The response of forests to the forecasted increase in climate stress occurrence is considered a key issue in climate change scenarios . Although forest productivity increased in most ecosystems during the 20th century [2,3], a review by Allen
The consequences of climatic events on forest health can be immediate but are often delayed up to 5 to 10 years [5,10], and may be significant for decades and sometimes irreversible on tree growth . Recent studies on tree architectural development and primary growth suggested that the long lasting impact of repeated droughts on tree crown development could be one of the causes of these delayed effects [12-14].
Primary growth corresponds to the creation of new tissues outside existing organs, and includes bole, branch and root length growth, branching (birth of new branches or roots), creation and growth of leaves or needles and rootlets, flowering and fruiting [15,16]. It is therefore fully linked to plant architectural development patterns and processes. In contrast, secondary growth for trees corresponds to the radial growth of existing branches, bole and roots. In single trees and forests, although secondary growth usually exceeds primary growth and leaf production , the total amount of biomass allocated to primary growth may be very important . As an example, leaf and fruit production measured through litter bags in French adult broadleaved stands (mainly beech and oak) reached 20 to 40% of wood production by stems . The relative allocation to new shoots was not correctly assessed up to now but it can be inferred from leaf production and the leaf mass fraction (LMF), ratio between leaves and whole twig mass. In a recent study, twig wood/leaf biomass ratio was found to reach from 25 to 50% and from 50 to 75% during respectively dry and humid years for
The mediterranean climate is characterized by high temperatures associated with low rainfall in summer, drought being the main environmental constraint for vegetation growth . For the 21st century, climatic models forecast that the Mediterranean basin will be prone to a faster warming than most other continental areas over the world, associated with a reduction of rainfall during the growth season [1,25]. Therefore, this area is a good place to detect and model any climate change impact on vegetation, all the more since a rapid decline in precipitation and higher temperatures were already noticeable in parts of this basin [26,27]. In Southeastern France, the period 1998-2007 was characterized by mean annual temperature and mean summer temperature 0.9°C and 1.3°C above the 30-year average (figure 1) Moreover, eight to ten of the twelve hottest years since 1850 were recorded during this time lapse  which give a foretaste of the climate forecasted for the next decades. In addition strong climatic events also recently occurred such as the 2003 heat
wave which significantly impacted French Mediterranean forests as well as most of Europe . Such extreme climatic events are likely to become frequent with global warming [1,30]. Scorching heat had direct and delayed negative effects on tree growth, especially on pine species . The resulting increase in summer and spring water stress may reduce tree growth in Mediterranean areas [32-35]. Raising temperature may also lead to phenological lags , particularly in the beginning and end of the growth season, with direct consequences on some primary growth processes and architectural development such as polycyclisme and branching rates [37,38].
This study aimed at quantifying the influence of recent climatic trend and events, particularly intense heat and drought, on the primary growth and architectural development of six conifers and one broadleaved species growing in Mediterranean plains (
2. Material and methods
2.1. Study area and species
The study area included 8 sites distributed between 80 and 1400 m of elevation, from the coast to hinterland mountains of Provence-Alpes-Côte d'Azur region in south-eastern France (figure 2).
Nearly 5150 twigs from 1050 branches of 210 trees and 7 species were sampled between 2005 and 2011 (Table 1). Their architectural development was retrospectively measured from morphological markers (figure 3) over a period of 15 to 25 years. For each tree, the sampling design considered separately three thirds of the crown (top, middle and base), branch orientation (north and south), but also branch hierarchy (principal or secondary axis – figure 4) and branch vigor. Secondary axes were chosen according to their relative vigor (strong vs weak axes) within the branch they belong to (figure 4). Twig absolute vigor for each species was later split in three groups of equal number (vigorous, medium, frail) according to their total length growth in the last three years before sampling date.
