Floods are the predominant natural risk in the world and the damage they cause is the biggest among natural disasters. One of the main drivers of floods is extreme rainfall. Due to climate change, the occurrence and intensity of climate change is likely to increase. As a result, the risk of flooding is likely to increase. The impact of changes in extreme rainfall is now an important issue for public safety and for the sustainability of infrastructure. Flood risk management strategies in the future climate are mainly based on simulations from digital climate models. In particular, a digital climate model provides a time series of precipitation for each of the grid points in its simulation space domain. Simulated time series may be daily or intraday time series and cover the entire simulation period, typically between 1961 and 2100. The spatial continuity of simulated physical processes leads to spatial consistency among time series. In other words, time series from neighbouring grid points often share similar characteristics. In general, the theory of extreme values is applied to these simulated time series to estimate the corresponding quantum at a certain level of risk. In most cases, the variance in estimation is considerable due to the limited number of extreme rainfall available and this can play a key role in the development of risk management strategies. Therefore, a statistical model for accurately estimating the quantum of extreme rainfall simulated by a digital climate model has been developed in this thesis. The model developed is specifically adapted to the data generated by a climate model. In particular, it uses the information contained in the continuous daily series to improve the estimation of non-stationary quantum, without making a binding assumption on the nature of non-stationarity. The model also exploits the information contained in the spatial coherence of extreme precipitation. This is modelled by a Bayesian hierarchical model, where a priori parameter laws are spatial processes, in this case Markov gaussiens fields. The application of the model developed to a simulation generated by the Canadian Regional Climate Model has significantly reduced the variance in the estimation of quantum in North America.