← Back to Sessions

WITHDRAWN Volcanic Eruption Forecasts Through Seismic Pattern Recognition: The 2023 Paroxysms of Shishaldin Volcano, Alaska

Description: 

WITHDRAWN Identifying patterns within geophysical data is critical for the detection of volcanic unrest and the forecasting of eruptions. Seismicity, encompassing volcano-tectonic earthquakes, long-period events, and tremor, among other geophysical signals, provides pivotal information about the state of unrest of a volcano. Tremor, specifically, tends to exhibit variations in its characteristics during eruption run-up, including changes in dominant frequency, the emergence or disappearance of overtones, and fluctuations in seismic amplitude. These variations raise the following question: Can we consistently quantify the probability of eruption based on the recognition of seismic tremor patterns? To address this question, we introduce a pattern recognition based-framework (building upon the work of Dempsey et al., 2020 [https://doi.org/10.1038/s41467-020-17375-2]; Ardid et al., 2023 [https://doi.org/10.21203/rs.3.rs-3483573/v1]; and Girona and Kyriaki, In Review) that merges seismic features from tremor data (e.g., dominant frequency, kurtosis, seismic amplitude, Shannon entropy) with several supervised machine learning models and Monte Carlo simulations to estimate the probability of an imminent eruption and its uncertainty. In particular, our framework integrates logistic regression, k-nearest neighbors, linear discriminant analysis, decision trees, random forest, support vector machine, and neural network models; and is designed to enable retraining, automatically updating the models whenever a new paroxysm takes place. The performance of our approach is assessed using the recent 13 paroxysms of Shishaldin Volcano, Alaska, occurring from July to November 2023, all of which were preceded by seismic tremor with escalating amplitude. We find that our framework captures successfully the seismic data trends that correlated with imminent paroxysms, reflected in a noticeable and consistent increase in the probability of eruption before the actual eruption occurs. This strategy exhibits promising potential for producing near real-time probabilistic eruption forecasts based on the recognition of patterns in seismic tremor data.

Session: Multidisciplinary Approaches for Volcanic Eruption Forecasting - II

Type: Oral

Date: 5/2/2024

Presentation Time: 05:15 PM (local time)

Presenting Author: Társilo

Student Presenter: No

Invited Presentation: 

Authors