In plain areas prone to tsunami, finding a way to shelter and escape from the inundation becomes a difficult task for residents. The 2011 Great East Japan Tsunami has shown that the horizontal evacuation using cars can compromise the safety of people. Another alternative is the vertical evacuation. In many cases, not only the capacity of these shelters plays an important role, but the spatial distribution and the evacuee preference for the nearest shelter. Such preference and location creates a conflict between capacity offer and demand. In this paper, we used an agent based model to simulate the evacuation of pedestrians and cars in La Punta, Peru. Twenty designated buildings for vertical evacuation are available for sheltering and escape from tsunami. The stochastic simulation of population spatial distribution and their refuge preferences revealed the over demand of some shelters. Finally, a capacity-demand map was created to share results with local authorities as a first step for future countermeasures in the district.