Ecossistemas campestres encontram-se fortemente impactados por conversão de hábitat e por espécies exóticas invasoras. É necessário restaurar os ecossistemas campestres ao redor do globo. Entretanto, para os ecossistemas campestres brasileiros há poucas experiências de restauração e precisamos testar a viabilidade de técnicas normalmente empregadas em outros lugares. O objetivo deste estudo foi testar diferentes combinações de técnicas para restauração de campos invadidos por Urochloa decumbens nos Campos Sulinos, sul do Brasil. Combinaram-se duas técnicas de controle da espécie invasora (aplicação de herbicida e remoção superficial de solo) e duas técnicas de introdução de espécies nativas (transposição de feno e semeadura direta). Foram estabelecidos oito blocos em um experimento bifatorial no Morro Santana, Porto Alegre, combinando dois fatores e cada um com três tratamentos (duas técnicas mais o controle). A cobertura de vários grupos de espécies, riqueza de espécies e composição de espécies foram avaliadas por análises de variância, e então também por análise de coordenadas principais. Adicionalmente, a relação entre a cobertura da invasora e das espécies nativas foi investigada. As técnicas de controle da invasora mostraram-se eficientes tanto na redução da cobertura da espécie, como em permitir a entrada de espécies nativas na comunidade. Comparando as duas técnicas, aplicação de herbicida pareceu ser melhor do que a remoção superficial de solo, pois parcelas que tiveram a aplicação tiveram menor cobertura da invasora e maior riqueza de espécies. Já as técnicas de introdução de espécies mostraram-se insuficientes para adicionar espécies nativas na comunidade para competir com U. decumbens. Padrões de composição de espécies diferiram entre os tratamentos. Técnicas de controle do invasor diferiram grandemente do seu controle, que foi dominado por U. decumbens. Uma clara relação existe entre a cobertura da invasora e a presença e cobertura de espécies nativas. Então o controle da espécie invasora é fundamental para uma maior recuperação da vegetação. Entretanto, os resultados aqui apresentados correspondem a apenas oito meses após a finalização da implementação do experimento, e ações futuras de manejo na área deverão combinar novamente o controle da invasora com introdução de espécies nativas. / Grasslands ecosystems are strongly impacted by land use and invasive species. It is necessary restore these ecosystems around the world. However, there are few experiences with ecological restoration for the Brazilian grasslands and we need to test the viability of techniques normally used in others grasslands ecosystems. The aim of this study was to test different combinations of techniques to the ecological restoration of grasslands invaded by Urochloa decumbens in Campos, Southern Brazil. We combine two techniques to control the invasive species (herbicide application and topsoil removal) and two techniques to introduce native species (hay transfer and direct sowing). We established nine blocks in a bi-factorial experiment on Morro Santana, Porto Alegre, Rio Grande do Sul, combining two factors and each one with three treatments (two techniques plus the control). The coverage of various groups of species, species richness, and species composition were evaluated by variance analyses, and the later also by a principal ordination analysis. Additionally, the relation between invader coverage and native species was investigated. Both techniques to control invasive species have shown to be efficient to reduce the coverage of the invasive species, as well as to allow the arrival of native species. Specifically comparing them, the herbicide application seems to be a better treatment than the topsoil removal, once plots with herbicide had lower invasive species coverage and higher species richness. However, the species introduction techniques failed efficiently to add native species to the community composition and to compete with U. decumbens. Species composition patterns differed among the treatments. The invader control techniques greatly differed from their control, which was dominated by U. decumbens. A clear relationship exists between the invader coverage and the presence and coverage of native species. Thus, the control of the invasive species is fundamental to further vegetation recovery. Nevertheless, results here presented correspond to just eight months from the experiment implantation and future monitoring and management actions on the area should combine once more the control of the invasive species and the introduction of native species.

