Les interactions entre la convection humide et la dynamique de grande échelle sont au cœur du climat tropical et de sa variabilité. Les processus associés aux nuages convectifs, tels que la condensation, l'évaporation, les processus radiatifs, et le transport d'énergie à petite échelle associé, génèrent des gradients de température dans l'atmosphère. Ces derniers engendrent des circulations de grande échelle qui contrôlent les distributions spatio-temporelles d'énergie et d'humidité, et donc en retour celle de la convection. Ces interactions forment probablement l'un des problèmes scientifiques majeurs de la modélisation de l'atmosphère. L'objectif de cette thèse est d'analyser la représentation de ces interactions dans le modèle de climat ARPEGE-Climat Version 6 et de comprendre le rôle qu'elles peuvent jouer dans les principaux biais du modèle sous les tropiques, notamment un biais froid de plusieurs degrés en moyenne et haute troposphère et un biais sec vers 850 hPa. Les impacts des processus convectifs sur la grande échelle sont souvent caractérisés par deux quantités, la source de chaleur apparente Q1 et le puits d'humidité apparent Q2 . Bien que difficilement observables, ces deux quantités peuvent être estimées en déployant un réseau de radiosondages permettant de déterminer les différents termes des bilans d'eau et d'énergie sur un quadrilatère donné. Un tel dispositif a été mis en œuvre d'octobre à décembre 2011 pendant la campagne CINDY2011/DYNAMO au cœur de l'Océan Indien. Les observations collectées et les données de Q1 et Q2 dérivées ont été utilisées dans cette thèse pour (i) caractériser le cycle de vie de la convection et (ii) mettre en place une configuration unicolonne du modèle ARPEGE-Climat sur les quadrilatères Nord et Sud du domaine CINDY2011/DYNAMO. Les résultats ont montré que le modèle ARPEGE-Climat est capable de reproduire de manière satisfaisante les transitions entre régimes de convection peu profonde, profonde et stratiforme, malgré une nette sous-estimation du flux d'évaporation en surface et de l'activité convective sur le domaine nord. Le modèle reproduit plus difficilement l'humidification de la troposphère pendant les phases de cumulus peu profonds. Les résultats obtenus dans ce cadre unicolonne ont ensuite été confrontés à des configurations 3D du modèle ARPEGE-Climat, à la fois en mode AMIP où le modèle est seulement forcé par les températures de la mer observées, et en mode "Transpose-AMIP" où le modèle est de plus initialisé à partir d'états réalistes de l'atmosphère. L'analyse de la dérive systématique du modèle dans ces simulations Transpose-AMIP a permis de montrer que les biais obtenus en mode AMIP étaient associés à des processus rapides (quelques jours). Ces biais sont généralement aussi très similaires à ceux documentés dans le cadre unicolonne. L'origine des biais thermodynamiques est analysée plus en détail, soulignant un rôle important des régimes de convection profonde, notamment dans sa phase stratiforme, pour le biais froid de la haute troposphère, et des défauts importants dans les régimes de cumulus peu profond et de congestus pour les biais d'humidité. Ces régimes mériteront une attention particulière dans les prochains développements de la physique d'ARPEGE-Climat. / The tropical climate and its variability at multiple timescales are dominated by interactions between moist convection and the large-scale atmospheric circulation. Small-scale processes associated with convective clouds such as condensation and evaporation, radiation, and vertical mixing all contribute to atmospheric temperature gradients which generate large-scale circulations. Such circulations exert a control on the spatio-temporal distribution of energy and humidity within the tropical atmosphere and, in turn, on moist convection. These twoway interactions represent one of the most difficult scientific challenge for global atmospheric modelling. The main objective of the present thesis is to analyse the representation of these interactions in Version 6 of the ARPEGE-Climat atmospheric general circulation model and to understand their possible contribution to the main model biases in the tropics, especially a cold bias in the mid and upper troposphere and a dry bias around 850 hPa. The large-scale impacts of moist convection are often characterized by two quantities, the apparent heat source, Q1, and the apparent moisture sink, Q2. Although difficult to observe, these two quantities can be estimated by deploying a sounding array of sufficient density to compute the different terms of the water and energy budgets over a selected domain. Such a strategy was implemented between October and December 2011 during the CINDY2011/DYNAMO field campaign in the middle of the tropical Indian Ocean. The collected observations and the derived Q1 and Q2 estimates are used in the present thesis to (i) characterize the life cycle of the tropical convection and (ii) set up a single column configuration of the ARPEGE-Climat model on the northern and southern domains of the campaign. Results show that the model is able to capture satisfactorily the transitions between different convective regimes, from shallow to deep and stratiform, despite a strong undestimation of surface evaporation and of the overall convective activity over the northern domain. The model however shows some difficulties at simulating the troposphere moistening during the shallow cumulus regime. The single column model results are then compared to 3D configurations of the ARPEGEClimat model, both in AMIP mode where the model is only driven by observed sea surface temperatures, and in " Transpose-AMIP " mode where the model is also initialized from realistic atmospheric conditions. Through the analysis of the systematic atmospheric drift across these Transpose-AMIP integrations, the dominant contribution of fast (within a few days) processes to the model biases found in AMIP mode is highlighted. Such biases also show some similarity with the errors simulated in the single-column framework. A more detailed analysis of the model systematic errors reveals a strong contribution of deep convection, especially in its stratiform regime, to the cold bias in the upper troposphere, and of deficiencies in the shallow cumulus regime to the moisture biases. These regimes will therefore deserve a particular attention during the next phase of development of the ARPEGE-Climat model.

