Gradients locaux et régionaux, 8 sites
(Chapitre III)
Gradients nationaux, Milliers de sites (Chapitres V et VI) Gradients locaux et régionaux,
18 sites (Annexe A)
Figure 1.9. Emplacement géographique des sites d’étude des programmes VISTA (rouge) et DivHerbe (bleu) utilisés dans cette thèse. Le site LAU, en mauve, était inclus dans les deux programmes. Abréviations du nom des sites, de l’ouest à l’est: MER = Métrola/Castro Verde; SUT = South Uist; MAG = Les Magnils; SLP = Saint-Laurent de la Prée; ERC = Ercé; MAR = Marcenat; THE = Theix; FAG = La Fage; HGM = Hautes Garrigues du Montpelliérais; TDV = Tour du Valat; LAU = Col du Lautaret; SAI = Les Saisies; BAT/BER = Båttjønndale/Berghøgda; MNP = Müritz National Park; OHR = Ohrazeni; BAL = South-east Baltic Sea; LAG = Lagadas; KDE = Karei Deshe
(niveau du site régional) et sur l’utilisation des terres et la qualité des sols (niveau des communautés locales). Ainsi, au sein de chaque site, les communautés ont été échantillonnées à différentes positions sur des gradients de variables environnementales locales telles que la disponibilité des nutriments, l’intensité des perturbations et l’intensité de sécheresse. Puis, en considérant tous les sites, il y a des gradients de variables environnementales régionales telles que la température et les précipitations moyennes. De plus, pour chaque communauté au sein de chaque site, la richesse spécifique, l’abondance des espèces et la valeur de nombreux traits – traits foliaires, traits de stature, traits phénologiques – pour les espèces dominantes (Pakeman et Quested, 2007) ont été collectés. Ces bases de données, qui contiennent de l’information précise sur des communautés prairiales sur une très grande échelle, donnent ainsi l’opportunité de tester de façon robuste des hypothèses d’importance capitale dans le domaine de l’écologie des communautés et de l’écologie fonctionnelle.
Durant cette thèse, j’ai aussi participé à un consortium visant à collecter, mettre en forme et analyser les données existantes sur la distribution géographique des espèces de prairies en France (DivGrass; description complète disponible sur le site du CESAB; www.cesab.org). Cette initiative s’opère à une échelle encore plus grande que les programmes VISTA et DivHerbe, ce dernier ayant été intégré à la base de données de DivGrass. Le consortium du projet DivGrass a donc amassé une quantité impressionnante de données sur la présence taxonomique, l’abondance et la distribution géographique des plantes des prairies françaises (Fig. 1.10) et collabore étroitement avec TRY – une initiative mondiale de collecte de données sur les traits fonctionnels (Kattge et al., 2011) – pour avoir de l’information sur les traits des espèces. De plus, de l’information sur plusieurs variables environnementales est disponible et ce, pour chaque cellule (8x8 km2) d’une grille virtuelle recouvrant la France. Les données contenues dans DivGrass comportent une certaine imprécision: les mesures d’abondance sont souvent en classe et non en pourcentage et les valeurs de traits sont pour la plupart des moyennes pour chaque espèce provenant de la base
Figure 1.10. Distribution spatiale des prairies françaises selon les 51 486 relevés botaniques de l’initiative DivGrass. Le gradient de couleur verte représente le pourcentage de la surface recouverte par les prairies permanentes dans chaque cellule de 5x5 km d’une grille recouvrant la France. Figure adaptée de Violle et al. (2015).
de données TRY au lieu d’être mesurées aux sites. Malgré tout, la grande échelle de cette initiative et la quantité de données inclues en font une opportunité unique pour évaluer certains enjeux cruciaux dans les domaines de l’écologie des écosystèmes et de la biogéographie fonctionnelle. Bien que ma participation à ce consortium fût secondaire, les projets auxquels j’ai participés sont en continuité directe avec la problématique principale de ma thèse – c’est-à-dire les mécanismes d’assemblage des communautés – et les deux
derniers chapitres du corps de la thèse (Chapitres V-VI) sont donc consacrés à certains résultats obtenus au sein de cette initiative.