|Site name||Species (number of trees)||Altitude (m)||exposition||Dates of mesures|
|Saint Mitre||PH (11) QI (6)||80||Flat||2008-2009|
|Font-Blanche||PH (58) QI (34)||420||Flat||2005-2011|
|Siou Blanc||PH (11) QI (6)||650||Flat||2008-2009|
|Sainte Baume||PS (5+5+5), PM (5)||950||North||2005-2006|
|Sainte Victoire 1||PS (5); PN (5)||650||North||2009-2010|
|Sainte Victoire 2||PM (5); PP (5)||500||Flat||2010|
|Ventoux||PS (5), PN (10), AA (19)||1100 to 1400m||North||2009-2010|
2.2. Growth and architectural parameters
When present, 5 to 10 needles were randomly chosen from the base to the end of each twig and all around it to measure their length and width. Needle number per twig was counted on a subsample of twigs (1/3) with consideration of missing needles which were counted using their scars on the twig. As needles are lined up in three to five lines or spirals along the shoot, counting needles along one or two of these lines or spirals and then bulking up the count proved to be a very reliable assessment (error < 5%)
In order to bridge primary and secondary growth, ring width was measured for all studied trees. Two cores per tree were collected perpendicularly at 1.3 m height. Ring widths of each core were measured using a micrometer (±0.002 mm, Velmex Inc., Bloomfield, NY). Some trees were also logged for stem analysis and rings were counted along 4 perpendicular directions. Ring width series were firstly cross-dated and standardized with the classical methods of dendroecology, to remove age-related tendencies from the growth curve and to obtain a homogenised variance. Then elementary raw and detrended series were respectively averaged for each tree and master chronologies were constructed for each species and for each plot by averaging tree series.
2.3. Branch modeling
The observed variations of architectural parameters, needle number and needle size were integrated into a 2D-model of pine branch architectural development to simulate the impact of climate on pine total leaf area (figure 5c). We designed the model for the more complex of studied conifers: Aleppo pine (
At each step of branch development and at the end of each year, active twigs (with needles) could be counted and sorted by vigor. The number and size of needles per active twig was set according to table 2. Total needle surface (length*width*needle number) was calculated for each twig and bulked up for the whole branch.
This model was used to simulate branch growth for 10 years, as all parameters for secondary axes were obtained for this time span. For longer periods, the interaction and competition with neighboring branches, twig self pruning, branch aging and accidents may significantly change these parameters, so that a 3D model taking these interactions into account is necessary.
2.4. Statistical analyses
As most architectural and growth parameters slowly evolve with branch aging, it was necessary to remove this natural trend. This was systematically done for each parameter using the difference measured at equal cambial age for branches of respectively top vs middle and middle vs base of the crown, for the period 1995-2000 considered as accident-free (figure 8, method in references [12,37]).
To quantify inter-annual variability between traits, an individual detrended coefficient of variation (dCV), for the period 1990-2010, was computed, for each trait of each tree, as follow: (i) individual trend was removed by taking the residuals of the linear or non-linear model with time as explanatory variable and (ii) the standard deviation of the detrended sequence was divided by the raw sequence mean (i.e. the mean trait value for each tree).
Thus, the detrended coefficient of variation of trait j for tree i can be written:with Xj,i the sequence of trait j values for tree i ; β0j,i and β1j,i the corresponding estimates of the model used for detrending and mean (Xj,i) the mean value of the trait j for tree i.
For a global assessment of the relationship between growth, architectural parameters and climate, a Principal Component Analysis (PCA) was performed considering years as observations and all detrended architectural and growth parameters as variables, species by species. PCA was also used to help sorting good years (favorable climate for tree growth) and bad years (figure 7). As some parameters were not common to all species (polycyclism, male flowering, needle number), each PCA was performed with and without these variables to check the stability of years and variables in PCA planes. Needle length was not always available for the same period than other factors. Thus each PCA was also performed with needle length for available years and without needle length on the whole studied period (1995-2010).