Radiative feedbacks associated with changes in water vapor, temperature, surface albedo and clouds remain a major source of uncertainty in our understanding of climate's response to anthropogenic forcing. In this dissertation climate model data is used to investigate variations in feedbacks that result from changing CO��� forcing and the time scales on which feedbacks operate, focusing on the applicability of one method in particular, the radiative kernel technique, to these problems. This computationally efficient technique uses a uniform, incremental change in feedback variables to infer top-of-atmosphere (TOA) radiative flux changes. The first chapters explore the suitability of the linear radiative kernel technique for large forcing scenarios. We show that kernels based on the present-day climate misestimate TOA flux changes for large perturbations, translating into biased feedback estimates. We address this issue by calculating additional kernels based on a large forcing climate state with eight times present day CO��� concentrations. Differences between these and the present-day kernels result from added absorption of radiation by CO��� and water vapor, and increased longwave emission due to higher temperatures. Combining present-day and 8xCO��� kernels leads to significant improvement in the approximation of TOA flux changes and accuracy of feedback estimates. While climate sensitivity remains constant with increasing CO��� forcing when the inaccurate present-day kernels are used, sensitivity increases significantly when new kernels are used. Comparison of feedbacks in climate models with observations is one way towards understanding the disagreement among models. However, climate change feedbacks operate on time scales that are too long to be evaluated from the observational record. Rather, short-term proxies for greenhouse-gas-driven warming are often used to compute feedbacks from observations. The third chapter of this dissertation examines links between the seasonal cycle and global warming using pattern correlations of spatial distribution of feedback variables and radiative flux changes. We find strong correlations between time scales for changes in surface temperature and climate variables, but not for TOA flux anomalies, reaffirming conclusions drawn in previous work. Finally, we investigate the fitness of the radiative kernel technique for evaluation of short-term feedbacks in a comparison with the more accurate, but more computationally expensive, partial radiative perturbations. / Graduation date: 2013

The background state and changes associated with the El Niño-Southern Oscillation (ENSO) in the tropical Pacific Ocean influence climate patterns all over the world. Understanding how the tropical Pacific will be impacted by climate change is therefore critical to accurate regional climate projections. However, sparse historical data coverage and strong natural variability in the basin make it difficult to assess the response of the tropical Pacific to anthropogenic climate change. Further, climate models disagree regarding the response of the basin to continued anthropogenic forcing into the future. Building off of the limited instrumental record, high-resolution records from coral and lake sediment archives can be used to assess the response of the tropical Pacific to past climate changes and to compare and assess climate model projections. In the present study, I use high-resolution coral and lake records from the equatorial Pacific to assess climate model projections and the response of the coupled ocean-atmospheric climate system in the basin (ocean temperature, salinity, winds, precipitation) to natural and anthropogenic forcing. Using a simple model of how climate is recorded by corals, we compare historical climate data and climate model simulations with coral paleoclimate records to assess climate model projections and address uncertainties in the historical data, models and paleoclimate records. We demonstrate that this simple model is able to capture variability and trend observed in the coral records, and show that the both sea surface temperature and salinity contribute to the observed coral trend. However, we find major discrepancies in the observed and climate model simulated trends in the tropical Pacific that may be attributed to uncertainties in model simulated salinity. We then assess 20th-century variability and trends in SST and salinity in the central tropical Pacific using replicated coral δ¹⁸O and Sr/Ca records from the Republic of Kiribati and the central Line Islands. We find that the coral records from these sites display a warming and freshening trend superimposed on strong interannual and low-frequency variability. Further, we demonstrate an apparent