Ecossistemas campestres encontram-se fortemente impactados por conversão de hábitat e por espécies exóticas invasoras. É necessário restaurar os ecossistemas campestres ao redor do globo. Entretanto, para os ecossistemas campestres brasileiros há poucas experiências de restauração e precisamos testar a viabilidade de técnicas normalmente empregadas em outros lugares. O objetivo deste estudo foi testar diferentes combinações de técnicas para restauração de campos invadidos por Urochloa decumbens nos Campos Sulinos, sul do Brasil. Combinaram-se duas técnicas de controle da espécie invasora (aplicação de herbicida e remoção superficial de solo) e duas técnicas de introdução de espécies nativas (transposição de feno e semeadura direta). Foram estabelecidos oito blocos em um experimento bifatorial no Morro Santana, Porto Alegre, combinando dois fatores e cada um com três tratamentos (duas técnicas mais o controle). A cobertura de vários grupos de espécies, riqueza de espécies e composição de espécies foram avaliadas por análises de variância, e então também por análise de coordenadas principais. Adicionalmente, a relação entre a cobertura da invasora e das espécies nativas foi investigada. As técnicas de controle da invasora mostraram-se eficientes tanto na redução da cobertura da espécie, como em permitir a entrada de espécies nativas na comunidade. Comparando as duas técnicas, aplicação de herbicida pareceu ser melhor do que a remoção superficial de solo, pois parcelas que tiveram a aplicação tiveram menor cobertura da invasora e maior riqueza de espécies. Já as técnicas de introdução de espécies mostraram-se insuficientes para adicionar espécies nativas na comunidade para competir com U. decumbens. Padrões de composição de espécies diferiram entre os tratamentos. Técnicas de controle do invasor diferiram grandemente do seu controle, que foi dominado por U. decumbens. Uma clara relação existe entre a cobertura da invasora e a presença e cobertura de espécies nativas. Então o controle da espécie invasora é fundamental para uma maior recuperação da vegetação. Entretanto, os resultados aqui apresentados correspondem a apenas oito meses após a finalização da implementação do experimento, e ações futuras de manejo na área deverão combinar novamente o controle da invasora com introdução de espécies nativas. / Grasslands ecosystems are strongly impacted by land use and invasive species. It is necessary restore these ecosystems around the world. However, there are few experiences with ecological restoration for the Brazilian grasslands and we need to test the viability of techniques normally used in others grasslands ecosystems. The aim of this study was to test different combinations of techniques to the ecological restoration of grasslands invaded by Urochloa decumbens in Campos, Southern Brazil. We combine two techniques to control the invasive species (herbicide application and topsoil removal) and two techniques to introduce native species (hay transfer and direct sowing). We established nine blocks in a bi-factorial experiment on Morro Santana, Porto Alegre, Rio Grande do Sul, combining two factors and each one with three treatments (two techniques plus the control). The coverage of various groups of species, species richness, and species composition were evaluated by variance analyses, and the later also by a principal ordination analysis. Additionally, the relation between invader coverage and native species was investigated. Both techniques to control invasive species have shown to be efficient to reduce the coverage of the invasive species, as well as to allow the arrival of native species. Specifically comparing them, the herbicide application seems to be a better treatment than the topsoil removal, once plots with herbicide had lower invasive species coverage and higher species richness. However, the species introduction techniques failed efficiently to add native species to the community composition and to compete with U. decumbens. Species composition patterns differed among the treatments. The invader control techniques greatly differed from their control, which was dominated by U. decumbens. A clear relationship exists between the invader coverage and the presence and coverage of native species. Thus, the control of the invasive species is fundamental to further vegetation recovery. Nevertheless, results here presented correspond to just eight months from the experiment implantation and future monitoring and management actions on the area should combine once more the control of the invasive species and the introduction of native species.

