Every year, floods have huge consequences both in terms of human casualties and damages to infrastructures. This thesis studies means to improve the response phase of the disaster management to flash floods. They are characterized by a fast rise of the water level which leaves limited time to rescue teams for anticipation measures. In collaboration with rescue teams from SDIS 31 (Service Départemental d'Intervention et de Secours), victim relief operations are studied. The problem is one of optimizing routes for vehicles between locations where victims need to be rescued. This problem is called a Vehicle Routing Problem. A major challenge is that rescue vehicles have limited capacities. Furthermore, interventions can be prioritized. In function of the victim to rescue, the intervention can be of various degree of urgency. With this priority is also associated a deadline. In addition, rescue teams from SDIS 31 use different categories of vehicles. These vehicles have different capacities and can handle different types of interventions. These constraints add complexity to the general Vehicle Routing Problem. However, rescue teams often do not possess enough vehicles to rescue all victims in one tour. A tour is counted every time a vehicle has to drive back victims to safety. Splitting service is allowed in the problem studied. It means routes can be planned to rescue victims at a single location with several interventions, which gives more flexibility than the classical Vehicle Routing Problem. First, the work of this thesis is to develop solutions to answer the Vehicle Routing Problem encountered by rescue teams and put it into perspective with related. An exact approach is tested and its performance are measured with the evolution of the problem size (number of intervention locations). Heuristic algorithms are also developed in order to cope with large problem sizes. Finally the resource distribution problem faced by rescue teams is studied. When a flood impacts several sectors separated with too much distance, rescue teams may need to dispatch vehicles in sectors. The optimization problem of such a resource distribution is presented. Different solutions are introduced in this thesis and compared. During crisis such as floods, all information is not known at the beginning of the rescue operations. Dynamic events of various nature are released during the crisis, while rescue vehicles are already on the road. Locations where victims need to be rescued, as well as the number of victims to rescue, may be revealed on-the-fly thanks to rescue teams scouting of the impacted area for instance. In addition, rescue teams operate in a degraded environment, which can imply detours on the planned route and delays on timing. Therefore, another class of problem is studied when dynamic information is considered, called Dynamic Vehicle Routing Problem. Solutions adapted to answer the Dynamic version of the Vehicle Routing Problem are developed and evaluated in this thesis. A graph generator is presented in the thesis to generate instances of a flooded territory and its issues. It enables to parameterize characteristics of the created instances. Hence, after data analysis of Experience Feedbacks from Luchon flash flood in the South of France in 2013, the case study of this thesis, instances with similar characteristics have been generated to conduct evaluation on. A simulator to run interactions between the different actors of the crisis management has also been developed. It allows to run dynamic crisis scenario in simulated time. It helps to speed the evaluation process conducted on a large number of scenarios. These scenarios have been generated using data from the study case as for instances. The evaluation process allows to observe which solutions developed in this thesis gives the best performances in terms of solutions quality and computation time.