京都大学 / 0048 / 新制・課程博士 / 博士(理学) / 甲第21578号 / 理博第4485号 / 新制||理||1644(附属図書館) / 京都大学大学院理学研究科地球惑星科学専攻 / (主査)教授 余田 成男, 教授 秋友 和典, 准教授 石岡 圭一 / 学位規則第4条第1項該当 / Doctor of Science / Kyoto University / DGAM

La neige est une composante essentielle du système climatique arctique. Au nord de l'Eurasie et de l'Amérique du Nord, la couverture neigeuse est présente de 7 à 10 mois par an et son extension saisonnière maximale représente plus de 40% de la surface terrestre de l'hémisphère nord. La neige affecte une variété de processus climatiques et de rétroactions aux hautes latitudes. Sa forte réflectivité et sa faible conductivité thermique ont un effet de refroidissement et modulent la rétroaction neige-albédo. Sa contribution au bilan radiatif de la Terre est comparable à celle de la banquise. De plus, en empêchant d'importantes pertes d'énergie du sol sous-jacent, la neige limite la progression de la glace et le développement du pergélisol saisonnier. Réserve d'eau naturelle, la neige joue un rôle essentiel dans le cycle hydrologique aux hautes latitudes, notamment en ce qui concerne l'évaporation et le ruissellement. La neige est l'une des composantes du système climatique présentant la plus forte variabilité. Le réchauffement de l'Arctique étant deux fois plus rapide que celui du reste du globe, la variabilité présente et future des caractéristiques de la neige est cruciale pour une meilleure compréhension des processus et des changements climatiques.Cependant, notre capacité à observer l'Arctique terrestre étant limitée, les modèles climatiques jouent un rôle clé dans notre aptitude à comprendre les processus liés à la neige. À cet égard, la représentation des rétroactions associées à la neige dans les modèles climatiques, en particulier pendant les saisons intermédiaires (lorsque la couverture neigeuse de l'Arctique présente la plus forte variabilité), est primordiale.Notre étude porte principalement sur la représentation de la neige terrestre arctique dans les modèles de circulation générale issus du projet CMIP5 (Coupled Model Intercomparison Project) au cours du printemps (mars-avril) et de l’automne (octobre-novembre) de 1979 à 2005. Les caractéristiques de la neige des modèles de circulation générale ont été validées par rapport aux mesures de neige in situ, ainsi qu’à des produits satellitaires et à des réanalyses.Nous avons constaté que les caractéristiques de la neige dans les modèles ont un biais plus marqué au printemps qu'en automne. Le cycle annuel de la couverture neigeuse est bien reproduit par les modèles. Cependant, les cycles annuels d'équivalent en eau de la neige et de sa profondeur sont largement surestimés par les modèles, notamment en Amérique du Nord. Il y a un meilleur accord entre les modèles et les observations dans la position de la marge de neige au printemps plutôt qu'en automne. Les amplitudes de variabilité interannuelle pour toutes les variables de la neige sont nettement sous-estimées par la plupart des modèles CMIP5. Pour les deux saisons, les tendances des variables de la neige dans les modèles sont principalement négatives, mais plus faibles et moins significatives que celles observées. Les distributions spatiales des tendances de la couverture neigeuse sont relativement bien reproduites par les modèles, toutefois, la distribution spatiale des tendances en équivalent-eau et en profondeur de la neige présente de fortes hétérogénéités régionales.Enfin, nous concluons que les modèles CMIP5 fournissent des informations précieuses sur les caractéristiques de la neige en Arctique terrestre, mais qu’ils présentent encore des limites. Il y a un manque d’accord entre l’ensemble des modèles sur la distribution spatiale de la neige par rapport aux observations et aux réanalyses. Ces écarts sont particulièrement marqués dans les régions où la variabilité de la neige est la plus forte. Notre objectif dans cette étude était d'identifier les circonstances dans lesquelles ces modèles reproduisent ou non les caractéristiques observées de la neige en Arctique. Nous attirons l’attention de la communauté scientifique sur la nécessité de prendre compte nos résultats pour les futures études climatiques. / Snow is a critical component of the Arctic climate system. Over Northern Eurasia and North America the duration of snow cover is 7 to 10 months per year and a maximum snow extension is over 40% of the Northern Hemisphere land each year. Snow affects a variety of high latitude climate processes and feedbacks. High reflectivity of snow and low thermal conductivity have a cooling effect and modulates the snow-albedo feedback. A contribution from terrestrial snow to the Earth’s radiation budget at the top of the atmosphere is close to that from the sea ice. Snow also prevents large energy losses from the underlying soil and notably the ice growth and the development of seasonal permafrost. Being a natural water storage, snow plays a critical role in high latitude hydrological cycle, including evaporation and run-off. Snow is also one of the most variable components of climate system. With the Arctic warming twice as fast as the globe, the present and future variability of snow characteristics are crucially important for better understanding of the processes and changes undergoing with climate. However, our capacity to observe the terrestrial Arctic is limited compared to the mid-latitudes and climate models play very important role in our ability to understand the snow-related processes especially in the context of a warming cryosphere. In this respect representation of snow-associated feedbacks in climate models, especially during the shoulder seasons (when Arctic snow cover exhibits the strongest variability) is of a special interest.The focus of this study is on the representation of the Arctic terrestrial snow in global circulation models from Coupled Model Intercomparison Project (CMIP5) ensemble during the melting (March-April) and the onset (October-November) season for the period from 1979 to 2005. Snow characteristics from the general circulation models have been validated against in situ snow measurements, different satellite-based products and reanalyses.We found that snow characteristics in models have stronger bias in spring than in autumn. The annual cycle of snow cover is well captured by models in comparison with observations, however, the annual cycles of snow water equivalent and snow depth are largely overestimated by models, especially in North America. There is better agreement between models and observations in the snow margin position in spring rather than in autumn. Magnitudes of interannual variability for all snow characteristics are significantly underestimated in most CMIP5 models compared to observations. For both seasons, trends of snow characteristics in models are primarily negative but weaker and less significant than those from observations. The patterns of snow cover trends are relatively well reproduced in models, however, the spatial distribution of trends for snow water equivalent and snow depth display strong regional heterogeneities.Finally, we have concluded CMIP5 general circulation models provides valuable information about the snow characteristics in the terrestrial Arctic, however, they have still limitations. There is a lack of agreement among the ensemble of models in the spatial distribution of snow compared to the observations and reanalysis. And these discrepancies are accentuated in regions where variability of snow is higher in areas with complex terrain such as Canada and Alaska and during the melting and the onset season. Our goal in this study was to identify where and when these models are or are not reproducing the real snow characteristics in the Arctic, thus we hope that our results should be considered when using these snow-related variables from CMIP5 historical output in future climate studies.