Ecossistemas campestres encontram-se fortemente impactados por conversão de hábitat e por espécies exóticas invasoras. É necessário restaurar os ecossistemas campestres ao redor do globo. Entretanto, para os ecossistemas campestres brasileiros há poucas experiências de restauração e precisamos testar a viabilidade de técnicas normalmente empregadas em outros lugares. O objetivo deste estudo foi testar diferentes combinações de técnicas para restauração de campos invadidos por Urochloa decumbens nos Campos Sulinos, sul do Brasil. Combinaram-se duas técnicas de controle da espécie invasora (aplicação de herbicida e remoção superficial de solo) e duas técnicas de introdução de espécies nativas (transposição de feno e semeadura direta). Foram estabelecidos oito blocos em um experimento bifatorial no Morro Santana, Porto Alegre, combinando dois fatores e cada um com três tratamentos (duas técnicas mais o controle). A cobertura de vários grupos de espécies, riqueza de espécies e composição de espécies foram avaliadas por análises de variância, e então também por análise de coordenadas principais. Adicionalmente, a relação entre a cobertura da invasora e das espécies nativas foi investigada. As técnicas de controle da invasora mostraram-se eficientes tanto na redução da cobertura da espécie, como em permitir a entrada de espécies nativas na comunidade. Comparando as duas técnicas, aplicação de herbicida pareceu ser melhor do que a remoção superficial de solo, pois parcelas que tiveram a aplicação tiveram menor cobertura da invasora e maior riqueza de espécies. Já as técnicas de introdução de espécies mostraram-se insuficientes para adicionar espécies nativas na comunidade para competir com U. decumbens. Padrões de composição de espécies diferiram entre os tratamentos. Técnicas de controle do invasor diferiram grandemente do seu controle, que foi dominado por U. decumbens. Uma clara relação existe entre a cobertura da invasora e a presença e cobertura de espécies nativas. Então o controle da espécie invasora é fundamental para uma maior recuperação da vegetação. Entretanto, os resultados aqui apresentados correspondem a apenas oito meses após a finalização da implementação do experimento, e ações futuras de manejo na área deverão combinar novamente o controle da invasora com introdução de espécies nativas. / Grasslands ecosystems are strongly impacted by land use and invasive species. It is necessary restore these ecosystems around the world. However, there are few experiences with ecological restoration for the Brazilian grasslands and we need to test the viability of techniques normally used in others grasslands ecosystems. The aim of this study was to test different combinations of techniques to the ecological restoration of grasslands invaded by Urochloa decumbens in Campos, Southern Brazil. We combine two techniques to control the invasive species (herbicide application and topsoil removal) and two techniques to introduce native species (hay transfer and direct sowing). We established nine blocks in a bi-factorial experiment on Morro Santana, Porto Alegre, Rio Grande do Sul, combining two factors and each one with three treatments (two techniques plus the control). The coverage of various groups of species, species richness, and species composition were evaluated by variance analyses, and the later also by a principal ordination analysis. Additionally, the relation between invader coverage and native species was investigated. Both techniques to control invasive species have shown to be efficient to reduce the coverage of the invasive species, as well as to allow the arrival of native species. Specifically comparing them, the herbicide application seems to be a better treatment than the topsoil removal, once plots with herbicide had lower invasive species coverage and higher species richness. However, the species introduction techniques failed efficiently to add native species to the community composition and to compete with U. decumbens. Species composition patterns differed among the treatments. The invader control techniques greatly differed from their control, which was dominated by U. decumbens. A clear relationship exists between the invader coverage and the presence and coverage of native species. Thus, the control of the invasive species is fundamental to further vegetation recovery. Nevertheless, results here presented correspond to just eight months from the experiment implantation and future monitoring and management actions on the area should combine once more the control of the invasive species and the introduction of native species.