All growth and architectural variables were averaged per species for bad and normal years. Bad years were defined as the four worst years in the 2003-2008 period for each individual variable. All other years were merged to compute the data for "normal years": as exceptionally low values due to repeated severe drought were excluded, we considered other data as normally good, mean or bad, representative of the normal interannual variability.
Partial least square (PLS) regressions were used to investigate relationships between architectural or growth parameters and climate. This method was chosen because it handles many variables with relatively few observations  and deals with correlated variables . The number of significant PLS components was chosen by a permutation test  with a 5% threshold for the explained variance. Variables were tested with a 1000-step cross-validation : they were retained only when the confidence interval (p<5%) for their partial correlation coefficient excluded zero. According to the phenology of the species in South-eastern France, climatic monthly parameters tested in each PLS were rainfall (P), maximum temperature (MaxT), minimum temperatures (MinT) and mean temperatures (MT) from January of previous year (n-1) to November of current year (n) over the period 1995–2010 [43,44]. The low number of observations (16) and high inter-annual variability of climate made grouping monthly climatic parameters necessary to obtain significant variables. We grouped the successive months having same signs for their individual partial correlation coefficients (sum of precipitations, average temperature). To compare exposition, position, status and vigour classes for each growth variable, normality was checked using a standard Shapiro-Wilks test. When the distribution was normal, a variance analysis and a multiple comparison test were performed to look for significant differences globally and further compare the different classes two by two. When the distribution was non-normal, these comparisons were performed respectively by a Kruskall-Wallis test and a Nemenyi test .
For all species, most growth and architectural variables showed decreasing values in the last 15 or 20 years, and also from the top to the base of the crown (figure 6
Whatever the species and variables used, PCA axis 1 and 2 were dominant with respectively 50-70% and 17-32% of explained variance, far over Axis 3 (4-12%). PCA results were highly coherent between species for the distribution of years in 3 groups (figure 7): bad years (2004 to 2007) and good years (1995 to 1998) were stable across all analyses, other years being more variable, generally in intermediate situation but sometimes good or bad (2003 and 2008). Some variables were stable for most species: shoot length, polycyclism and needle number were correlated with Axis 1 and between each other. Fruiting rate and needle length were better correlated to axis 2. Male flowering was available for
The period 2004-2007 was characterized by a reduction of all growth and architectural parameters: shoot length, ring widths, needle length and number, and branching rate (figures 6
Branching rate, one of the important architectural development indicator, is a good example of the common pattern between all species (figure 9
Polycyclism, a fundamental growth trait for some pines species, confirmed this pattern (figure 10
All architectural variables were positively correlated to branch vigour. Figure 11 shows the example of needle number for
The relative fall of growth and architectural variables after 2003 was generally more severe for vigorous and principal axes and at the top of the crown than on weaker and secondary axes and in the middle and bottom of the crown (figures 6, 11
Although branch vigour was correlated to branch hierarchy and position within the crown, each of these factors significantly influenced branch architectural development (figure 13). For the same vigour (same mean length growth during the last 3 years), a branch had higher values on branch principal axes than on secondary axes and on axes of following orders, and decreasing values from the top to the base of the crown. Vigour, hierarchy and position were always highly significant (P<0.01%). South-exposed branches had sometimes higher values than north-exposed ones, but the difference was rarely significant (P<5%) and always at the very limit.
