Le rôle du phosphore (P) en agriculture est indéniable: le P est un nutriment essentiel dont tous les êtres vivants ont besoin pour fonctionner, et est donc nécessaire pour maintenir les rendements agricoles à l’échelle globale dans les années à venir. Une grande partie du P utilisé pour fertiliser les cultures se présente sous forme d’engrais chimique et provient de mines de roches phosphatées. Cette ressource finie est gérée de manière non-optimale: dans certains endroits du Monde, le P est utilisé de manière excessive et peut nuire à l’environnement, alors qu’à d’autres endroits, le P apporté est insuffisant et conduit à des baisses de rendement importantes. Cette hétérogénéité, combiné à des problématiques d’accès à la ressource, qui dépend également de facteurs économiques et politiques, conduit à de sérieuses questions sur les impacts potentiels du P sur la sécurité alimentaire mondiale. Des études récentes se sont penchées sur les principaux facteurs limitant les rendements agricoles dans le Monde, mais présentent des difficultés à séparer la contribution de ces différents facteurs, et en particulier du P. Dans un premier temps, j’ai combiné des simulations de la distribution du P dans les sols agricoles et des simulations de croissance des céréales dans des conditions idéales (i.e. non limitantes en eau, azote, etc.), tout en prenant en compte, de manière fine, les mécanismes de transfert du P entre le sol et la plante. J’ai montré que le P pourrait contribuer de manière significative à une baisse de rendement par rapport au rendement potentiel de 22, 55 et 26 % en blé d’hiver, maïs et riz. Cette diminution n’est que partiellement impactée quand les apports actuels de P par fertilisants chimiques sont considérés et ceci s’explique principalement par l’historique du bilan en P des sols (qui a contribué à fortement augmenter les stocks de P des sols). Cependant, la non prise en compte de certains processus, à savoir ceux liés aux ajustements des plantes dans des conditions limitantes en P, ont pu fortement biaisé ces estimations. Pour mieux représenter ces processus d’ajustements, j’ai ensuite développé un modèle d’allocation du carbone (C) et du P basé sur des principes d’optimisation d’utilisation des ressources au sein de la plante. Le modèle est capable de simuler la réponse de la plante à une limitation en P: augmentation du ratio racines / biomasse aérienne, diminution de la biomasse totale et de la concentration en P. Le modèle a été testé dans un gradient de disponibilité en P à différentes échelles (plante en hydroponie et au champ) et reproduit raisonnablement le comportement des plantes. Malgré des hypothèses simplistes qui ne permettent pas de capturer la nature exacte de l’allocation, le modèle présenté peut être introduit dans un modèle de végétation plus physique, permettant l’étude de la limitation en P de manière plus générique. Le couplage du modèle d’allocation idéalisé à un modèle de végétation physique a été réalisé en utilisant ORCHIDEE, un modèle de végétation dynamique utilisé pour étudier les interactions végétation-climat. Les paramétrisations de processus fondamentaux au sein d’ORCHIDEE (assimilation, etc.) ont été utilisées pour piloter le modèle d’allocation en fonction de la disponibilité en C et en P, et les simulations ont été comparées à deux jeux d’observations sur maïs irrigué. Les résultats ont montré le potentiel de la combinaison de ces deux modèles pour simuler de fonctionnement des cultures dans différents environnements. Le modèle ainsi obtenu pourra être utilisé pour mieux quantifier, à l’échelle mondiale, la contribution du P à la baisse de rendement des cultures par rapport à leur potentiel. / The global role of phosphorus (P) in agriculture is undeniable: P is an essential nutrient required by all living beings to function, and thus necessary for sustaining yields worldwide in the time to come. In global agriculture, most of the P used to grow crops comes in form of chemical fertilizer which is mined from existing soil deposits. This in itself would not be an issue, was it not for the way we globally (mis)manage this potentially finite resource. While some places use P to the point of harming the environment, others do not have enough to sustain their yields and feed themselves. Combined with uncertainties of equitable P supply in the future which depend on economical and political factors as well, serious questions arise on the potential impacts of P on global food security. Recent studies have looked into the main drivers of yield worldwide, but have difficulties separating P’ contribution, as they lack the information to do so due to their empirical nature. As an initial step, we