La restauration écologique est considérée comme un des moyens susceptibles d’enrayer la perte de la biodiversité. Les changements d’occupation du sol peuvent être des opportunités pour restaurer des écosystèmes dégradés par les activités agricoles. C’est notamment le cas du projet participatif du domaine du Cassaïre, situé dans le delta du Rhône, qui vise à recréer sur d’anciennes rizières des écosystèmes méditerranéens favorables à l’activité cynégétique. Deux écosystèmes sont plus particulièrement visés, les marais temporaires et les pelouses méso-xériques. Les objectifs de la thèse sont de mettre en évidence les principaux mécanismes concourant à l’installation d’une communauté végétale, de tester des techniques de restauration et d'en évaluer les conséquences pour les communautés végétales mais aussi pour d'autres compartiments de l'écosystème. En l’absence d’espèces cibles dans le pool régional d’espèces, l’introduction de ces espèces est nécessaire en addition de la restauration des conditions abiotiques. L’étrépage et le transfert de sol pour la communauté des marais permettent une augmentation des espèces cibles et de la similarité avec la communauté de référence. Cette technique apparaît moins pertinente pour la communauté d’invertébrés aquatiques. Le succès contrasté du transfert de sol souligne les risques à ne pas privilégier qu’un nombre restreint d’indicateurs de restauration, ne reflétant pas l’ensemble de l’écosystème. L’étrépage de sol suivi d’un transfert de foin semble être une combinaison pertinente pour recréer la communauté végétale de pelouses, même si les résultats obtenus, probablement en raison d’une compétition élevée, sont moins convaincants que pour les marais temporaires. Nos résultats obtenus pour la grande majorité en mésocosmes, s’ils ne concernent que les premières étapes de la restauration, mettent cependant en évidence des techniques de restauration qui paraissent pertinentes pour installer certaines composantes des deux écosystèmes de référence. Ces résultats, par leurs limites, suggèrent néanmoins de privilégier la conservation in situ des habitats naturels plutôt que de chercher à les restaurer après qu’ils aient été détruits. / Ecological restoration is considered as one approach to slow down the loss of biodiversity. Changes in land-uses may be an opportunity to restore ecosystems degraded by agricultural activities. This is the case of the participatory project of the Cassaïre site, located in the Rhône delta, which aims at recreating Mediterranean ecosystems favorable to hunting on former ricefields. Two ecosystems are targeted, temporary wetlands and meso-xeric grasslands. The aims of the thesis are to highlight the main drivers of plant community establishment, to test restoration techniques and to evaluate their effects on plant communities but also on other compartments of the ecosystem. In the absence of target species in the regional species pool, the introduction of these species is necessary in addition to the restoration of abiotic conditions. Topsoil removal and soil transfer for wetland communities allow an increase of target species and of similarity with the reference community. This technique appears to be less relevant for aquatic invertebrate community. The contrasted successful of soil transfer highlights the risks of favoring some indicators of restoration success, as they may not reflect the entire ecosystem. Topsoil removal and hay transfer seem to be a relevant combination to recreate grassland plant community, although the results obtained are less convincing than for temporary wetland, probably due to high competition. Our results, obtained in mesocosms, even if they relate only to the early stages of recovery, provide restoration techniques that seem relevant to establish some components of the two reference ecosystems. These results, by their limitations, however, suggest focusing on in situ conservation of natural habitats rather than trying to restore them after they were destroyed.Keywords: Biodiversity, Community ecology, Restoration

Global environmental changes, especially land-use changes, have profound effects on both ecosystem functioning and biodiversity, having already altered many ecosystem services. These losses emphasize the need to preserve what remains; however when conservation programs are not sufficient, restoring areas that have been destroyed or disturbed can improve conservation efforts and mitigate damages. This work focuses on campos rupestres, Neotropical grasslands found at altitudes, which are part of the Cerrado (Brazilian savannas). They host a great biodiversity with a high level of endemism and, like other mountain ecosystems, provide valuable ecosystem services, such as water purification and recreational services. They have been and still are being impacted by human activities, such as civil engineering construction, quarrying or mining. The first objective of