Tsunami pedestrian evacuation simulation for Camaná, Peru: Perspectives for improving evacuation performance

Abstract

Optimizing pedestrian evacuation in the face of a tsunami remains a critical challenge for safeguarding human lives. Agent-based models combined with reinforcement learning techniques offer a powerful framework to simulate complex evacuation scenarios, where agents learn to make decisions and identify safe routes in real time. This study focuses on improving evacuation efficiency along the coast of Camaná, Arequipa, Peru. We propose the use of numerical simulations to model pedestrian movement under the guidance of a reinforcement learning-based system. Under current transportation network conditions, only 16.6% of the population is able to reach a safe area in a tsunami scenario similar to the 2001 event. To address this, several modifications to the transportation network were proposed, including the addition of new evacuation paths and the construction of vertical evacuation structures. With the incorporation of 12 new paths and 6 vertical evacuation structures, the percentage of the population reaching safety increases to 73%. These findings provide a scientific basis for planning and implementing improvements to evacuation infrastructure in tsunami-prone areas.