Repeating quakes during multiple phases of unrest and eruption

Video Transcript
Download Video
Right-click and save to download

Detailed Description

The video describes work by Jay Wellik, a USGS seismologist and his colleagues with the Volcano Disaster Assistance Program (VDAP). They assisted Indonesia's CVGHM in responding to the 2017 activity/eruption at Mt. Agung, on the island of Bali. The work describes use of software to characterize the different types of earthquakes at the volcano, and how that revealed important infrormation about magma ascent and eruption.


Date Taken:

Length: 00:39:38

Location Taken: Agung Volcano, BA, ID


Jay:    All right. So this is mostly the story of the 2017 eruption at Agung in Bali, Indonesia, and then I'll also talk about my retrospective work with the repeating earthquakes during the unrest and through the eruption. I've been studying these repeating earthquakes at Agung and then San Miguel since 2017. This eruption--- VDAP was heavily involved in it over the course of about two and a half months during the initial unrest. I think there were about 15 people from VDAP that were in Bali at one point or another installing equipment or helping monitor activity. Of course, this wasn't VDAP's response. Obviously, CVGHM was leading the way and they were recognized for their efforts at the last COV meeting in Naples. I'm just here to tell you that story. Why are we telling you the story beyond the fact that VDAP was heavily involved in it? Well, from a volcanologist's perspective, this eruption was an example of a long dormant stratovolcano, long dormant basaltic-andesitic  stratovolcano that evolved slowly leading up to the eruption. So from a forecasting perspective, it was an example of how the long-term forecasts was pretty easy, but the short term forecast was really difficult because we'll see that there was an incredible amount of unrest in terms of seismicity for a couple of months, followed by about a month of declining seismicity, leading into the onset of eruptions. Socially, this was also an important eruption and a significant crisis because of the history of multiple VEI-5s at Agung. The last eruption at Agung before this unrest was in 1963, and it produced within a few days of detected activity---- within a few days of the first felt earthquakes at Agung (in 1963), lava had filled the summit crater, breached the rim, and caused pyroclastic flows that were fatal. So as unrest started at this volcano, that eruption was very much on people's minds. If you want to read more about this eruption, I'm going to give you the overview, and then a few things that we've learned after the fact. But if you want to read the full story, if you want to know everything we know about these eruptions, there are two papers you can refer to. The first is Syahbana et al. This is just an overview of the unrest, the monitoring, the crisis response, the evacuations that took place, and then the eruption itself. There's a link to that article here. There's also a paper written about InSAR modeling done after the fact that really helped to refine our model of what happened during this eruption. So there's a link to that article here as well. Let's see. We'll do a little bit of background on Agung, the 1963 eruption. What happened since then between 1963 and 2017,  the unrest itself, what we were keeping track of in real time, then these retrospective analyses that were done with InSAR, and then what I've done with the repeating earthquakes. So in 1963, again, it was a really fast eruption.  In mid-February, there were felt earthquakes. Three days later on the 19th, there were loud noises heard and explosions. Lava started to fill the crater. On the 20th, it had breached that summit crater, caused pyroclastic flows, lahars, and there are a lot of fatalities from these events. Then about a month later in mid-March, there was a VEI-5 paroxysm, and there were continued explosions for at least a year that continued to cause fatalities. There was no monitoring network prior to this eruption (in 1963). There was a network installed during the eruption as a part of the response, but as far as I know, all co-eruptive seismic records from this eruption had been lost. I've never seen them, and I never heard anything about them. The other thing about this eruption was a year later, Batur, which is small stratocone inside the large caldera, just right off the flanks of Agung. A year later in 1964, it erupted. Then that plus other petrologic work has led people to wonder about the deep magmatic connection between those two volcanoes. So keeping our eye on Batur was a part of this crisis response as well. Since 1963, this is still very densely populated area. Bali, in general, is a huge tourist destination, and some of those people will hike to the summit of Agung.  