this thesis was to describe the reference ecosystem in order to aim for a clear restoration target and to monitor progress and success. We show that campos rupestres are composed of at least two distinct plant communities (i.e. sandy and stony grasslands), each having a specific composition and structure, hosting a great biodiversity. Several phenological patterns occur among the herbaceous communities: the majority of species flowers and fruits appear during the rainy season but other patterns can be observed. During our 2-year survey, some dominant species belonging to Poaceae, among others, were not observed reproducing, which implies limited chances to disperse on degraded areas. Campo rupestre vegetation is not resilient following a strong disturbance: several years after the disturbance, almost no native species are encountered on the degraded areas, soils are completely altered and seed bank recomposes only with non-target ruderal species. According to the filter model, a local community is a subset of the regional species pool determined by a set of dispersal, abiotic and biotic filters. Acting on the different filters to influence the plant community was the core of our restoration interventions. We then applied three in-situ restoration protocols (hay transfer, species translocation and turf translocation) to restore both kinds of grassland. Hay transfer does not allow the restoration of campo rupestre vegetation because of soil alteration and mainly because of poor seed quality. Indeed, germination studies show that, while some Xyridaceae and Velloziaceae have a high germinability, some dominant Poaceae, Cyperaceae or Asteraceae species have embryoless, unviable or dormant seeds, which makes seeding less efficient. There is no evidence that fire-related cues enhance germination in campos rupestres. Species translocation is successful for only one species, Paspalum erianthum; for the others, root damages probably impede survival. Finally, turf translocation is the most successful method, since numerous species are re-introduced on degraded areas. However due to the low resilience of pristine campos rupestres where turfs are taken from, turf translocation can only be considered in the case of habitat rescue, in circumstances when complete habitat destruction is otherwise unavoidable. Face to the difficulty to restore these peculiar grasslands, the protection and the conservation of campos rupestres must be made a high priority

Les écosystèmes herbacés méditerranéens, caractérisés par une importante biodiversité, ont subi de considérables changements d‟usage. Désormais, leur conservation est devenue insuffisante au regard de l‟augmentation des dommages causés par les activités humaines. La restauration écologique apparaît donc comme un outil approprié pour participer à leur préservation. Des recherches préalables sur les différents compartiments de l‟écosystème ainsi que sur les facteurs responsables de l‟assemblage de leurs communautés avant et après dégradations sont donc nécessaires afin de mettre en place des méthodes de restauration adaptées à ces écosystèmes. Ainsi, en utilisant comme modèle biologique les Coussouls de Crau et leurs friches post-culturales (Bouches-du-Rhône, France), représentatif de nombreux écosystèmes steppiques méditerranéens, nous avons choisi de déterminer et d‟étudier les principaux filtres potentiellement responsables de l‟organisation de cette communauté végétale. Le pâturage ovin, les changements trophiques du sol, les phénomènes de compétition et la capacité de reproduction et/ou de dispersion des espèces, identifiés lors d‟une première étude sur la régénération de la steppe, ont ainsi été étudiés dans trois expérimentations différentes. Deux protocoles de restauration écologique in situ (transfert de foins et semis d‟espèces structurantes) et un protocole d‟expérimentation ex-situ testant les interactions d‟une sélection d‟espèces de la steppe ont permis de réaliser ces recherches. Nos principaux résultats montrent que les filtres de reproduction/dispersion, le niveau trophique du sol, la compétition et le régime de perturbations récurrentes lié au pâturage ovin ont un rôle déterminant dans les changements de la composition et de la structure de la communauté et ils expliquent ainsi en partie le ralentissement de sa dynamique successionnelle vers la steppe de référence. Nous avons également pu démontrer que ces filtres interagissaient entre eux. Au niveau de la recherche appliquée, les protocoles testés favorisent le retour de la végétation steppique, néanmoins la composition des parcelles restaurées reste toujours très différente de l‟écosystème de référence à très court terme (deux à trois années après la restauration). Cela démontre que même si nous comprenons mieux aujourd‟hui les facteurs expliquant l‟organisation de cette communauté, il est cependant toujours difficile de la restaurer dans son intégralité car il