Thesis: The modeling of the combinations of flood hazard phenomena is a current issue for the scientific community which is primarily interested in urban and nuclear sites. Indeed, it is very likely that the deterministic approach exploring several scenarios has certain limits because these deterministic scenarios ensure an often excessive conservatism. Probabilistic approaches provide additional precision by relying on statistics and probabilities to complement deterministic approaches. These probabilistic approaches aim to identify and combine many possible hazard scenarios to cover many possible sources of risk. The Probabilistic Flood Hazard Assessment (PFHA) proposed in this thesis allows to characterize a quantity(ies) of interest (water level, volume, duration of immersion, ect.) at different points of interest of a site based on the distributions of the different phenomena of the flood hazard as well as the characteristics of the site. The main steps of the PFHA are: i) screening of the possible phenomena (rainfall, sea level, waves, ect.), ii) identification and probabilization of the parameters representative of the selected flood phenomena, iii) propagation of these phenomena from their sources to the point of interest on the site, iv) construction of hazard curves by aggregating the contributions of the flood phenomena. Uncertainties are an important topic of the thesis insofar as they will be taken into account in all the steps of the probabilistic approach. The work of this thesis is based on the study of the conjunction of rain and sea level and provide a new method for taking into account the temporal phase shift between the phenomena (coincidence). An aggregation model has been developed to combine the contributions of different flood phenomena. The question of uncertainties has been studied and a method based on the theory of belief functions has been used because it has various advantages (faithful modeling in cases of total ignorance and lack of information, possibility to combine information of different origins and natures, ect.). The proposed methodology is applied on the site of Le Havre in France.