From a seismic perspective, since 1963, there was no local seismicity recorded on the Agung network. It was just a two station short-period analog network, but no local earthquakes recorded since 1963. There was, however, quite a bit of cultural noise on this small network that proved to be a headache throughout this crisis. This is an example of some of that cultural noise. This is all from sand mining on the flanks of the volcano. You can see, it starts at about 7:30 local time, and it goes to about 5:00 or 5:30 local time, and there's about an hour lunch break in between. So this made it really difficult to detect some signals and keep our eyes out for smaller subtle signals or things like tremor throughout the crisis. The first local earthquakes are recorded here at Agung were actually in June or July, and it was only a couple of events per day, and not every day. But considering the lack of seismicity for the previous decades, this caught people's attention. By mid-September, things had just increased up to tens of earthquakes per day, and based on that CVGHM, the monitoring agency in Indonesia, they raised the alert level to level 2 on the 14th. After the 14th, for the next several days, seismic rates continued to increase, and fumarolic activity at the summit of Agung increased as well. It was really the fumarolic activity at Agung that really helped us. It really helped us decide that Agung was our focus here and not Batur, because all these earthquakes that we're being recorded and felt, they were all happening right in between Agung and Batur. The regional earthquake monitoring agency, BMKG, they were locating some of the largest earthquakes, and those earthquake locations confirmed that they were happening right in between those two volcanoes. But there were no changes at Batur, so our main focus was always Agung throughout this crisis. So on the 18th, based on these changes, CVGHM raised the alert level to 3. At that time, seismicity continued to increase, fumarolic activity increased, there were now satellite-based thermal anomalies that showed activity in the crater, and in retrospect, we also know that there is a deformation happening. A five station GPS network had been installed with VDAP's help in 2012, but the telemetry stopped in 2014, so we really only had the most recent data that was still being saved on sight. Later on this crisis when there is an opportunity to do field work, we were able to retrieve those data, revive the network, and process those data. So we know retrospectively that during this time, there's some deep inflation source between 10, 15, 20 kilometers depth. On the 22nd, based on all these changes, based on the increasing felt earthquakes, the seismic rates, and increase in fumarolic activity, CVGHM raise their alert level to four, which is their highest alert level-- it's a four level system. I think it's important to note that this was pretty unusual for CVGHM. Typically, the alert level 4 in Indonesia is reserved for imminent eruptions. But at this point, we only had these VTs that were distal to the volcano. There were really no signs seismically of shallowing magma. There were no low-frequency earthquakes, no tremor, anything like that. Of course, there was the increased fumarolic activity at the summit. Because of this huge VT swarm and the lack of shallow signs of magma movement, and then along with the history of large, fast eruptions at Agung, there was a lot of uncertainty, plus the large number of people that needed to be evacuated, CVGHM decided to go to alert level 4 here. If you go to the Syahbana paper, the overview paper, there's a whole section on our eruption forecasts during this crisis and how they evolved, and you can read more about the uncertainty during this crisis. After the alert level was changed to 4, seismicity began to decrease, It waxed and waned, but for the most part, the trend was downwards, and fumarolic activity decreased as well, and we know in retrospect that deformation was holding pretty stable at this point also. After about a month of decrease in seismicity and a lack of further changes that indicated that interruption was imminent, CVGHM was under a lot of pressure to change the alert level, so they did. We lowered the alert level to 3 by late October after seismicity had stabilized. It was really only about a week after that, that we started to see the first signs of shallow magma movement-- the first-time that we started to see that seismically. There was a spread in the hypocenters of earthquakes. Instead of just being located off to the northwest between Agung and Batur, seismicity began to spread to the northeast. Then on November 9th, there was a large magnitude 4.9 earthquake with a large aftershock sequence. You can see that in the top right of this helicorder here. That earthquake and all those aftershocks were up in the northeast,  more or less like a new location. We interpreted that as some redistribution of stress in the crust as some change in the system that produced these earthquakes in a new location. It also appeared that seismicity was shallowing and coming closer to the summit. Given the network at this time, we'd installed more instruments by this time-- but our locations for the smallest events still weren't that great. We