combined simulated global information on agricultural soil P and cereal growth in ideal conditions, while accounting for mechanisms of soil-plant P transfer more faithfully. We have found that P could significantly contribute to existing global production gaps with an average yield gap of 22, 55 and 26 % in winter wheat, maize and rice; lowering only slightly with today’s P fertilizer use. This is mainly to be due to the global P management history or the net soil P balance up to date. But the idealized nature of the employed models ignored other processes, namely plant adjustment in P limited environments, which have a significant potential to change our diagnostic estimates. To better represent plant adjustment, we have then developed an carbon (C) & P allocation model based on optimal functioning principles. The idealized model is capable of simulating primary plant response to a P limited environment: root-shoot ratio change, biomass and P concentration decrease. It was compared to plant growth across a P availability gradient at different scales (hydroponic to field) and has been found to reasonably predict observed plant behaviour. In spite of its simplistic assumptions which do not capture the exact nature of P flow within a plant, the idealized model could be introduced into a more physical vegetation one to allow the study of P limitation in a generic growing environment. The coupling of our idealized allocation model to a physical vegetation one was performed using ORCHIDEE, a dynamic vegetation model used to study global vegetation-climate interaction. Its parameterizations of fundamental plant processes were used to drive our model as function of C and P availability, and compared to two irrigated maize observation datasets. The results have shown the potential of their combination to simulate crops in different growing environments, which is to be used on a global scale and finally help us better understand contribution of P to crop productivity globally.

This PhD thesis presents the spatio-temporal distribution of terrestrial carbon fluxes for the time period of 1982 to 2002 simulated by a combination of the process-based dynamic global vegetation model LPJ and a 21-year time series of global AVHRR-fPAR data (fPAR – fraction of photosynthetically active radiation). Assimilation of the satellite data into the model allows improved simulations of carbon fluxes on global as well as on regional scales. As it is based on observed data and includes agricultural regions, the model combined with satellite data produces more realistic carbon fluxes of net primary production (NPP), soil respiration, carbon released by fire and the net land-atmosphere flux than the potential vegetation model. It also produces a good fit to the interannual variability of the CO2 growth rate. Compared to the original model, the model with satellite data constraint produces generally smaller carbon fluxes than the purely climate-based stand-alone simulation of potential natural vegetation, now comparing better to literature estimates. The lower net fluxes are a result of a combination of several effects: reduction in vegetation cover, consideration of human influence and agricultural areas, an improved seasonality, changes in vegetation distribution and species composition. This study presents a way to assess terrestrial carbon fluxes and elucidates the processes contributing to interannual variability of the terrestrial carbon exchange. Process-based terrestrial modelling and satellite-observed vegetation data are successfully combined to improve estimates of vegetation carbon fluxes and stocks. As net ecosystem exchange is the most interesting and most sensitive factor in carbon cycle modelling and highly uncertain, the presented results complementary contribute to the current knowledge, supporting the understanding of the terrestrial carbon budget. / In der vorliegenden Arbeit wird anhand der Kombination eines dynamischen globalen Vegetationsmodells mit einer Zeitreihe von 21 Jahren optischer Satellitendaten eine realistische Abschätzung der terrestrischen Quellen und Senken von CO2 ermöglicht. Grundlage des hier vorgestellten neuen Modells stellt das dynamische globale Vegetationsmodell LPJ dar, ein prozessorientiertes Vegetationsmodell, das basierend auf ökophysiologischen Grundlagen die Vegetationsverteilung und -dynamik, Störungen (z.B. Feuer) und den Kohlenstoff- sowie den Wasserkreislauf modelliert. Die Kopplung des LPJ-DGVM erfolgte mit einer Zeitreihe globaler AVHRR-fPAR Daten (fPAR – Anteil photosynthetisch aktiver Strahlung), für den Zeitraum 1982 bis 2002 in einer räumlichen Auflösung von 0.5°. Als Ergebnis liegt nun eine globale raum-zeitliche