Needle length was highly variable from one year to the other (figures 7
Although we only have short series of data for
Table 2 presents
These values were used to compute the total leaf area of a branch during 10 years with and without a 4-year accident, corresponding to the four worst years observed for many variables in this study (figure 6). According to the branch model, the deficit of active twigs (carrying needles) and of leaf area amounted respectively to 59 % and 78 % for
|needles number per twig|
|needles length (mm)|
|Needle width (mm)|
|Needle surface (mm²)|
|Leaf area per twig (mm²)|
4.1. Endogenous and climatic effects on crown architecture components
Most of the studied morphological traits present a progressive decreasing trend along time (figures 6, 9, 10, 11 and 12). This decrease corresponds to a morphogenetical gradient named axis drift [15,46,47]. Morphological traits values are also driven by axis vigor, hierarchy and position within the crown (figure 13
Theses endogenous constraints can be viewed as a variational module sensu Wagner
Pre-induction of the primary growth characteristics (and when relevant of several flushes) in buds induces a strong dependence of primary growth on the global health status of trees and branch vigour at the time of bud formation, i.e. in previous year. Except in case of successive intense stress or extreme events, tree vigour and health rarely sharply changes between two years as trees can use non-structural carbohydrates stored in the stem and branches for growth and respiration. Such an autocorrelation between growth units is well known in ring width series [5,49,50] and was logically found for primary growth in this study.
Climatic effect on tree growth are generally analyzed using classical dendroclimatic analyses based on tree-ring data . They usually conclude that tree radial growth is affected by continuous changes in climatic conditions (trends) and by strong climatic events [3,26,52]. Height growth was more occasionally used as indicator due to time-consuming measurements in the field, but gave reliable results . This study revealed that a temporal survey of branch elongation and architectural parameters is of interest to this aim.
The occurrence of successive drought years between 2003 and 2007 led to a clear fall in all growth and architectural parameters: branch length and tree height growth, branching rate and polycyclism, needle number and needle length reached very low values during two to five years. These reductions could be clearly attributed to climate as the natural trends induced by branch or tree aging were slower, and all parameters finally recovered, at least partly, with more favorable climatic conditions.
As an extreme event with deleterious effects on forest health and productivity throughout Europe  and on studied species in Southeastern France , the 2003 summer heat-wave was supposed to have deeply impacted tree architectural development and growth. The direct impact of 2003 was mainly visible on needle length and individual leaf area (figures 8
All the traits did not respond similarly to climate variability and accidents (figures 8, 12, 15) for a species on a given site. Each of them is driven by many factors, related to climate or to functional relations between organs within a tree.
Needle length is mainly determined by the climate of the year of their development. For polycyclic species, needles of the later cycle can be very short when they lack time to complete their lengthening before the end of the growth season. Conversely, needle width and thickness mainly depend on twig diameter and vigour  and therefore, at tree level, on the climate of previous years and more generally on branch and tree health status. Although they are slightly sensitive to climate conditions of their growing year, they do not follow the rapid changes of needle length. Consequently, needle area is a compromise between its length and width which do not vary synchronously. Needle number and branching rate on a given twig are predetermined in the terminal bud of the twig. For monocyclic species, they are fully controlled by the climate of the previous year and by twig vigour. For polycyclic species, several growth units can be predetermined by previous year conditions but additional cycles may be formed later in the growing year according to climate and other constraints . These new growth units can give birth to both needles and new branches, or only needles or branches. Branches are, however, also controlled by the climate of their first growth season: some of the preformed buds remain dormant and young shoots abort in case of unfavourable conditions. The total leaf area at branch level is thus complex to model due to the many factors at stake.
The variable response of traits in time and intensity may also be linked with phenology: secondary growth occurs longer than shoot extension in monocyclic species , but the opposite is observed in some polycyclic species when shoot growth occurs late in autumn and even in winter without cambial activity . Early or late climate-related stresses may not have the same impact.
Ring width seemed to present a reduced plasticity compared with annual shoot lengths for
The inertia of branching patterns, driving leaf number and total leaf area, may explain the increasing length of the integration period with climate aridity.