strengthening of the E-W SST gradient across the dateline (between 173°E and 160°W) and a slight weakening of the N-S SST gradient due to enhanced warming along the equator and west of the dateline relative to other sites. However, we find no evidence of increased variability in the central Pacific, suggesting that there has not been an increase in central Pacific style ENSO events. Finally, we show that the salinity response to climate change may be very patchy within the basin. Using a new ~90 year coral Mn/Ca record from the central Pacific, we investigate variability and trends in tropical Pacific trade winds. First, we demonstrate a strong association between westerly wind anomalies and coral skeletal Mn/Ca, which recorded all of the major historical El Niño events of the 20th century. In this new long Mn/Ca record, we find a reduction in the amplitude and frequency of Mn/Ca pulses between 1893 and 1982, suggesting a decrease in westerly wind anomalies in the western equatorial Pacific Ocean. Finally, we use a sediment record from Bainbridge Crater Lake, Galápagos Archipelago to assess variability in the eastern tropical Pacific over the past ~6 thousand years. Based on results from long-term monitoring of the lake, we propose a new climate interpretation of the sediment record and find further evidence reduced mid-Holocene ENSO variability and a ramp up of ENSO variability starting around 1775 cal. years BP.

A changing climate can severely perturb regional hydrology and thereby affect human societies and life in general. To assess and simulate such potential hydrological climate change impacts, hydrological models require reliable meteorological variables for current and future climate conditions. Global climate models (GCMs) provide such information, but their spatial scale is too coarse for regional impact studies. Thus, GCM output needs to be downscaled to a finer scale either through statistical downscaling or through dynamic regional climate models (RCMs). However, even downscaled meteorological variables are often considerably biased and therefore not directly suitable for hydrological impact modeling. This doctoral thesis discusses biases and other challenges related to incorporating climate model output into hydrological studies and evaluates possible strategies to address them. An analysis of possible sources of uncertainty stressed the need for full ensembles approaches, which should become standard practice to obtain robust and meaningful hydrological projections under changing climate conditions. Furthermore, it was shown that substantial biases in current RCM simulations exist and that correcting them is an essential prerequisite for any subsequent impact simulation. Bias correction algorithms considerably improved RCM output and subsequent streamflow simulations under current conditions. In addition, differential split-sample testing was highlighted as a powerful tool for evaluating the transferability of bias correction algorithms to changed conditions. Finally, meaningful projections of future streamflow regimes could be realized by combining a full ensemble approach with bias correction of RCM output: Current flow regimes in Sweden with a snowmelt-driven spring flood in April will likely change to rather damped flow regimes that are dominated by large winter streamflows.

In the analysis of the signals of natural origin, we are often confronted with situations where we do not know if a change occurred, or where the possible point of change can be located(localized). However, diverse methods in signal processing rest(base) implicitly on a hypothesis of stationarity, because the still case is defined well in a theoretical prospect(perspective). On the other hand, all the processes of the real world are a priori non-still and, in the majority of the cases, this supposition shows itself true. Given that there are numerous manners by which the property of stationarity can be broken, various tests of stationarity were developed to test the various forms of non-stationarity. This thesis(theory) concentrates on the conception(design)

This thesis addresses a series of questions related to the problem of achieving reliable and physically consistent representations of aerosol-cloud interaction in global circulation models (GCM). In-situ data and modeling tools are used to develop and evaluate novel parameterization schemes for the process of aerosol activation for applications in GCM simulations. Atmospheric models of different complexity were utilized, ranging from detailed Lagrangian parcel model simulations of the condensational growth of droplets, to one-dimensional single column model with aerosol and cloud microphysics, and finally GCM simulations performed with the Community Atmosphere Model (CAM). A scheme for mapping the sub-grid scale variability of cloud droplet number concentrations (CDNC) to a number of microphysical process rates in a GCM was tested, finding that neglecting this impact can have substantial influences in the integrated cloud properties. A comprehensive comparison and evaluation of two widely used, physically-based activation parameterizations was performed in the framework of CAM5.1. This was achieved by utilizing a numerical adjoint sensitivity approach to comprehensively investigate their response under the wide range of aerosol and dynamical conditions encountered in GCM simulations. As a result of this, the specific variables responsible for the observed differences in the physical response across parameterizations are encountered, leading to further parameterization improvement.