Verteilung aller relevanten Kohlenstoffflüsse vor: Nettoprimärproduktion, Bodenrespiration, Nettoökosystemproduktion, durch Feuer und Ernte emittierter Kohlenstoff, sowie der in Biomasse und Boden gespeicherte Kohlenstoff. Verglichen mit dem Originalmodell haben sich durch die Einspeisung der Satellitendaten alle relevanten Kohlenstoffkomponenten verringert und zeigen nun bessere Übereinstimmung mit Literaturwerten. Die geringeren Kohlenstoffflüsse resultieren aus einer Kombination verschiedener Effekte: geringere Vegetationsbedeckung, Berücksichtigung der landwirtschaftlichen Nutzfläche, realistischere Abbildung der Saisonalität, Veränderung der Vegetationsverteilung und Verschiebung der Artenzusammensetzung. Die globalen Kohlenstoffflüsse werden mit dem vorgestellten Modell realistischer abgebildet als mit Ansätzen, die nur die potentiell natürliche Vegetation simulieren. Insbesondere die Quellen- und Senkendynamik unterliegt vielfältigen Prozessen, die mit einem Modell, dass auch die Bodenrespiration prozessorientiert berücksichtigt, verlässlich geschätzt wird.

Climate change is no longer debated in the context of whether or not it is occurring, but rather in the context of how rapid and extensive that change will be. This is the global situation to which the biomes of national parks in Canada and the United States must adapt. Through the use of the MC1 Dynamic Global Vegetation Model (DGVM) this thesis constructs projections of possible vegetation response of ten biome classifications to the impacts of continental-scale climate change in seven regions: Atlantic, Great Lakes, Mountain, Northern, Pacific, Prairie, and Southern. It then analyzes the potential ways in which DGVMs can be utilized by park management schemes in accommodating for future climate change in the selection, creation, and maintenance of national parks. As the latest generation of vegetation modelling systems, the advantages of Dynamic Global Vegetation Models over pre-existing equilibrium biogeography models are examined in this thesis. DGVMs highlight the degree to which ecosystems are interconnected, and are able to provide continental-scale data necessary in coordinating an integrated planning approach for national parks in North America. They are utilized in this study for generating projections of future biome distribution, based on climate information from three General Circulation Models: CGCM2, CSIRO Mk2, and HadCM3. Following the generation of possible climate scenarios, the impact of changes to biome distribution within national parks is discussed. The thesis findings provide valuable modelling analysis and scenarios for use in future planning by the US National Park System and Parks Canada. Utilization of DGVMs will help in creating flexible, coordinated management strategies that take into account projected vegetation responses to climate shifts that lie ahead.

Climate change is no longer debated in the context of whether or not it is occurring, but rather in the context of how rapid and extensive that change will be. This is the global situation to which the biomes of national parks in Canada and the United States must adapt. Through the use of the MC1 Dynamic Global Vegetation Model (DGVM) this thesis constructs projections of possible vegetation response of ten biome classifications to the impacts of continental-scale climate change in seven regions: Atlantic, Great Lakes, Mountain, Northern, Pacific, Prairie, and Southern. It then analyzes the potential ways in which DGVMs can be utilized by park management schemes in accommodating for future climate change in the selection, creation, and maintenance of national parks. As the latest generation of vegetation modelling systems, the advantages of Dynamic Global Vegetation Models over pre-existing equilibrium biogeography models are examined in this thesis. DGVMs highlight the degree to which ecosystems are interconnected, and are able to provide continental-scale data necessary in coordinating an integrated planning approach for national parks in North America. They are utilized in this study for generating projections of future biome distribution, based on climate information from three General Circulation Models: CGCM2, CSIRO Mk2, and HadCM3. Following the generation of possible climate scenarios, the impact of changes to biome distribution within national parks is discussed. The thesis findings provide valuable modelling analysis and scenarios for use in future planning by the US National Park System and Parks Canada. Utilization of DGVMs will help in creating flexible, coordinated management strategies that take into account projected vegetation responses to climate shifts that lie ahead.

