Probabilistic Coastal Tsunami Evacuation Modelling Using Agent-based Modelling in Catania, Italy

The eastern coast of Sicily, including Catania’s harbor and the tourist beach, is highly vulnerable to tsunami hazards, with a history of major events such as the January 11^th, 1693 earthquake. Due to its geographic location and the region’s seismic activity, Catania remains at significant risk of similar catastrophic events. Evacuation is widely recognized as the most effective means of saving lives in an imminent tsunami event.

The Catania coastal area is densely populated and there is a significant proportion of elderly people among the residents who may face greater difficulties during evacuation. Moreover, there is a significant seasonal variation in the population since small coastal towns host many tourists during spring and summer.

In this study, we propose a probabilistic simulation-based framework for evacuation modelling.  In the framework, we use Agent-Based Modeling (ABM) to develop a high-resolution digital model of the evacuation environment, including the location of people, residences, roads, and the shelters that are defined in the advisory/watch tsunami evacuation maps designed for Italian coasts (Tonini et al. 2021). We have modeled human behaviors using data collected from questionnaires and other open-source statistical databases. The ABM model simulates human behavior in response to 92 detailed tsunami inundation scenarios derived from Probabilistic Tsunami Hazard Analysis (PTHA) results (Gibbons et al., 2020).

The probability of safe evacuation is assessed for various scenarios such as daytime or nighttime exposure and the presence or absence of a tsunami early warning. This assessment is evaluated using a Monte Carlo simulation workflow, incorporating all uncertain modeling parameters. These parameters range from tsunami source characteristics (e.g., magnitude and slip) to various human response factors influenced by different behavioral patterns, such as immediate escape, freezing, or seeking information as well as choices like deciding between driving a car or walking. The model incorporates different types of agents to capture the complexity of human behavior. These agents include residents, both individuals and families across various age groups, and tourists, each characterized by distinct response patterns and decision-making processes. The probabilistic evacuation modeling results are derived by sampling the agents’ response parameters, such as individual velocity and response delays, to account for variability while maintaining computational feasibility. Preliminary results from selected scenarios with simple human behavior show that tsunami scenario parameters such as magnitude and tsunami impact (e.g., flow depth), can significantly influence the probability of safe evacuation.