CHAPITRE II –
OCCUPANCY AND OVERLAP IN TRAIT SPACE ALONG A SUCCESSIONAL GRADIENT IN MEDITERRANEAN OLD-FIELDS
par
Jessy Loranger, Benjamin Blonder, Éric Garnier, Bill Shipley, Denis Vile et Cyrille Violle article soumis à American Journal of Botany
2.1 – Présentation de l’article et contribution des auteurs
L’Annexe A présente comment différentes variables environnementales interagissent ensemble pour affecter plusieurs types de traits fonctionnels et comment ces relations traits- environnement peuvent être complexes. Dans ce deuxième chapitre, nous nous intéressons maintenant à comment la sélection de ces traits par l’environnement (abiotique et biotique) influence l’assemblage de communautés d’herbacées appartenant à différents stades de succession. Plus précisément, nous nous sommes intéressés aux variations de l’espace fonctionnel occupé par des friches herbacées représentatives d’une chronoséquence couvrant un intervalle de temps de 2 à 42 ans après abandon. Contrairement aux autres études sur l’assemblage des communautés le long d’une succession, nous avons analysé ici trois aspects de la variation fonctionnel (trajectoire de la moyenne, chevauchement fonctionnel entre les stades et convergence ou divergence fonctionnelle), ce qui nous permet d’évaluer avec plus de confiance la nature des processus de sélection des traits par l’environnement menant à l’assemblage des communautés. Un des points fort du chapitre est notre capacité à y tester l’impact sur les résultats d’utiliser différents traits appartenant à la même dimension fonctionnelle. Ceci permet de tester la pertinence et la généralité du
concept de dimension fonctionnelle et de démontrer l’importance de bien considérer les fonctions spécifiques représentées par chaque trait pour interpréter les résultats. Ceci renvoie aussi à l’importance de l’Annexe A où une description des relations traits- environnement permet de mieux identifier ces fonctions.
Je suis à l’initiative de ce travail, tant d’un point de vue des questions posées que des analyses menées. J’ai souhaité me former et appliquer la méthode originale de l’« hypervolume » pour appréhender la niche fonctionnelle des espèces d’un point de vue multivarié, plutôt qu’univarié. Éric Garnier et Denis Vile ont fourni les données. Éric Garnier, Cyrille Violle et moi-même avons développé le plan d’analyse des données et la structure de la présentation de celles-ci. J’ai effectué toutes les analyses et la représentation des résultats et j’ai rédigé une première version complète du chapitre. Benjamin Blonder a été d’une aide précieuse dans l’analyse des données puisque j’ai utilisé la technique d’hypervolume qu’il a lui-même développée. Tous les auteurs ont participé à l’amélioration de la première version écrite, avec une participation particulièrement importante de Bill Shipley et Benjamin Blonder.
2.2 – Abstract
Premise of the study: Secondary succession is a worldwide phenomenon affecting plant communities. Studying functional variation during succession allows understanding the mechanisms through which environmental shifts drive succession. Here we investigated the changes in the functional space occupied by herbaceous communities during succession. Furthermore, since different traits are differently affected by environmental conditions, we asked how considering different sets of plant traits impacts those changes.
Methods: Using a chronosequence of Mediterranean old-fields (2-42 years after abandonment) we presented a threefold analysis investigating the displacement of occupied functional space during succession, how the volume of occupied functional space varies compared to null expectations, and the functional overlap between communities of different successional status. We repeated these analyses considering i) separately and together the Leaf-Height-Seed functional dimensions and ii) different sets of traits representing those dimensions.
Key results: From early to late succession, dominance shifted towards conservative- competitive species. Functional strategies of mid-successional communities appeared more diverse than expected by chance and less diverse than expected for early and late communities. Early and middle stages overlapped the most. These patterns were generally robust to the choice of functional axes, though important trait-specific exceptions occurred.
