8: Vegetation and soil parameters used to run SUREAU simulations. Fig. S9: Calculation of the physiological water deficit indices (WDIs). Fig. S10: Relationships between the water deficit indices and time. Fig. S11: RW, BAI and RWI chronologies. Fig. S12: Relationships between the primary and secondary growth traits, and time. Fig. S13: Pairwise correlations between the studied traits and time. Fig. S14: Characteristics of the shoot length data. Fig. S15: Characteristics of the number of ramifications data. Fig. S16: Characteristics of the polycyclism data. Fig. S17: Characteristics of the needle length data. Fig. S18: Competition effect on primary growth. Fig. S19: Competition effect on secondary growth. Fig. S20: Correlations between each growth trait and climatic water deficit (CWD) computed over various timeframes. Fig. S21: Correlations between each of the studied traits and mean temperature averaged over various timeframes. Fig. S22: Relationships between each growth trait and the WDI that predicts it best estimated for the 2009–2022 period. Fig. S23: Drought duration in the rainfall exclusion relative to the control. Table S1: Soil parameters. Table S2: Summary of the models for shoot length. Table S3: Summary of the models for number of ramifications. Table S4: Summary of the models for polycyclism. Table S5: Summary of the models for needle length. Table S6: Results of the Moran tests performed on the mixed model residuals. Table S7: Pearson correlation coefficients with the best CWD windows.