In this project, we analyse the ability of two regional climate models to simulate meteorological conditions on Nordenskiöldbreen, a glacier in Svalbard. To do so, regional climate model output is compared with in situ measurements from an automatic weather station. Detailed information about the weather conditions on Nordenskiöldbreen is important for simulating the glacial mass balance in a changing climate. The parameters analysed were the following: temperature, air pressure, relative humidity, precipitation, cloud cover, wind speed and wind direction. The weather station did not measure all parameters, cloud cover was instead estimated through the incoming longwave radiation and temperature, while precipitation was calculated from snow depth. The results show that the models represent certain parameters better than others. Temperature, air pressure and wind speed and direction are found to be simulated with high precision. Poorest agreement is found for precipitation, which appears to be both difficult to simulate and observe. Relative humidity and cloud cover show average agreement with the station. The conclusion of the project is that the estimation of some of the parameters is satisfactory, while others are lacking. None of the models can be determined to have performed significantly better than the other. / I det här projektet analyserades två regionala klimatmodellers förmåga att simulera meteorologiska förhållanden på Nordenskiöldbreen, en glaciär på Svalbard. Detta gjordes genom jämförelser av data från regionala klimatmodeller mot lokala mätningar från en automatisk väderstation. Detaljerad information om väderförhållandena på Nordenskiöldbreen är viktigt för att kunna simulera glaciärens massbalans i ett föränderligt klimat. Parametrarna som jämfördes var temperatur, lufttryck, relativ luftfuktighet, nederbörd, molntäcke samt vindhastighet och vindriktning. Stationen mätte inte alla parametrar, molntäcket uppskattades istället genom inkommande långvågig strålning och temperatur, medan nederbörd beräknades via snödjup. Resultatet visar att modellerna representerar vissa parametrar bättre än andra. Temperatur, lufttryck, vindhastighet och vindriktning simuleras med hög precision. Parametern med lägst samband är nederbörd, somverkar vara svår både att simulera och observera. Relativ luftfuktighet och molntäcke har ett medelmåttigt samband till stationen. Slutsatsen av projektet är att modellernas uppskattning av några parametrar är tillräckligt bra, medan andra är bristfälliga. Ingen av modellerna kan bedömas ha presterat signifikant bättre än den andra.