A central issue in climate change research is to identify what species will be most affected by variations in temperature, precipitation or CO2 and via which underlying mechanisms. Dynamic global vegetation models (DGVMs) have been used to address questions of habitat shifts, extinctions and changes in carbon and nutrient cycling. However, DGVMs have been criticized for assuming full migration and using the most generic of plant functional types (PFTs) to describe vegetation cover. My doctoral research addresses both of these concerns. In the first study, I added two new tropical PFTs to an existing regional model (LPJ-GUESS) to improve vegetation representation in Central America. Although there was an improvement in the representation of some biomes such as the pine-oak forests, LPJ-GUESS was still unable to capture the distribution of arid ecosystems. The model representations of fire, soil, and processes unique to desert vegetation are discussed as possible explanations. The remaining three chapters deal with the assumption of full migration, where plants can arrive at any location regardless of distance or physical barriers. Using LPJ-GUESS, I imposed migration limitations by using fat-tailed seed dispersal kernels. I used three temperate tree species with different life history strategies to test the new dispersal functionality. Simulated migration rates for Acer rubrum (141 m year-1) and Pinus rigida (76 m year-1) correspond well to pollen and genetic reconstructed rates. However, migration rates for Tsuga canadensis (85 m year-1) were considerably slower than historical rates. A sensitivity analysis showed that maturation age is the most important parameter for determining rates of spread, but it is the dispersal kernel which determines if there is any long distance dispersal or not. The final study demonstrates how northerly refugia populations could have impacted landscape recolonization following the retreat of the last glacier. Using three species with known refugia (Acer rubrum, Fagus grandifolia, Picea glauca), colonization rates were faster with a northerly refugia population present. The number of refugia locations also had a positive effect on landscape recolonization rates, which was most pronounced when populations were separated. The results from this thesis illustrate the improvements made in vegetation-climate models, giving us increasing confidence in the quality of future climate change predictions.

A central issue in climate change research is to identify what species will be most affected by variations in temperature, precipitation or CO2 and via which underlying mechanisms. Dynamic global vegetation models (DGVMs) have been used to address questions of habitat shifts, extinctions and changes in carbon and nutrient cycling. However, DGVMs have been criticized for assuming full migration and using the most generic of plant functional types (PFTs) to describe vegetation cover. My doctoral research addresses both of these concerns. In the first study, I added two new tropical PFTs to an existing regional model (LPJ-GUESS) to improve vegetation representation in Central America. Although there was an improvement in the representation of some biomes such as the pine-oak forests, LPJ-GUESS was still unable to capture the distribution of arid ecosystems. The model representations of fire, soil, and processes unique to desert vegetation are discussed as possible explanations. The remaining three chapters deal with the assumption of full migration, where plants can arrive at any location regardless of distance or physical barriers. Using LPJ-GUESS, I imposed migration limitations by using fat-tailed seed dispersal kernels. I used three temperate tree species with different life history strategies to test the new dispersal functionality. Simulated migration rates for Acer rubrum (141 m year-1) and Pinus rigida (76 m year-1) correspond well to pollen and genetic reconstructed rates. However, migration rates for Tsuga canadensis (85 m year-1) were considerably slower than historical rates. A sensitivity analysis showed that maturation age is the most important parameter for determining rates of spread, but it is the dispersal kernel which determines if there is any long distance dispersal or not. The final study demonstrates how northerly refugia populations could have impacted landscape recolonization following the retreat of the last glacier. Using three species with known refugia (Acer rubrum, Fagus grandifolia, Picea glauca), colonization rates were faster with a northerly refugia population present. The number of refugia locations also had a positive effect on landscape recolonization rates, which was most pronounced when populations were separated. The results from this thesis illustrate the improvements made in vegetation-climate models, giving us increasing confidence in the quality of future climate change predictions.