couldn't say for sure that earthquakes are starting to happen under the summit, but there did appear to be a shallowing of earthquake activity. There was also around the 9th of November, some small episodes of tremor too started to catch our attention. But then it was on November 12th that we really saw like first clear indication of shallow magmatic activity, seismically. There were large tremor signals that were larger than the cultural noise from the mining that you can see here. It's most clear during these lunch hours when this cultural noise would generally decrease, but there's still a lot of tremor there. There's also a couple of large LPs that happened during this time. This was our first indication seismically that something had shallowed in the system. But this was really the only day where we had these indications. We are still working in this distal VT model based on the White and McCausland paper from 2016 that we had a volcano that hadn't erupted since 1963. We were working under the assumption that it was fairly well sealed up. A fairly good interpretation was that these earthquakes off to the flanks of the volcano were being caused by pressurized aquifers, activating regional faults off to the flank of the volcano, and then we expected-- based on that model --that as soon as magma starts to break into the hydrothermal system and start to shallow, we should see a lot of noise from that. We should see a lot of VTs proximal to the summit as rock starts to break it, and we should see a lot of tremor and LPs as magma starts to mix with that hydrothermal system. We saw this on November 12th and thought, "Okay. Here's our first sign of that shallowing activity. We should see more if things progress," but then we didn't. Things continued to quiet down for the next week. During this time, VDAP's gas team was now in Bali, and they were working with CVGHM. By the 21st, they had finally put together this drone-based DOAS measurement. On the morning of the 21st, they pioneered their first gas flights with this Multi-GAS machine and flew it over the summit of Agung, and they detected a huge CO2 anomaly from the plume, which they interpreted as a significant sign of unrest. Sure enough, about nine hours later, the volcano erupted. We had the first phreatomagmatic eruption. The first eruption was fairly small, and it was not recorded seismically on our network. Our closest station at this time was about four to four-and-a-half kilometers away, and we didn't record it on the network. It wasn't until the later eruptions later that afternoon and the next day that we started to record tremor from those eruptions. Here are some pictures of that first eruption, and then some of the later eruptions that we did record seismically. These small phreatomagmatic eruptions continued for a couple of days, and then on the 25th, there was another large phreatomagmatic eruption, and then lava started to fill the crater on this date. All of our observations of lava at the surface are satellite-based, so we don't know exactly when lava started to effuse onto the crater floor, but it's likely coincident with the swarm of approximately 20 large LPs that occurred around 7:00 in the morning on November 25th. These were by far the largest LPs we had seen so far at Agung, and they all have incredibly repetitive waveforms,  but different amplitudes. We interpret these as being the signs of magma effusion onto the crater floor or at least closely associated with that. After these larger eruptions and the effusion of magma onto the crater, CVGHM raise the alert level back to 4 and remained there for several months. Then after this, most of the seismicity that we're recording at Agung was dominated by low frequency earthquakes. The VT swarm died away and we just had these large LPs that were occurring. That's the overview of the eruption starting with the unrest and then leading all the way through the first magmatic activity. Just to review the challenges from this crisis response: there was the large amounts of seismicity on our fairly limited network, the month long decrease in activity prior to the onset of eruption, and then the large population that we had to evacuate. After the fact, we were able to revise our model of how we thought about this unrest based on some InSAR work that Albino et al. did. This paper is from 2019 and there's a link earlier in the presentation. Based on this InSAR work, we know that there was a shallow dike intrusion between seven and 15 kilometers depth right in between Agung and Batur. Then we also know from their InSAR data that between the 8th of November and the 21st of November, there was deflation from this dike source. Interpretation is that magma started to withdraw sometime between the 8th and the 21st. These are figures from that paper. This is their model dike here in between Agung and Batur. These are their InSAR images that they use to model that deformation. Starting on the 21st of November all the way through the 8th of November, there's inflation at this dike source. Then the next