
by Leoncio Cabrera (Universidad de Chile and Pontificia Universidad Católica de Chile)
Jun 28, 2024
Forecasting Volcanic Eruptions Using Machine Learning
Anticipating volcanic eruptions remains a significant challenge in Earth science, despite extensive research efforts (for a recent review, see Acocella et al., 2023). Traditional monitoring systems, which assess volcano-alert levels based on seismic activity, surface deformation, and other factors, often do not include eruption forecasts. This is critical given the severe human, economic, and environmental impacts of eruptions, as demonstrated by the Chaitén (Chile) and Whakaari (New Zealand) volcano disasters. The Chaitén eruption necessitated the complete evacuation and subsequent relocation of the town, while the Whakaari eruption resulted in the tragic deaths of 22 out of 47 tourists and guides on the volcano. Recent advancements in machine learning techniques offer a new approach to identifying pre-eruptive patterns (for an example in New Zealand, see Dempsey et al., 2020), potentially improving the reliability of forecasts and aiding in the timely implementation of disaster mitigation strategies.
In this work (Cabrera et al., 2024), we developed a short-term eruption forecasting model for the Copahue Volcano using seismic data and machine learning techniques. Copahue, located on the border between Chile and Argentina in the Southern Andes, is notable for having experienced six eruptive pulses between 2020 and 2022 and for its proximity to urban centers. Our pipeline, summarized in Fig. 1A, includes steps for precursor extraction, classification modeling, and decision-making to issue eruption alerts. By testing our model on the six recent eruptions of Copahue Volcano, we demonstrated its ability to forecast eruptions 5 to 75 hours in advance (see an example in Fig. 1B). The model showcases a high true negative rate, indicating robust performance in distinguishing between eruption and non-eruption periods.

We also analyzed long-term geodetic data (GNSS and InSAR) of the volcano to study its physical processes over several years. Our findings indicate that the volcano has experienced periods of inflation and deflation in recent years. Specifically, there was an uplift episode of approximately 3–4 cm from September 2017 to March 2020, followed by a subsidence episode of about 2 cm from March 2020 until late 2021. No significant displacements were associated with the short-duration eruptive pulses during this period.
The combined short-term sensitivity of seismic data and long-term sensitivity of geodetic data, coupled with the advanced capabilities of various machine learning technics, presents a significant opportunity to enhance existing monitoring systems. This integration can substantially reduce risk in complex environments, ultimately contributing to safer and more secure communities.
Citation
Cabrera, L., A. Ardid, I. Melchor, S. Ruiz, B. Symmes- Lopetegui, J. Carlos Báez, F. Delgado, P. Martinez-Yáñez, D. Dempsey, and S. Cronin (2024). Eruption Forecasting Model for Copahue Volcano (Southern Andes) Using Seismic Data and Machine Learning: A Joint Interpretation with Geodetic Data (GNSS and InSAR), Seismol. Res. Lett. XX, 1–16, doi: 10.1785/ 0220240022.
More information
Leoncio Cabrera is a researcher in the Department of Geophysics of the Universidad de Chile, and will start soon as Assistant Professor at the Pontificia Universidad Católica de Chile. For more information, please contact him at lecabrera@uchile.cl or visit his personal webpage.
Other references
Acocella, V., Ripepe, M., Rivalta, E., Peltier, A., Galetto, F., & Joseph, E. (2024). Towards scientific forecasting of magmatic eruptions. Nature Reviews Earth & Environment, 5(1), 5-22.
Dempsey, D. E., Cronin, S. J., Mei, S., & Kempa-Liehr, A. W. (2020). Automatic precursor recognition and real-time forecasting of sudden explosive volcanic eruptions at Whakaari, New Zealand. Nature communications, 11(1), 3562.