Dans les modèles globaux de la biosphère continentale, toute la végétation mondiale est généralement représentée par une dizaine de grand groupes fonctionnels (PFT-Plant Functional Type), dont les caractéristiques (traits) sont fixes. Cette rigidité ne permet pas de représenter correctement l'évolution de la végétation face aux pressions environnementales et anthropiques grandissantes, et est à l'origine de nombreuses incertitudes pour l'estimation des cycles bio-géochimiques associés. Trois approches complémentaires axées sur l'utilisation des traits fonctionnels ont été explorées à l'aide du modèle dynamique global de végétation ORCHIDEE afin d'améliorer la représentation des PFTs forestiers. La première approche consiste à augmenter le nombre de PFTs à partir d'une classification hiérarchique des espèces. La seconde approche permet d'extrapoler les traits observés pour chaque PFT existant grâce à des relations empiriques calibrées à partir de plusieurs variables environnementales. La dernière approche utilise la théorie de la coordination de la photosynthèse afin d'estimer des distributions continues de traits en conditions optimales de photosynthèse. En parallèle, cette étude s'interroge sur les capacité d'un modèle global à représenter correctement les traits fonctionnels lorsqu'il est optimisé pour un flux de carbone. L'augmentation du nombre de PFTs permet d'améliorer significativement les caractéristiques et la représentativité spatiale des peuplements simulés de plus de 50 %. Les deux autres approches permettent d'estimer des distributions de traits réalistes et mettent en évidence un rôle ``tampon'' important de la plasticité des traits sur les flux de carbone futurs. Les trois approches abordées ont mis en évidence certaines faiblesses du modèle liées à la représentation de la phénologie, de l'allocation de la biomasse ou encore du stress hydrique pour les conifères. Ces résultats ont menés à la mise en place d'une représentation explicite des processus phénologiques pour les conifères sempervirents dans ORCHIDEE, qui à présent reproduit les dynamiques de LAI observées par télédétection. Enfin, le modèle ORCHIDEE ne peux pas être paramétré avec des observations directes de traits, privilégiant l'approche théorique pour simuler les distributions de traits. Cependant, l'assimilation de données d'observations de flux de carbone permet de faire le lien entre les traits mesurés à l'échelle foliaire et leur intégration à l'échelle de la canopée. Elle permet de retrouver des distributions de traits cohérentes avec les observations, ainsi que des relations trait-trait et trait-environnement qui sont observées à l'échelle foliaire. / Earth system models currently use a discretized representation of vegetation, grouping together the whole world species into a dozen of Plant Functional Types (PFT), whose characteristics (traits) are fixed. This rigidity does not allow to accurately represent the evolution of ecosystems and their associated bio-geochemical cycles, while vegetation is facing stronger environnemental and anthropogenic pressures. Three complementary approaches based on functional traits were tested in order to improve the representation of forests in the dynamic global vegetation model ORCHIDEE. Based on a hierarchical classification of species, the first approach is to increase the number of PFTs. The second approach extrapolates observed traits for each PFT from empirical relationships calibrated against different environmental variables. The last one uses the photosynthesis coordination theory which allows the estimation of functional traits in optimal photosynthesis conditions. In addition, this study explores the capacity of a global model to represent functional traits when optimized against a given carbon flux. Increasing the number of PFTs significantly improves the estimations of stand characteristics and their spatial distribution by more than 50 %. The two other approaches managed to reproduce realistics traits distributions and higlighted the ``buffer'' role of traits plasticity on futur carbon fluxes. Some weaknesses of the model linked to phenological processes, biomass allocation or hydric stress, emerged for conifers species. This led to the implementation of an explicit representation of the phenology for evergreen needleleaves PFTs in ORCHIDEE, which can now reproduce the LAI dynamic observed from remote sensing data. Finally, the ORCHIDEE model cannot be calibrated with in situ observations, emphasizing the theoretical approach to simulate continuous traits distributions. However, the assimilation of observed carbon fluxes allows to bridge the gap between traits measured at the leaf scale and the canopy. It reproduced consistent trait distributions with observations, and led to trait-trait and trait-environment relationships similar to those observed at the leaf scale.