The work presented in the thesis focuses on systematically documenting the multi scale nature of the temporal persistence and spatial coherence of tropical rainfall. There are three parts to the thesis: The first two parts utilize satellite-retrieved rainfall at multiple observational resolutions to characterize the space-time organization of rain; the third part assesses the ability of state-of-the-art coupled models to reproduce some of the observed features. In the first part of the study, which focuses on the temporal persistence of rain, we analyze the Tropical Rainfall Measurement Mission (TRMM) satellite-based observations to compare and contrast wet and dry spell characteristics over the tropics (30 S-30 N). Defining a wet (dry) spell as the number of consecutive rainy (nonrainy) days, we find that the distributions of wet spells (independent of spatial resolution) exhibit universality in the following sense. While both ocean and land regions with high seasonal rainfall accumulation (humid regions) show a predominance of 2-4 day wet spells, those regions with low seasonal rainfall accumulation (arid regions) exhibit a wet spell duration distribution that is essentially exponential in nature, with a peak at 1 day. The behaviour that we observed for wet spells is reversed for dry spell distributions. The total rainfall accumulated in each wet spell has also been analyzed, and we find that the major contribution to seasonal rainfall for arid regions comes from very short length wet spells; however, for humid regions, this contribution comes from wet spells of duration as long as 30 days. An exhaustive sensitivity study of factors that can potentially affect the wet and dry spell characteristics (e.g., resolution) shows that our findings are robust. We also explore the role of chance in determining the 2-4 day mode, as well as the inuence of organized convection in separating reality from chance. The second part deals with the spatial coherence of tropical rain. We take two different approaches, namely, a global and local view. The global view attempts to quantify the con-ventional view of rain, i.e., the dominance of the intertropical convergence zone (ITCZ), while the local view tries to answer the question: if it rains, how far is the influence felt in zonal and meridional directions? In both approaches, the classical e-folding length for spatial decorrelation is used as a measure of spatial coherence. The major finding in the global view approach is that, at short timescales of accumulation (daily to pentad to even monthly), rain over the Equator shows the most dominant zonal scale. It is only at larger timescales of accumulation (seasonal or annual) that the dominance of ITCZ around 7 N is evident. In addition, we also find a semi-log linearity between the spatial scales, seen from afar, and timescale of accumulation, with a break in linearity around typical synoptic timescales of 5-10 days. The local view quantifies the dominance of the zonal scale in the tropical ocean convergence zones, with an anisotropy value (ratio of zonal to meridional scales) of 3-4. Over land, on the other hand, the zonal and meridional scales are comparable in magnitude, suggesting that rain tends to be mostly isotropic over continental regions. This latter finding holds true, irrespective of the spatial and temporal resolutions at which rain is observed. Interestingly, the anisotropy over ocean, while invariant with spatial resolution, is found to be a function of temporal resolution: from a value of 3-4 at daily timescale, it decreases to around 1.5 at 3-hourly resolution, suggesting that perhaps rain fundamentally might be isotropic in nature at an event scale. The final part analyses a few models from the suite of Coupled Model Intercomparison Project (CMIP5) models, to evaluate their ability to reproduce some of these aforementioned features. For all the strong biases that models are known to have, some of the observed features are captured well by the models. Specifically, on the temporal persistence front, the observed 2-4 day mode of wet (dry) spells of rain over humid (arid) regions is also seen in models. The overestimation of longer duration wet spells appears to be the primary cause of a positive bias in the number of rainy days from the models. In general, the tendency of models to not stop raining results in lower and higher number of shorter and longer duration wet spells, respectively, and consequently an overall reduction in dry spells of all durations. On the spatial coherence front, the main finding from the global view approach is that the observed semi-log linearity of the zonal spatial scale of rainfall as a function of timescale of accumulation is strikingly well-reproduced by the models. Even more remarkable is that the models are able to mimic the break in this linearity around 5 days (typical synoptic scale). What the models fail to do prominently is the transition of the dominance of equatorial rain at smaller timescales of accumulation to the dominance of ITCZ at around 7 N at higher timescales of accumulation. Based on the local view approach, we find that, in general, even though the zonal and meridional scales are overestimated, the observed isotropy of continental rain is captured very well by the models. Over the oceans, the success is less prominent, especially with the core of the ITCZ showing much larger ratios than those observed. Thus, the models seem to be able to reproduce the anisotropy for the wrong reasons, and the proposed anisotropy ratio could be a useful metric in further diagnosis of climate models.