Conclusions: Different traits typically grouped under one functional dimension can substantially affect the results, discouraging the use of surrogate traits from the same dimension. Despite that the choice of traits matters for several conclusions, we showed evidence for a well-defined history of successive dominance of different assembly mechanisms along succession, resulting in a generally stronger functional diversification in mid-succession.
Key words: community assembly; competitive exclusion; environmental filtering; functional convergence and divergence; limiting similarity; secondary succession
2.3 – Introduction
Characterizing the variation in functional traits during secondary succession is essential to understanding the underlying mechanisms linking taxonomic composition to environmental variation and predicting successional outcomes (Odum, 1969; Connell and Slatyer, 1977; Pickett et al., 1987; Prach et al., 1997; Fukami et al., 2005). Environmental conditions change during succession (Bazzaz, 1979; Huston and Smith, 1987), imposing different constraints for plants of each successional stage to which they must respond with specific functional trade-offs (Tilman, 1985, 1990). For example, biomass normally increases during succession with a corresponding increase in competition for light resulting in positive relationships between successional age, plant height and species fitness. This functional perspective is still a major component of modern succession ecology and is necessary to advance the discipline (Meiners et al., 2015).
Given the importance of functional shifts during succession, the variation in the size of the occupied community functional space is an important and yet rarely assessed aspect of successional studies that generates precise hypotheses relating functional distribution to various assembly mechanisms (Weiher and Keddy, 1995; Enquist et al., 2015). These include i) abiotic environmental filtering (Keddy, 1992; Kraft, Adler, et al., 2015), which is the sorting of species by the abiotic environment leading to the restriction of viable functional strategies and potentially to the convergence of traits conferring adaptation to the environment, ii) competitive exclusion (Hardin, 1960; Keddy, 1992), which is the exclusion of weaker competitors by stronger ones for a given limiting resource, leading to the convergence of competitive traits (Grime, 2006; see Mayfield and Levine, 2010 for a similar argument for phylogenetic diversity), or iii) limiting similarity (MacArthur and Levins, 1967), where co-existing species avoid competition by having different functional strategies leading to functional divergence. Using a null model approach we can test patterns of functional distribution by comparing the volume of occupied functional space in
a community to null expectations (see e.g. Cornwell et al. 2006, Bernard-Verdier et al. 2012): Smaller (larger) occupied volumes than expected reflect functional convergence (divergence) and the traits considered can be related to a given assembly mechanism.
Understanding of successional processes also requires the consideration of multiple functional dimensions (Weiher et al., 1999; Grime, 2001; Laughlin, 2014; Maire et al., 2015), such as the Leaf-Height-Seed (L-H-S) framework (Westoby, 1998). The L-H-S dimensions represent important functions of the plant (Westoby, 1998; Garnier et al., 2016) – resource acquisition and use (leaf), plant stature with a role in light/space competition (height/size), and regeneration (seed) – each reflecting different trade-offs experienced by plants. For example, the leaf and size dimensions were repeatedly found to independently – but strongly – impact community assembly during succession, their respective relative importance depending on environmental conditions (Douma et al., 2012; Raevel et al., 2012). In contrast, Navas et al. (2010) showed that changes in functional composition in a Mediterranean secondary succession represented the evolution of a whole-plant strategy against drought, combining physiological, morphological and phenological traits. Therefore, it is clear that one functional dimension may not be enough to capture strategy changes (Falster and Westoby, 2005) and that analyzing both separately and together how different functional dimensions vary with succession is desirable (Bhaskar et al., 2014). This requires the ability to describe and quantify how plants occupy a multidimensional trait space and how this space occupancy changes during succession, something that has not been possible until recently (see Edwards et al., 2014 for an example with insects along a disturbance gradient).
Here we explore 1) the mean of multivariate functional trajectories along a successional sequence considering several functional combinations, 2) the degree of functional overlap between successional stages, and 3) the variation in the occupied volume of multivariate trait spaces. This threefold analysis should yield novel understanding of successional
mechanisms. For example, functional overlap across successional stages gives complementary insights relative to the variation in the rate of functional displacement during succession (Purschke et al., 2013). In this study we revisit a previously studied secondary succession in a series of Mediterranean abandoned fields (Garnier et al., 2004; Shipley et al., 2006; Vile et al., 2006a; b) and apply the proposed threefold analysis while considering together, as well as separately, the L-H-S dimensions.
