← Back to Sessions

The What, When and Whys of Alert Progression During Tsunamigenic Events: A Simple Generative Approach to Forecasting Decision Points and Developing Heuristics

Description: 

Due to the possibility of imminent tsunami inundation after large earthquakes, tsunami warning centers make fast, real-time decisions about which regions to alert. For initial alerts, the centers typically assume a worst-case scenario as there often aren’t sufficient preliminary constraints to resolve event tsunamigenesis. This strategy mitigates the risk posed by the tsunami and simplifies the decision making process. However, it also communicates serious impacts that may not come to pass, and can thereby be a source of frustration for alerted populations. These populations, some of whom may have evacuated, may have to wait hours before the worst-case assumptions can be refuted and the event resolved. Here, in order to alleviate the uncertainty inherent in event progression and to better explain/simplify event decision-making, we propose a method focused on identifying when, where, and why plausible hazard likelihoods are expected to update during an individual event. We develop a simple generative model that maps event observations to a probabilistic description of future impacts at the coast, and show how the model can be used to both highlight upcoming decision points and develop straightforward event-specific heuristics (rules of thumb) that simplify the decision making process. For example, the model can recognize that a specific open ocean water-level observation exceeding some threshold will be the first determining factor in whether or not a warning level observation (>1 m waves) will be realized at a coastal location. The model is trained using an array of precomputed Alaska Tsunami Forecast Model (ATFM) outputs, including forecasted tsunami amplitudes for large subduction zone earthquakes in the Pacific. This model does not improve final forecast accuracy, but rather provides a real-time, decision-first framework to highlight what amplitude thresholds need to be measured when in order to reduce initial forecast uncertainty and crystallize anticipated impacts. The goal of this model is to provide a basic, explainable, robust, data-oriented framework that informs the decision making process.

Session: Six Decades of Tsunami Science: From the Source of the 1964 Tsunami to Modern Community Preparedness [Poster Session]

Type: Poster

Date: 5/2/2024

Presentation Time: 08:00 AM (local time)

Presenting Author: Ben

Student Presenter: No

Invited Presentation: 

Authors