← Back to Sessions

Lahar Early Warning at Volcano Santiaguito: A Classical and a Deep Learning Approach

Description: 

The Santiaguito Volcanic Complex is located in Guatemala’s Western Volcanic Highlands, along the westernmost section of the Central American Volcanic Arc. Santiaguito presents multiple hazards (including explosions, local earthquakes, lahars, mudflows, and pyroclastic flows) to the local population, which numbers 1.6 million people within 30 km² around the volcano. Lahars are flows of a mixture of a large amount of water and pyroclastic debris (that can include gases) which can rapidly initiate and reach speeds of tens of meters per second down wide barrancas (canyons) making them highly destructive. Lahar occurrence strongly correlates with rainfall at the volcano and they are commonplace in the long rainy season. They pose a great hazard to local inhabitants who regularly cross the channels as they live and work on farms on or near the flanks of the barrancas. INSIVUMEH, the national seismic and volcano monitoring agency, has recently built a network of 10 seismic stations that can monitor these flows in collaboration with external agencies. We used the open seismic data from 2022 and 2023 to consolidate a Lahar catalogue, having a total of 43 Lahars. We then use the Lahar data (91 waveforms) from 2022 to train a Siamese Neural Network that uses 5-minute single-station waveforms to produce a Deep Learning Lahar detection. We use the same data to develop a classical Lahar detector that uses ratios between two different short-term/long-term average amplitude (STA/LTA) characteristic functions implemented in SeisComP. We then test both approaches on continuous data from 2023 and compare them. We demonstrate the advantages and disadvantages of both approaches and show that both could be combined for optimal Lahar Early Warning. We show how the classical method, implemented through SeisComP, is being used to operate an early warning messaging system. We expect that the developed tools can be portable to other volcanic areas and even other flow types. Furthermore, we demonstrate the usefulness of Siamese Neural Networks for the development of Deep Learning models when a small number of training data is available.

Session: Detecting, Characterizing and Monitoring Mass Movements - I

Type: Oral

Date: 5/2/2024

Presentation Time: 09:00 AM (local time)

Presenting Author: Dario

Student Presenter: No

Invited Presentation: 

Authors