We formulated four specific hypotheses centered on two main research questions. The first research question is focused on the mechanisms of successional patterns and is represented by three hypotheses. First (H1), we expect to find a functional shift towards a more nutrient conservative and competitive strategy during succession, affecting all functional dimensions (Grime, 1977; Garnier et al., 2004; Vile et al., 2006a). Second (H2), we expect each separate dimension to show significant functional convergence/divergence at one successional stage or another, depending on the importance of this dimension for the assembly in a given stage. However, only when considering the L-H-S dimensions together should we detect significant patterns for each stage because selection during succession is likely to operate on integrated phenotypes rather than individual axes (Navas et al., 2010). Since environmental filtering (sensu Kraft et al. 2014) is expected to be relatively more important in early succession while competitive exclusion would be more important latter during succession (Connell and Slatyer, 1977; Tilman, 1990; Schleicher et al., 2011), we hypothesize that both early and late succession will be characterized by functional convergence – i.e. smaller occupied space than expected from random drawings from the regional species pool – allowing fast colonization earlier (seed and leaf dimensions) and a better competition for nutrients and light later (size and leaf dimensions). In mid-succession where the strength of competition should be intermediate, both early-colonizers and good competitors could co-occur and we should observe greater functional divergence (Navas and Violle, 2009) in all functional dimensions, i.e. a larger occupied functional space than expected. Third (H3), the functional overlap should be greater between middle and late stages of succession than between early and middle stages. Indeed, since a shift towards a
conservative strategy is expected, inherent growth rate should be reduced which should translate into a slower displacement rate and an extended species coexistence in later stages according to classical theory (Huston, 1979; Huston and Smith, 1987).
The second research question, with only one related hypothesis, is rather focused on the way we represent ecological strategies, i.e. the traits used to represent the L-H-S dimensions. Originally, Westoby (1998) proposed the use of specific leaf area (L), plant height (H) and seed mass (S). However, there are several other functional traits that can represent the general functional dimensions of resource acquisition and use, plant stature, and regeneration. At the same time different traits from the same dimension can reflect different specific plant functions (Garnier et al., 2016). For example, specific leaf area (SLA) – one of the most commonly measured plant functional trait and the most widely used proxy for the leaf dimension (Westoby, 1998) – relates to photosynthesis, while leaf dry matter content – another trait representing the leaf dimension – relates more directly to tissue density and leaf palatability (Deraison et al., 2015; Garnier et al., 2016). Consequently, community assembly processes could act differently on different traits of the same dimension. Despite the potential effect that these differences can have on functional analyses, the impact of considering different traits for a given dimension has never been investigated. We therefore investigate this issue in the framework of our threefold analysis and, assuming that the L-H-S dimensions are meaningful groups of functional traits co- varying together (as presented in Garnier et al. 2016), we should (H4) find similar trends regarding our second and third hypotheses regardless of the trait chosen to represent any given dimension.
2.4 – Methods
2.4.1 – Study site
The fields, located in southern France (43o51’N, 3o56’E, 100-160 m a.s.l.) in a sub-humid Mediterranean climate, consisted of twelve abandoned vineyards located within 4 km2 and which had similar soil texture (calcic cambisol; FAO-Unesco 1997) and pH (between 8.1 and 8.6), and were abandoned between 2 and 42 years prior to our study (see Garnier et al. 2004 for more details). Herbaceous species were dominant in all plots (Annex B, see Supplemental Data with the online version of this article) and the environmental similarities between the fields make this site a good chronosequence to study the effect of succession on plant communities (Pickett, 1989; Foster and Tilman, 2000; Garnier et al., 2004). The twelve fields were grouped into three main successional stages: early (2 years after abandonment), middle (7-12 years after abandonment, average = 9.2) and late (26-42 years after abandonment, average = 34.4). This grouping was based on plot cover – from largely uncovered to almost completely covered – and on occurring species that were found to be indicative of successional status in these regions (Braun-Blanquet et al., 1952; Escarré et al., 1983) – from the dominance of annuals to the arrival of small woody species.