<p>At present, carbon sequestration in terrestrial ecosystems slows the growth rate of atmospheric CO₂ concentrations, and thereby reduces the impact of anthropogenic fossil fuel emissions on the climate system. Changes in climate and land use affect terrestrial biosphere structure and functioning at present, and will likely impact on the terrestrial carbon balance during the coming decades - potentially providing a positive feedback to the climate system due to soil carbon releases under a warmer climate. Quantifying changes, and the associated uncertainties, in regional terrestrial carbon budgets resulting from these effects is relevant for the scientific understanding of the Earth system and for long-term climate mitigation strategies.</p> <p>A model describing the relevant processes that govern the terrestrial carbon cycle is a necessary tool to project regional carbon budgets into the future. This study (1) provides an extensive evaluation of the parameter-based uncertainty in model results of a leading terrestrial biosphere model, the Lund-Potsdam-Jena Dynamic Global Vegetation Model (LPJ-DGVM), against a range of observations and under climate change, thereby complementing existing studies on other aspects of model uncertainty; (2) evaluates different hypotheses to explain the age-related decline in forest growth, both from theoretical and experimental evidence, and introduces the most promising hypothesis into the model; (3) demonstrates how forest statistics can be successfully integrated with process-based modelling to provide long-term constraints on regional-scale forest carbon budget estimates for a European forest case-study; and (4) elucidates the combined effects of land-use and climate changes on the present-day and future terrestrial carbon balance over Europe for four illustrative scenarios - implemented by four general circulation models - using a comprehensive description of different land-use types within the framework of LPJ-DGVM.</p> <p>This study presents a way to assess and reduce uncertainty in process-based terrestrial carbon estimates on a regional scale. The results of this study demonstrate that simulated present-day land-atmosphere carbon fluxes are relatively well constrained, despite considerable uncertainty in modelled net primary production. Process-based terrestrial modelling and forest statistics are successfully combined to improve model-based estimates of vegetation carbon stocks and their change over time. Application of the advanced model for 77 European provinces shows that model-based estimates of biomass development with stand age compare favourably with forest inventory-based estimates for different tree species. Driven by historic changes in climate, atmospheric CO₂ concentration, forest area and wood demand between 1948 and 2000, the model predicts European-scale, present-day age structure of forests, ratio of biomass removals to increment, and vegetation carbon sequestration rates that are consistent with inventory-based estimates. Alternative scenarios of climate and land-use change in the 21^st century suggest carbon sequestration in the European terrestrial biosphere during the coming decades will likely be on magnitudes relevant to climate mitigation strategies. However, the uptake rates are small in comparison to the European emissions from fossil fuel combustion, and will likely decline towards the end of the century. Uncertainty in climate change projections is a key driver for uncertainty in simulated land-atmosphere carbon fluxes and needs to be accounted for in mitigation studies of the terrestrial biosphere.