n‟est pas aisé de manipuler et de trouver un équilibre entre l‟action de ces différents facteurs. Il est donc désormais plus que nécessaire de poursuivre ces recherches fondamentales et appliquées notamment au niveau des processus de maturation et de structuration des vieilles communautés végétales herbacées. / Mediterranean herbaceous ecosystems which are characterized by a high biodiversity have been submitted to a lot of changes in use. Their conservation is not sufficient enough and ecological restoration seems to be an adapted solution to improve conservation efforts as it can mitigate damages induced by human activities. To develop restoration methods adapted to these ecosystems, preliminary researches on the different communities composing an ecosystem and the filters which are responsible for community assemblage before and after disturbance are necessary. Representative of numerous Mediterranean steppe ecosystems, the “Coussouls de Crau” and their ex-arable-fields (Bouches-du-Rhône, France) are the biological models used to identify and study the principal filters potentially responsible for organizing the plant communities. Sheep grazing, trophic changes, competition and reproduction and /or species dispersal, were identified in a first study on steppe regeneration; these filters were then the core of the following three research experiments. We carried out two in-situ restoration protocols (hay transfer and sowing foundation species) and one ex-situ experiment testing interactions between a selection of steppe species. The main results show that reproduction, seed dispersal, soil trophic level, competition, extensive sheep grazing are filters which have a determinant role in structuring the communities, in determining plant composition and partly in explaining the slow dynamics towards the reference steppe. Also, we show that filters interact between them. For applied restoration, the tested methods promote steppe vegetation come back, but in the short term (two or three years after restoration), the composition of restored areas remains different to the reference ecosystem. Even if we understand better the different factors which explain community organization, these results demonstrate that it is difficult to restore it integrally because it is not easy to manipulate and find an equilibrium between the different filters actions. Now it is necessary to continue these fundamental and applied researches, in particular on the maturation and structure of the old herbaceous plant communities.

La restauration écologique a été identifiée comme une approche permettant notamment de ralentir la perte de biodiversité et de maintenir tous les biens et services issus de cette biodiversité desquels dépend le bien être de notre civilisation actuelle. Cette restauration des écosystèmes se base sur des connaissances provenant à la fois de l'écologie des communautés et de l’écologie de la restauration. Les objectifs de la thèse sont donc de comprendre la dynamique d’une steppe méditerranéenne après changements d’usage ainsi que la mise en oeuvre de techniques qui pourraient être appliquées à la restauration de cet écosystème après une perturbation anthropique sévère. La thèse a pour objet d'étude la steppe méditerranéenne de la plaine de Crau, et notamment d’anciennes cultures pour étudier la recolonisation spontanée après perturbation et le projet de réhabilitation à grande échelle de Cossure pour les expérimentations sur les techniques de restauration. En ce qui concerne la dynamique après une perturbation anthropique exogène sévère, nous avons confirmé la faible résilience de la communauté végétale steppique à la fois à moyen (30-40 ans) et long terme (150 ans), tandis que les paramètres du sol et le taux d'infestation des mycorhizes sont résilients sur le long terme. En outre, nous avons confirmé le rôle joué par les trois filtres dans la recolonisation des communautés végétales. En ce qui concerne la steppe de la Crau, la recolonisation est déterminée en premier par le filtre abiotique, puis par le filtre de dispersion et enfin par le filtre biotique. Compte tenu de la faible résilience de la communauté, nous avons testé plusieurs techniques de restauration appliquées à grande échelle au sein du projet de réhabilitation de Cossure: le semis d’espèces nurses, l'étrépage de sol, le transfert de foin et le transfert de sol. Afin d'évaluer l'efficacité des techniques de restauration, nous avons développé des indices pour mesurer « l 'intégrité » de la structure de la communauté permettant de distinguer les abondances inférieures des abondances supérieures par rapport à la communauté de référence. Les meilleurs résultats ont été obtenus avec le transfert du sol, suivi par l’étrépage de sol, puis le semis d’espèces nurses et enfin le transfert de foin. Ces résultats ont toutefois confirmé la difficulté de restaurer totalement la communauté végétale steppique. Les recherches