2.4.2 – Sampling and functional traits
In each field, herbaceous areas were chosen for sampling. Small woody species only occurred in the oldest fields. Total aboveground standing biomass was harvested twice a year in February and May 2000 in four 0.5 m2 plots per field for a total harvested area of 1 m2 per field. The live biomass was sorted by species and oven dried to determine relative
abundances by calculating the dry mass proportion of each species in the fields (complete sampling details in Garnier et al. 2004).
In each field, species’ means of nine functional traits (Table 2.1) were measured in 2000 or 2002 following standard protocols (Cornelissen et al., 2003) for all species making up at
Table 2.1. Definition of abbreviations and units of the nine functional traits.
Abbreviation Definition Units
Leaf dimension
SLA Specific leaf area m2/kg
LDMC Leaf dry matter content mg/g
LNC Leaf nitrogen concentration mg/g
Size dimension
H_max Reproductive (maximum) plant height cm VegMass Pre-reproductive (before flowering) vegetative mass g Lm Plant leaf mass at first seed dispersal g Seed dimension
SeedM Mean individual seed mass g
SN Number of seeds produced per individual per year
OnsFlo Flowering date of ~50% of plants in the population Julian day Note: All masses are dry masses
least 80% of the total biomass harvested in that field, as recommended by Pakeman and Quested (2007). A total of 30 species were selected (Annex B) and the measurements were conducted at the spring peak of growth in this Mediterranean system. A unique aspect of our trait dataset is that the three leading dimensions of interspecific functional variation (originally specific leaf area-plant height-seed mass; Westoby 1998) were each represented
by three traits. Specifically, the “resource acquisition and use (leaf)” traits were 1) specific leaf area (SLA) as the ratio of one-sided area of a fresh leaf to its oven-dry mass (m2/kg), 2) leaf dry matter content (LDMC) as the oven-dry mass of a leaf divided by its water- saturated fresh mass (mg/g) and 3) leaf nitrogen concentration in mg/g; The “size (height)” traits were 1) the reproductive plant height as the distance between the ground level and the upper boundary of the reproductive organs (cm), 2) the pre-reproductive vegetative mass, which is the oven-dried mass of all vegetative above-ground organs before flowering (g) and 3) the plant total leaf mass which is the oven-dried mass of all the leaves of one individual (g); The “regeneration (seed)” traits were 1) the mean seed mass (g), 2) seed number produced per individual per year and 3) the onset of flowering as the first day where at least half of the individuals of a species are flowering (Julian days). All values of a given trait for a given species were averaged; thus every species was attributed only one mean value per trait for the whole site. See Garnier et al. (2004) and Vile et al. (2006a) for more details on trait measurements.
2.4.3 – Hypervolume calculations and null model
The hypervolume method (Blonder et al., 2014) uses presence data to estimate a multi- dimensional volume via thresholded kernel density estimation. Contrary to other multidimensional functional richness methods (e.g. Cornwell et al. 2006), this hypervolume is not sensitive to outliers and allows holes in space, giving a more accurate approximation of the real occupied functional space. Hypervolumes were calculated via the “hypervolume” package in R (R Core Team, 2013) with computational parameters set to 1000 repetitions per data point, a quantile threshold of 5% to exclude outliers, and a fixed bandwidth of 1.1 in accordance with the Silverman bandwidth estimator of the hypervolume package which suggested the same bandwidth for all traits. The values of the reproductive plant height, pre-reproductive vegetative mass, plant leaf mass, seed mass and seed number were loge-transformed to decrease the effect of few species with very high
values. Then all traits were standardized (divided by their standard deviation) prior to analyses.
Different stages of succession did not have the same species richness (early = 17, middle = 26, late = 11). As hypervolumes calculated using fixed bandwidths necessarily increase with species richness (Lamanna et al., 2014) it is important to use a null model approach to compare observed hypervolumes to a null sampling expectation. We used a null lottery model of species selection from a regional species pool, without replacement, to generate