Mudanças climáticas podem afetar a distribuição espacial e temporal das variáveis hidrológicas, tendo como consequências alterações nos regimes de precipitação e vazão dos rios. Aumentos ou reduções no volume de escoamento de uma bacia hidrográfica podem, por exemplo, produzir danos aos ecossistemas, afetar a produção de alimentos, abastecimento de água, navegação e geração de energia. Atualmente buscam-se relações que permitam entender os processos de mudanças climáticas a fim avaliar os impactos e mitigá-los, assim como avaliar as incertezas inerentes ao processo de modelagem hidrológica de tais mudanças. Neste contexto este trabalho apresenta uma metodologia de quantificação e análise de incertezas para estudos de mudanças climáticas, tomando como estudo de caso a bacia hidrográfica do Rio Uruguai (BHRU) com área aproximada de 110,000 Km². Para tanto três fontes de incerteza foram analisadas e comparadas: o modelo hidrológico, técnicas de remoção de viés e modelos climáticos. O modelo hidrológico MGB-IPH foi avaliado quanto ao processo de parametrização, utilizando diferentes períodos de simulação para calibração: (i) Período de calibração 1 – MGB/P1: representando a série completa de observações de 1960-1990 com verificação no período de 1992-1999; (ii) Período de calibração 2 - MGB/P2: calibração em período seco e verificação de período de cheias (iii) Período de calibração 3 – MGB/P3: calibração em período característico de cheias e verificação de período de estiagem. Três diferentes técnicas de remoção de viés foram aplicadas para analisar o grau de incerteza que a escolha de um determinado método de correção pode agregar ao resultado final: (i) RV1 - Técnica de Mapeamento Quantil-Quantil; (ii) RV2 - Técnica de Escalonamento Linear e (iii) RV3 - Técnica Delta change. Os modelos climáticos globais (GMC’s) foram analisados quanto a sua estrutura, comparando projeções de cinco diferentes modelos: MPEH5 (ECHAM5/MPIOM), GFCM21 (GFDL-CM2.1), MRCGCM (MRI-CGCM2.3.2), HADCM3 (UKMO-HadCM3) e NCCCSM (CCSM3). Adicionalmente, também foram analisadas as projeções climáticas de cinco diferentes versões do modelo climático regional (RCM) ETA/CPTEC: CT20, CT40, LOW, MID e HIGH. Inicialmente os resultados das simulações provenientes de cada uma destas fontes foram comparados de maneira isolada e em seguida de maneira combinada. Portanto, a metodologia foi dividida em Etapa (1) e Etapa (2). A Etapa (1) teve por objetivo responder a seguinte pergunta: Qual dentre as fontes de incerteza selecionadas agrega maior variação ao resultado final? Ou seja, qual destas fontes propaga maior incerteza em termos de impactos de mudanças climáticas na BHRU? Os resultados obtidos por cada uma das fontes de incerteza foram comparados em termos de anomalias de vazões médias de longo período (QMLP), máximas e mínimas anuais. Na Etapa (2) foi realizada a análise total de incerteza, ou seja, a análise combinada dos resultados obtidos na Etapa (1). As anomalias de vazões foram apresentadas utilizando as curvas de distribuição acumulada (CDF’s) e a incerteza total expressa pela diferença entre os percentis 5% e 95%. Considerando os resultados obtidos para as vazões médias de longo período (QMLP), as fontes podem ser ordenadas de forma decrescente, em relção ao grau de incerteza que propagam: modelos climáticos globais > modelos climáticos regionais > técnicas de remoção de viés > modelo hidrológico. Para as vazões extremas os RCM’s apresentam as maiores variações de anomalias se comparadas às dos modelos hidrológicos e técnicas de remoção de viés, inclusive para ambos os extremos, máximos e mínimos. Esta variação se dá principalmente, pelos resultados de LOW e MID. Estas informações podem ajudar os gestores e tomadores de decisão no adequado gerenciamento e planejamento dos recursos hídricos sob condições de mudanças climáticas, assim como o entendimento da incerteza associada. / Climate change can affect the spatial and temporal distribution of hydrological variables, with the consequences of changes in precipitation regimes and river flows. Increase or decrease the flow of rivers, for example, can cause damage to ecosystems, affecting food production, water supply, navigation and power generation. Currently seeking to relationships that allow understand climate change processes in order to assess the impacts and mitigate them, and assess the uncertainties inherent in hydrologic modeling process of such climate change. This thesis aimed at the development of a methodology for quantification and analysis of uncertainties for climate change studies in hydrology , taking as a case study the basin of the Uruguay River (BHRU) with a drainage area near 110,000 km². For that three sources of uncertainty were analyzed and compared: the hydrologic model, bias removal techniques and climate models. The hydrological model MGB-IPH was evaluated for parameterization, using different simulation periods for calibration: (i) MGB /P1: full range with calibration period (1960-1990) and validate (1992-1999); (ii) MGB / P2: calibrated in the period of dry and validated in the flood season (iii) MGB/P3: calibrated in the period of floods