menées au sein de cette thèse montrent que les connaissances actuelles en matière de restauration écologique permettent de restaurer au moins partiellement certaines composantes de cet écosystème, mais suggèrent de mettre un maximum de moyens pour la conservation in situ des habitats naturels plutôt que de devoir les restaurer après qu'ils aient été détruits / Ecosystem restoration has been identified as one approach to slow down the loss of biodiversity and to protect all the biodiversity-based goods and services from which humankind benefits. Restoration feeds from knowledge coming from both community ecology and restoration ecology. The objectives of the thesis are to provide insights on both the dynamics of a mediterranean steppe after changes in land-use and the implementation of techniques which could be applied to restore this ecosystem after severe anthropogenic disturbances. The thesis takes as a study object the La Crau Mediterranean steppe, and especially former cultivated fields to study the recovery after cultivation and the Cossure large scale rehabilitation project to experiment rehabilitation and restoration techniques. Concerning dynamics after severe exogenous anthropogenic disturbances, we confirmed the low resilience of the steppe plant community both at mid- (30-40 years) and long-term (150 years) while the resilience of soil parameters and mycorrhizal infestation rate are effective on the long-term. Moreover we confirmed the role played by the three filters in the plant community recovery and found that for the La Crau steppe, this is firstly driven by the abiotic filter, then by the dispersion filter and finally by the biotic filter. Given this low resilience, we tested several restoration techniques applied at large-scale within the Cossure rehabilitation project: nurse species seeding, topsoil removal, hay transfer and soil transfer. In order to assess the efficiency of restoration techniques we developed indices to measure the community structure integrity, disentangling lower and higher abundances compared to the reference. The best results were obtained with soil transfer, followed by topsoil removal, then nurse species seeding and finally hay transfer. The research conducted for this thesis shows that current knowledge in ecological restoration makes it possible to restore at least partially some La Crau ecosystem components, but ought to lead us to understand the importance of in situ conservation of natural habitats as a better alternative to restore them after they were destroyed

Les changements globaux, notamment les changements d'usage des terres, modifient profondément le fonctionnement des écosystèmes et la biodiversité et, ont déjà impacté de nombreux services écosystémiques. La disparition de ces écosystèmes souligne la nécessité de préserver les zones intactes, et la restauration des zones détruites ou perturbées peut permettre de venir en appui aux efforts de conservation et minimiser les dommages. Ce travail a pour objet d’étude les campos rupestres, des pelouses néotropicales d’altitude, faisant partie du Cerrado (savane brésilienne), qui recèlent une importante biodiversité et qui, comme bien d'autres écosystèmes de montagne, fournissent de précieux services écosystémiques tels que la filtration de l’eau. Ils ont été, et sont encore, grandement affectés par les activités humaines (les travaux de génie civil, les carrières ou les mines). Le premier objectif de cette thèse était de décrire l'écosystème de référence, afin de définir clairement un objectif de restauration et mesurer les progrès et le succès de la restauration. Nous avons montré que les campos rupestres sont composés d'au moins deux communautés végétales distinctes (une avec un substrat caillouteux et l’autre avec un substrat sableux), chacune ayant une composition en espèces et une structure particulières ainsi qu’une grande biodiversité. La phénologie reproductive varie au sein des communautés herbacées: la majorité des espèces fleurissent et fructifient pendant la saison des pluies, d'autres se reproduisent en revanche durant la saison sèche. Tout au long de nos 2 années de suivis phénologiques, certaines espèces dominantes, notamment des Poaceae, n'ont pas été observées en fleur impliquant une dispersion limitée de ces espèces vers les zones dégradées. Les communautés végétales de campos rupestres ne sont pas résilientes aux fortes perturbations: plusieurs années après, presque aucune des espèces cibles n’ont été trouvées en zones dégradées, les sols ont complètement été modifiés et les banques de graines ne se sont recomposées qu’avec des espèces rudérales non désirées. Selon le modèle des filtres, une communauté résulte d’un pool régional d’espèce sélectionné par un ensemble de filtres : de dispersion, abiotique et biotique. Les interventions de restauration mises en place avaient pour but d’agir sur ces différents filtres afin de diriger la dynamique