Analyse probabiliste régionale des précipitations : prise en compte de la variabilité et du changement climatique

Extreme precipitations and their consequences (floods) are one of the most threatening natural disasters for human beings. In engineering design, Frequency Analysis (FA) techniques are an integral part of risk assessment and mitigation. FA uses statistical models to estimate the probability of extreme hydrological events which provides information for designing hydraulic structures. However, standard FA methods commonly rely on the assumption that the distribution of observations is identically distributed. However, there is now a substantial body of evidence that large-scale modes of climate variability (e.g. El-Niño Southern Oscillation, ENSO; Indian Ocean Dipole, IOD; etc.) exert a significant influence on precipitation in various regions worldwide. Furthermore, climate change is likely to have an influence on hydrology, thus further challenging the “identically distributed” assumption. Therefore, FA techniques need to move beyond this assumption. In order to provide a more accurate risk assessment, it is important to understand and predict the impact of climate variability/change on the severity and frequency of hydrological events (especially extremes). This thesis provides an important step towards this goal, by developing a rigorous general climate-informed spatio-temporal regional frequency analysis (RFA) framework for incorporating the effects of climate variability on hydrological events. This framework brings together several components (in particular spatio-temporal regression models, copula-based modeling of spatial dependence, Bayesian inference, model comparison tools) to derive a general and flexible modeling platform. In this framework, data are assumed to follow a distribution, whose parameters are linked to temporal or/and spatial covariates using regression models. Parameters are estimated with a Monte Carlo Markov Chain method under the Bayesian framework. Spatial dependency of data is considered with copulas. Model comparison tools are integrated. The development of this general modeling framework is complemented with various Monte-Carlo experiments aimed at assessing its reliability, along with real data case studies. Two case studies are performed to confirm the generality, flexibility and usefulness of the framework for understanding and predicting the impact of climate variability on hydrological events. These case studies are carried out at two distinct spatial scales: * Regional scale: Summer rainfall in Southeast Queensland (Australia): this case study analyzes the impact of ENSO on the summer rainfall totals and summer rainfall maxima. A regional model allows highlighting the asymmetric impact of ENSO: while La Niña episodes induce a significant increase in both the summer rainfall totals and maxima, the impact of El Niño episodes is found to be not significant. * Global scale: a new global dataset of extreme precipitation including 11588 rainfall stations worldwide is used to describe the impact of ENSO on extreme precipitations in the world. This is achieved by applying the regional modeling framework to 5x5 degrees cells covering all continental areas. This analysis allows describing the pattern of ENSO impact at the global scale and quantifying its impact on extreme quantiles estimates. Moreover, the asymmetry of ENSO impact and its seasonal pattern are also evaluated.