</p> / <p>Kohlenstoffspeicherung in terrestrischen Ökosystemen reduziert derzeit die Wirkung anthropogener CO₂-Emissionen auf das Klimasystem, indem sie die Wachstumsrate der atmosphärischer CO₂-Konzentration verlangsamt. Die heutige terrestrische Kohlenstoffbilanz wird wesentlich von Klima- und Landnutzungsänderungen beeinflusst. Diese Einflussfaktoren werden sich auch in den kommenden Dekaden auf die terrestrische Biosphäre auswirken, und dabei möglicherweise zu einer positiven Rückkopplung zwischen Biosphäre und Klimasystem aufgrund von starken Bodenkohlenstoffverlusten in einem wärmeren Klima führen. Quantitative Abschätzungen der Wirkung dieser Einflussfaktoren - sowie der mit ihnen verbundenen Unsicherheit - auf die terrestrische Kohlenstoffbilanz sind daher sowohl für das Verständnis des Erdsystems, als auch für eine langfristig angelegte Klimaschutzpolitik relevant.</p> <p>Um regionale Kohlenstoffbilanzen in die Zukunft zu projizieren, sind Modelle erforderlich, die die wesentlichen Prozesse des terrestrischen Kohlenstoffkreislaufes beschreiben. Die vorliegende Arbeit (1) analysiert die parameterbasierte Unsicherheit in Modellergebnissen eines der führenden globalen terrestrischen Ökosystemmodelle (LPJ-DGVM) im Vergleich mit unterschiedlichen ökosystemaren Messgrößen, sowie unter Klimawandelprojektionen, und erweitert damit bereits vorliegende Studien zu anderen Aspekten der Modelunsicherheit; (2) diskutiert unter theoretischen und experimentellen Aspekten verschiedene Hypothesen über die altersbedingte Abnahme des Waldwachstums, und implementiert die vielversprechenste Hypothese in das Model; (3) zeigt für eine europäische Fallstudie, wie Waldbestandsstatistiken erfolgreich für eine verbesserte Abschätzung von regionalen Kohlenstoffbilanzen in Wäldern durch prozessbasierten Modelle angewandt werden können; (4) untersucht die Auswirkung möglicher zukünftiger Klima- und Landnutzungsänderungen auf die europäische Kohlenstoffbilanz anhand von vier verschiedenen illustrativen Szenarien, jeweils unter Berücksichtigung von Klimawandelprojektionen vier verschiedener Klimamodelle. Eine erweiterte Version von LPJ-DGVM findet hierfür Anwendung, die eine umfassende Beschreibung der Hauptlandnutzungstypen beinhaltet. </p> <p>Die vorliegende Arbeit stellt einen Ansatz vor, um Unsicherheiten in der prozessbasierten Abschätzung von terrestrischen Kohlenstoffbilanzen auf regionaler Skala zu untersuchen und zu reduzieren. Die Ergebnisse dieser Arbeit zeigen, dass der Nettokohlenstoffaustausch zwischen terrestrischer Biosphäre und Atmosphäre unter heutigen klimatischen Bedingungen relativ sicher abgeschätzt werden kann, obwohl erhebliche Unsicherheit über die modelbasierte terrestrische Nettoprimärproduktion existiert. Prozessbasierte Modellierung und Waldbestandsstatistiken wurden erfolgreich kombiniert, um verbesserte Abschätzungen von regionalen Kohlenstoffvorräten und ihrer Änderung mit der Zeit zu ermöglichen. Die Anwendung des angepassten Modells in 77 europäischen Regionen zeigt, dass modellbasierte Abschätzungen des Biomasseaufwuchses in Wäldern weitgehend mit inventarbasierten Abschätzungen für verschiede Baumarten übereinstimmen. Unter Berücksichtigung von historischen Änderungen in Klima, atmosphärischem CO₂-Gehalt, Waldfläche und Holzernte (1948-2000) reproduziert das Model auf europäischer Ebene die heutigen, auf Bestandsstatistiken beruhenden, Abschätzungen von Waldaltersstruktur, das Verhältnis von Zuwachs und Entnahme von Biomasse, sowie die Speicherungsraten im Kohlenstoffspeicher der Vegetation. Alternative Szenarien von zukünftigen Landnutzungs- und Klimaänderungen legen nahe, dass die Kohlenstoffaufnahme der europäischen terrestrischen Biosphäre von relevanter Größenordnung für Klimaschutzstrategien sind. Die Speicherungsraten sind jedoch klein im Vergleich zu den absoluten europäischen CO₂-Emissionen, und nehmen zudem sehr wahrscheinlich gegen Ende des 21. Jahrhunderts ab. Unsicherheiten in Klimaprojektionen sind eine Hauptursache für die Unsicherheiten in den modellbasierten Abschätzungen des zukünftigen Nettokohlenstoffaustausches und müssen daher in Klimaschutzanalysen der terrestrischen Biosphäre berücksichtigt werden.</p>