Although some differences can be observed in the response of species to inter-annual climate variability, this study showed globally consistent trends in time for all traits and all species (figures 7, 9
As already stated, within the six studied conifers, only two (
4.2. Effect of crown architecture components on tree health
Tree leaf area is the product of numerous architectural components. It is the result of the number and size of leaves per growth unit and the number of growth units per annual shoot and the branching rate. At tree level, it also depends on competition between branches and between trees, and on crown shape related to age and tree history. At all scales, it depends on twig, branch and tree vigour and health status, not to mention external factors as defoliators and diseases. Up to now, leaf area deficit and tree growth (radial, height, or volume) were the main factors used to quote tree health [63,64]. If leaf area deficit can initially occur as an avoidance mechanism to maintain a favorable water balance by reducing transpiration, it also induces a reduction in carbon assimilation . Consequently leaf area deficit may be the early warning of a sequence leading to tree death [63,66], and could be used to predict tree mortality . But this deficit was always assessed globally without desentangling its distinct causes. In this study, we quantified with our branch model an extreme leaf area real deficit, reaching nearly 80% after 4 very bad years (figure 16). This is far over the "leaf deficit" usually quoted by crown transparency, reaching such an extreme only for dying trees which is far from being the case in our plots. This discrepancy can be partly explained by the reduction of branch length growth and of the distance between leaves or needles along the twigs, which concentrates them in a smaller crown volume.
According to the branch model, during the two first bad years, branch deficit remained under 10% and could not explain the leaf area reduction which reached 35%. Thus, leaf number and size were the main factors at stake. With the lengthening of the drought period, the deficit in branch number became dominant, explaining most of the gap in total leaf area. Finally, during the recovery period after the end of bad years, branching shortage entirely matched leaf area deficit. This is consistent with figures 8
According to the concept of functional equilibrium [68,69] plants allocate biomass in priority to organs concerned by the most limiting factors : roots if case of nutrients or water shortage, shoots and leaves when their environment is deficient in CO2 or light. Accordingly, plants show remarkable resilience when part of their leaves, branches or roots are destroyed or artificially removed. They generally recover a normal leaf area or root length and biomass in a few years . This may be true when repeated or long lasting climatic stresses reduce first their aerial growth (shoot and needle length) (figures 8
5. Conclusion and prospects
Modeling tree responses to climate change and particularly dieback hazard is a key issue since strong changes in tree productivity, survival and recruitment were observed recently [4,63]. A main concern is the assessment of tree vulnerability to increasing drought periods. Empirical models based on statistical relationships are not reliable as they cannot accurately take thresholds and extreme events into account. In contrast, mechanistic models explicitly represent the processes by splitting them into different blocks, which describe the response of the process to some input variables. However the variability of architectural components is poorly represented up to now in process-based models of individual tree growth. Most of them ignore their spatial variations and differentiated temporal response according to axis position in the crown, hierarchy in the branch and vigour. Their improvement with these new findings is urgently needed.
Our analysis made on many sites and species in Southeastern France revealed common patterns of response of tree architectural development to climate change and accidents. The role of long lasting delayed consequences of climate accidents on branching rate, holding back the potential leaf area for years, is one of the key issues to be tackled. Low leaf area, through carbon shortage, may contribute to forest decline and die-back.
This study highlights the necessity of more thorough investigations, in terms of field work and modeling. Our preliminary results must be confirmed for new species and climates and with longer data series to disentangle the multiple and contradictory effects of climate change on tree architectural development.
We would like to thank Christian Ripert, Roland Estève, Willy Martin, Aminata N’Diaye Boucabar, Frédéric Faure-Brac, Maël Grauer, Hendrik Davi, Nicolas Mariotte, William Brunetto and Florence Courdier for their assistance in the field and laboratory work. This research was funded by the French National Research Agency (DROUGHT+ project N° ANR-06-VULN-003-04, and DRYADE project n° ANR-06-VULN-004), the French Ministry for Ecology, Energy and Sustainable Development (GICC–REFORME project, No MEED D4ECV05000007), the Conseil Général des Bouches-du-Rhône (CG13), ECCOREV Research Federation (FR3098), the ‘‘F-ORE-T’’ LTER network and Cemagref.