and validated in the dry season. Three different bias correction methods were applied to analyze the degree of uncertainty that the choice of a particular method of correction can add to the final result: (i) RV1 - Quantil-Quantil Mapping; (ii) RV2 - Linear Scaling, and (iii) RV3 - Delta Change Technique. Global climate models (GMC's) were analyzed for their structure, comparing projections of five different models: MPEH5 (ECHAM5/MPI-OM), GFCM21 (GFDLCM2.1), MRCGCM (MRI-CGCM2.3.2), HADCM3 (UKMO-HadCM3) e NCCCSM (CCSM3). Additionaly, climate projections from five different versions of the regional climate model (RCM) ETA / CPTEC were also analyzed: CT20, CT40, LOW, MID e HIGH. Initially the simulation results from each of the sources of uncertainty were compared individually (single propagation) and then in a combined way (multi propagation). Therefore, the methodology was divided in Step (1) and Step (2). Step (1) aimed to answer: Which of the selected sources of uncertainty adds more variation to the final result? Which of these sources propagates greater uncertainty in terms of impacts of climate change on BHRU? The results for each of the sources of uncertainty were compared in terms of long-term mean flow (QMLP), maximum and minimum annual flow. In Step (2) total uncertainty analysis was performed, therefore the combined analysis of the results obtained in Step (1). The anomalies in discharge were presented using the cumulative distribution function (CDF's) and the total uncertainty expressed by the difference between the percentiles 5% and 95%. Throughout the application of the proposed methodology it was concluded that: (i) for the extremes (maximum and minimum) annual discharges the largest source of uncertainty are the projections of the RCM's, followed by the the bias removal technique and finally the hydrological model; (ii) for the QMLP the largest source of uncertainty are followed global climate models, then the regional climate models. This information can help managers and decision makers in the proper management and planning of water resources under climate change conditions, as well as the understanding of the associated uncertainty.

Dans l’analyse des signaux d’origine naturelle, nous sommes souvent confrontés à des situations où nous ne savons pas si un changement s’est produit, ni où le possible point de changement peut être localisé. Cependant, diverses méthodes en traitement du signal reposent implicitement sur une hypothèse de stationnarité, car le cas stationnaire est bien défini dans une perspective théorique. D’un autre côté, tous les processus du monde réel sont a priori non-stationnaires et, dans la majorité des cas, cette supposition se révèle vraie. Etant donné qu’il existe de nombreuses façons par lesquelles la propriété de stationnarité peut être enfreinte, différents tests de stationnarité ont été développés pour tester les différentes formes de non-stationnarité. Cette thèse se concentre sur la conception et l’amélioration des techniques qui peuvent être appliquées aux signaux environnementaux, plus spécifiquement, les signaux hydrométéorologiques. Les techniques qui ont été développées présentent certaines caractéristiques qui sont préférables pour tester les données environnementales (i.e. être non-parametrique, être capable d’extraire automatiquement les informations des données disponibles, être capable d’identifier un changement dans les moments statistiques du premier et du second ordre). Dans cette thèse, le test de stationnarité et la détection de point de changement ont été abordés séparément: les tests de stationnarité rejettent la stationnarité de tout l’intervalle d’observation, tandis que pour détecter les points de changement, nous testons les signaux pour les quels la stationnarité a déjà été rejetée. Dans cette thèse, de nombreuses contributions et de nouvelles approches de ces sujets sont proposées. La dernière partie de la thèse consiste à appliquer toutes les approaches développées sur des données environnementales. Les données ont été générées par le Canadian Regional Climate Model (CRCM), un modéle très réaliste qui prend en compte de nombreuses interactions physiques complexes.La cohérence des résultats obtenus confirme le potentiel des approches proposées au regard des approches concurrentes. / In the analysis of the signals of natural origin, we are often confronted with situations where we do not know if a change occurred, or where the possible point of change can be located(localized). However, diverse methods in signal processing rest(base) implicitly on a hypothesis of stationarity, because the still case is defined well in a theoretical prospect(perspective). On the other hand, all the processes of the real world are a priori non-still and, in the majority of the cases, this supposition shows itself true. Given that there are numerous manners by which the property of stationarity can be broken, various tests of stationarity were developed to test the various forms of non-stationarity. This thesis(theory) concentrates on the conception(design)