Les prairies de montagne offrent de nombreux services ecosystémiques qui sont menacés par le changement global. Les traits fonctionnels constituent un outil prometteur pour caractériser les réponses des communautés à des changements de conditions environnementales et leurs répercussions sur les services associés. Cependant, des résulats de plus en plus nombreuses soulignent l’importance de la variabilité intra-spécifique des traits a également été mise en évidence. Pour étudier ces effets, je propose un nouveau modèle à base d’agents, MountGrass, qui combine la modélisation de communautés végétales riches en espèces avec des processus de plasticité phénotypique. Ces deux éléments au coeur du modèle sont associés grâce à des compromis d’allocation basés sur des patrons empiriques établis de stratégies d’utilisation des resources.Avec MountGrass, j’ai exploré l’impact de la plasticité phénotypique sur la croissance individuelle et les propriétés principales des communautés prairiales. À l’échelle individuelle, le modèle paramétré a révélé un fort impact positif de la plasticité phénotypique sur la croissance mais aussi sur la niche fondamentaledes espèces. Des phénomènes de convergence et de réduction de la sensibilité aux variations de conditionsexpliquent ces effets. À l’échelle des communautés, les simulations ont confirmé de forts effets de la plasticité sur la structure des communautés et leur diversité spécifique. Ces effets sont expliqués par l’effet combiné de la réduction du filtre abiotique et de la réduction des différences de compétitivité. Cependant, aucun effet majeur sur la stratégie dominante ou la productivité n’a pu être mis en évidence.Des implémentations alternatives ou des extensions du modèle devraient permettre de tester la robustesse des résultats obtenus et d’analyser d’autres schémas de dynamiques des communautés. En conclusion, ce travail ouvre la voie à une meilleure considération et une meilleure compréhension du rôle des variabilités intra-spécifiques dans les dynamiques des communautés végétales. / Mountain grasslands provide numerous ecosystem services that are likely to be impacted by global change. Plant functional traits hold great promise to succinctly characterise plant community response to changing environmental conditions and its effect on associated services; with growing evidence of the importance of intra-specific trait variability. I propose here a novel agent-based model, MountGrass, that combines the modelling of species rich grassland communities with phenotypic plasticity. These two key components are integrated via allocation trade-offs based on established empirical patterns of strategic differentiation in resource-use.With MountGrass, I explored the impact of phenotypic plasticity on individual plant growth and on main properties of grassland communities. At the individual level, the parametrised model revealed a strong impact of plasticity on growth and species’ fundamental niches, with potentially large impacts on community properties. These effects are explained by the convergence of species’ strategies and the reduction of the sensitivity to variable conditions. At the community level, simulations confirmed the strong effect of plastic allocation on community structure and species richness. These effects are driven by the cumulative effect of a reduction of both abiotic filtering and fitness differences between species. However, no clear effect on the dominant strategy or productivity could be detected.Going further, the robustness of these findings and other patterns of community dynamics should be analysed with alternative or extended implementations of MountGrass. In sum, this work opens a door towards a better integration and understanding of the role of the intra-specific variability in complex plant community dynamics.