Using Template Matching to Detect Hidden Fluid Release Episodes Beneath Crater Lakes in Ruapehu, Copahue, and Kawah Ijen Volcanoes

The problem

Some active volcanoes hold a lake in their crater, above the vent. Eruptions that break through such a lake can usually be detected by sight or by acoustic sensors (Edwards et al., 2017; Scott, 2013). However, smaller jets of hydrothermal fluid released into the lake may go unnoticed (Montanaro et al., 2022), although they may still pose a risk to people near the summit. Detecting these small jets would provide observations of how often a seal forms and fails below the lake.

Ruapehu, in New Zealand, has a crater lake about 400 m across that holds roughly 10 million cubic metres of water. Our goal was to find seismic signs of a seal forming below this lake, followed by a release of fluid into its base.

What we did

We used seismic data from 2006 to 2021, recorded at one station about 3 km from the crater. After removing earthquake signals from the record, we measured volcanic tremor in three frequency bands: 2 to 5 hertz (Hz, vibrations per second), 4.5 to 8 Hz and 8 to 16 Hz. The average tremor amplitude in the 2 to 5 Hz band is called the real-time seismic amplitude measurement (RSAM). The displacement seismic amplitude ratio (DSAR) is the ratio of ground displacement in the middle band to that in the high band. We took its median over rolling 48-hour windows.

At Whakaari (White Island), this median rose over several weeks before the 2019 eruption. Ardid et al. (2022) suggested that the rise reflected a low-permeability seal forming over the hydrothermal system. In that interpretation, the seal suppresses the shallow fluid movement that generates high-frequency tremor. We used the four weeks of DSAR median before that eruption as a template. We flagged every four-week period of the Ruapehu record whose correlation coefficient with the template exceeded 0.3, the top 5% of all correlations. We kept a flagged period only if (1) RSAM peaked more than three standard deviations above its mean near the match, (2) the middle band stayed above the high band in the days before the peak, to exclude noise from ski fields, rain and wind, and (3) RSAM decayed exponentially in the following 24 hours.

The third check is based on the RSAM decay after the 2006 and 2007 eruptions. We compared it with a simple model of a pressurised reservoir draining into the lake through a permeable pathway. As fluid leaves, reservoir pressure falls and the discharge rate declines exponentially. Girona et al. (2019) showed, in a physical model, that tremor amplitude is proportional to degassing rate. We assumed that RSAM after a release is proportional to the discharge rate.

What we found

At Ruapehu, six events passed every check. Two are the 2006 and 2007 eruptions. A third, in 2009, was described as phreatic activity by the GeoNet monitoring network (Scott, 2013). The other three, in 2010, 2016 and 2021, were to our knowledge unreported and have no corroborating disturbance at the surface. However, each coincided with a regional earthquake and with rain storms. Rain-driven changes in lake level make these events difficult to confirm from lake-level data.

We classified the 2009 event as a fluid release because it was likely too small to count as an eruption. In the month before, lake level and temperature rose and then stabilised while the DSAR median increased. This is consistent with a seal restricting the entry of hot fluid into the lake. The event appeared as a spike in RSAM and a sudden rise in lake level of about 0.2 m, roughly 20,000 cubic metres. By our arithmetic, that is about 0.2% of the lake's volume.

After the 2006 and 2007 eruptions, RSAM decayed with discharge times of about 10 to 40 hours. The shortest among all the Ruapehu events was 12 hours. In the model, a 12-hour discharge time leaves tremor at about 14% of its peak after 24 hours, as our illustration below shows.

At Kawah Ijen in Indonesia we found three potential events in about 2.7 years of data. At Copahue, on the border of Chile and Argentina, four small ash-producing eruptions were reported during a 9-month record, and we identified three of them. Both records were noisier than Ruapehu's.

These observations are consistent with a conceptual model of pressure cycling below an active crater lake. In this model, a seal forms at the top of the hydrothermal system and pressure builds in the fluid trapped beneath it. When the seal is disrupted, fluid is released into the lake at a rate that declines over hours to days. We suggest that many of these releases do not produce a phreatic eruption.

Limitations

Our method is unlikely to find all events, because it assumes they resemble the Whakaari template and develop over about a week. The link between tremor and discharge rate is also an assumption. Frequent gas-flux measurements in the crater could test it.

Two of our checks use the RSAM peak and its decay. Both are recorded only during and after an event. Template matching alone is therefore not sufficient for accurate forecasting. We have set up a system that applies the template matching to Ruapehu in real time, every 10 minutes. It could help track possible sealing events, although a sealing event can dissipate through passive outgassing without an eruption. We will evaluate its performance over the next few years.

Glossary

A crater lake is a body of water, often warm, acidic and saline, that forms in the crater of an active volcano.

A hydrothermal system is hot water and gas circulating through rock, here between the magma column below and the crater lake above.

A phreatic eruption is an explosive release of pressurised hydrothermal fluid, such as the 2006 and 2007 Ruapehu eruptions.

A hydrothermal seal is a low-permeability layer that restricts the flow of gas and water toward the surface.

Permeability measures how easily fluid flows through rock.

Volcanic tremor is continuous low-level shaking of the ground at an active volcano.

RSAM (real-time seismic amplitude measurement) is the average tremor amplitude in the 2 to 5 Hz band over 10-minute windows.

The displacement seismic amplitude ratio (DSAR) is the ratio of ground displacement in the 4.5 to 8 Hz band to that in the 8 to 16 Hz band.

A template is a reference segment of a record, searched for elsewhere by matching its shape.

A standard deviation measures the typical spread of values around their mean.

A correlation coefficient measures how similar two curves are in shape, equal to 1 for the same shape and near 0 for unrelated curves.

Exponential decay is a decline by the same fraction in each equal interval of time.

The discharge time is the time over which the discharge rate, and in our model RSAM, falls to about 37% of its starting value.