Au cours du XXIème siècle, une augmentation des précipitations est attendue dans les régions polaires. En Antarctique, cette augmentation devrait se traduire par une accumulation de neige sur le continent, contribuant ainsi positivement au bilan de masse de la calotte polaire, et par conséquent négativement au niveau des mers. Les modèles utilisés pour simuler le climat du XXIème siècle prédisent presque tous une augmentation des précipitations en Antarctique, mais l'importance de ce changement diffère fortement d'un modèle à l'autre. De plus, les taux de précipitation actuels reproduits par ces mêmes modèles divergent également beaucoup. Cependant, faute d'observation fiable de précipitation en Antarctique, il était jusqu'à présent difficile de vérifier la capacité des modèles à simuler ces dernières. Dans cette étude, les données issues du radar météorologique embarqué à bord du satellite CloudSat ont été utilisées afin de produire la première climatologie de précipitation en Antarctique à partir d'observations. Cette climatologie couvre la période août 2006 - avril 2011, et a montré de très bons accords avec les réanalyses ERA Interim qui n'utilisent pas d'observations issues de CloudSat. Le taux de chute de neige obtenu avec CloudSat sur le continent Antarctique jusqu'à 82°S est en moyenne de 171 mm/an. L'automne austral est la saison avec les chutes de neige les plus importantes, et le printemps austral, la saison avec les chutes de neige les plus faibles. Par ailleurs, une expérience de mesure in-situ des précipitations a été développée sur la base de Dumont d'Urville en Antarctique, des observations in-situ étant nécessaires à la validation des algorithmes de télédétection. Un système de profilage utilisant des capteurs optiques a été installé sur un mât de 73 m afin d'identifier les chutes de neige et les évènements de transport de neige par le vent. Les flux de neige mesurés à différentes hauteurs devraient être similaires lors de chute de neige sans transport de neige, alors qu'un gradient devrait apparaître si de la neige est transportée depuis la surface. Le système a été évalué et comparé aux analyses opérationnelles d'ECMWF. Enfin, les simulations des modèles de climat utilisés pour la production du rapport du GIEC ont été comparées aux observations satellites obtenues. Tous les modèles simulent un taux de chute de neige supérieur à celui observé avec CloudSat. Le changement de précipitation en Antarctique durant le XXIème siècle simulé varie de -6.0 % à +39.4 % en fonction des modèles et des scénarios d'émission de gaz à effet de serre. Les modèles de climat simulant des taux de chute de neige proches de ceux observés par satellite pour la période actuelle prédisent en moyenne un changement plus important de précipitation au cours du XXIème siècle, et donc un impact sur le niveau des mers plus conséquent. / During the 21st century, precipitation is expected to increase in polar regions. In Antarctica, this would lead to an increase in snow accumulation over the continent, which would represent a positive contribution to the ice sheet mass balance, and thus a negative contribution to sea level. Almost all the climate models predict a precipitation increase in Antarctica during the 21st century, but this change differs widely according to the models. Moreover, the current precipitation rate simulated by these models diverge greatly. However, because no reliable observation of Antarctic precipitation was available so far, it was not possible to benchmark climate models. In this study, data from the cloud profiling radar onboard CloudSat satellite have been used to produce the first climatology of Antarctic precipitation from observations. This climatology agrees well with ERA Interim reanalysis, the production of which is constrained by various in situ and satellite observations, but does not use any data from CloudSat. The mean snowfall rate from CloudSat observations is 171 mm/an over the Antarctic ice sheet, north of 82°S. The maximum snowfall rate is observed during the fall, while the minimum snowfall rate occurs in spring. Because in-situ measurements are necessary to evaluate remote sensing observations, a field experiment has been developed at Dumont d'Urville station in Antarctica for measuring precipitation. Optical sensors have been set up at different levels on a 73-meter tower in order to separate snowfall from blowing snow events. Snow flux measured at different heights should be similar during snowfall without blowing snow, whereas a gradient shoud be observed if blowing snow occurs. The system has been evaluated and compared to the ECMWF operational analysis. Finally, simulations from the climate models used for the last IPCC report have been compared to the new satellite climatology. All the models produce a higher snowfall rate than the snowfall observed with CloudSat. Precipitation increase predicted in Antarctica varies from -6.0 % to +39.4 % according to the models and the greenhouse gas emissions scenarios.Climate models which reproduce a current snowfall rate close to the snowfall rate observed by satellite predict on average a larger increase in Antarctic precipitation during the 21st century, and thus a stronger impact on sea level.