des communautés végétales. Nous avons mis en place trois protocoles de restauration in-situ (le transfert de foin, la translocation d’espèce et la translocation de plaque de végétation) pour restaurer les deux types de communautés de campos rupestres identifiées. Le transfert de foin n’a pas permis la restauration des communautés végétales de campos rupestres en raison de l’importante altération des sols et, surtout, à cause de la mauvaise qualité des graines. En effet, nos études de germination ont montré que, alors que certaines espèces de Xyridaceae et Velloziaceae germent très bien, certaines espèces dominantes de Poaceae, de Cyperaceae ou d’Asteraceae ont des graines soit vides, soit non viables, soit dormantes ; le semis se révèle alors peu efficace. Nous n’avons pas mis en évidence d’effet positif du feu sur la germination des espèces de campos rupestres. La translocation d'espèces s’avère un succès pour une seule espèce, Paspalum erianthum. Pour les autres, les dommages causés au niveau des racines lors de la translocation limitent probablement leur survie. Enfin la translocation de plaques de végétation s’avère être la méthode la plus efficace permettant à de nombreuses espèces d’être réintroduites en zones dégradées. En raison de la faible résilience des campos rupestres dans lesquels les plaques de végétation ont été prélevées, cette méthode ne peut être envisagée que pour sauver des habitats dans le cas extrême où leur destruction est inévitable. Face à la difficulté de restaurer les campos rupestres, leur protection et leur conservation doit être une priorité / Global environmental changes, especially land-use changes, have profound effects on both ecosystem functioning and biodiversity, having already altered many ecosystem services. These losses emphasize the need to preserve what remains; however when conservation programs are not sufficient, restoring areas that have been destroyed or disturbed can improve conservation efforts and mitigate damages. This work focuses on campos rupestres, Neotropical grasslands found at altitudes, which are part of the Cerrado (Brazilian savannas). They host a great biodiversity with a high level of endemism and, like other mountain ecosystems, provide valuable ecosystem services, such as water purification and recreational services. They have been and still are being impacted by human activities, such as civil engineering construction, quarrying or mining. The first objective of this thesis was to describe the reference ecosystem in order to aim for a clear restoration target and to monitor progress and success. We show that campos rupestres are composed of at least two distinct plant communities (i.e. sandy and stony grasslands), each having a specific composition and structure, hosting a great biodiversity. Several phenological patterns occur among the herbaceous communities: the majority of species flowers and fruits appear during the rainy season but other patterns can be observed. During our 2-year survey, some dominant species belonging to Poaceae, among others, were not observed reproducing, which implies limited chances to disperse on degraded areas. Campo rupestre vegetation is not resilient following a strong disturbance: several years after the disturbance, almost no native species are encountered on the degraded areas, soils are completely altered and seed bank recomposes only with non-target ruderal species. According to the filter model, a local community is a subset of the regional species pool determined by a set of dispersal, abiotic and biotic filters. Acting on the different filters to influence the plant community was the core of our restoration interventions. We then applied three in-situ restoration protocols (hay transfer, species translocation and turf translocation) to restore both kinds of grassland. Hay transfer does not allow the restoration of campo rupestre vegetation because of soil alteration and mainly because of poor seed quality. Indeed, germination studies show that, while some Xyridaceae and Velloziaceae have a high germinability, some dominant Poaceae, Cyperaceae or Asteraceae species have embryoless, unviable or dormant seeds, which makes seeding less efficient. There is no evidence that fire-related cues enhance germination in campos rupestres. Species translocation is successful for only one species, Paspalum erianthum; for the others, root damages probably impede survival. Finally, turf translocation is the most successful method, since numerous species are re-introduced on degraded areas. However due to the low resilience of pristine campos rupestres where turfs are taken from, turf translocation can only be considered in the case of habitat rescue, in circumstances when complete habitat destruction is otherwise unavoidable. Face to the difficulty to restore these peculiar grasslands, the protection and the conservation of campos rupestres must be made a high priority