image is on the 20th of November, one day before the first phreatomagmatic eruption, and there's a significant decrease in volume here. The idea is that there's some ascension of magma between those two dates. This is their working model. They interpret some deep storage zone and then intrusion into a dike, and then from that dike magma then ascends to the surface. The model presented in the Syahbana paper is a little different. It has deep storage zone directly underneath Agung and then a dike going off in between Agung and Batur, and then magma continuing to ascend up to the summit of Agung. After we revised our model based on these InSAR results and after this crisis had finished, I wanted to know what more I could learn from this earthquake data at Agung, and if there is anything we could do to better identify changes in activity at Agung with the network that we had. I wanted to look at repeating earthquakes. I wanted to do this because you can study repeating earthquakes with just one station if that's all you have. We did have more stations here, but you could use any network you wanted. We don't need to know the earthquake locations, which is going to be difficult given our network and some of the small magnitude for some of these earthquakes. We also didn't have a long time series of data, you have to analyze for some other methods. I'll share with you some of my results from this work. I used the REDPy software, which is freely available on GitHub and was written by another USGS employee in California, Alicia Hotovec-Ellis. I know for the seismologist in Ecuador that were at lavas last year in Colombia, they heard a lot more about the details of REDPy and things like that. Just to review, the repeating earthquakes are any earthquakes that have very similar waveforms, and are thus interpreted to be the result of the same source process happening in the same location. So you know that if you've got repeating earthquakes, you know that they're in the same location even though you don't know their absolute location. These repeating earthquakes occur everywhere. They occur in tectonic settings, in glacial settings, and in volcanic settings. Since doing this work, I've dived into the literature on repeating earthquakes at volcanoes, and I like to talk about these four specific examples of different ways that repeating earthquakes have been used at volcanoes. The first one is actually Shishaldin, and here they use repeating earthquakes to study repetitive gas driven explosions. They're actually looking at repeating seismic signals from explosions to help track this stable source at the summit. Another example is Augustine, which is probably the example most similar to Agung. The eruption at Augustine in 2006 was preceded by a large VT swarm and locations and repeating earthquakes after the fact helped identify how this vertical dike propagated up towards the summit. Then Soufriere Hills in Talica, these are places where tracking repeating earthquake families helped show how the system was changing or evolving based on repeating earthquake families continuing throughout explosions, thus indicating that the shallow system had not been disrupted significantly by explosions that were happening. Then during certain explosions, the earthquake families would all change and there would be new sets of families. So it helps tell that there's been some more significant change in the shallow system. At Agung, I started looking at the repeating earthquakes starting in mid-October. I chose mid-October because this is when we had our larger network installed. I ended up using about five stations to do the study. You don't need five stations to study repeating earthquakes, but there was a lot of telemetry issues and stations going in and out for short periods of time. So using five stations helped me fill all those time gaps. Mid-October is also when seismic activity started decreasing. This was the time that I was interested in what could we say about changes in the system based on seismicity as earthquake rates are actually decreasing, leading to these eruptions. Here's a little graphic of earthquake families and what they're doing starting mid-October and then leading through the onset of eruptions. This top chart here is just the frequency of all earthquakes detected. The red circles are the repeating earthquakes, the ones that actually cross correlate with another earthquake and create a family. The black ones are orphans, so-called orphan events, meaning they don't cross correlate with any other earthquake in this entire sequence. On the y-axis here, this is the frequency index and this is just some arithmetic way to define the peak frequency of these earthquakes, basically as a proxy for telling whether or not they're VTs or low frequency earthquakes. Then in this bottom figure, these are all the earthquake families that are identified, and all of the blue families, these are earthquakes that have a high frequency index, meaning they're probably VTs. The red ones are earthquakes with a low-frequency index, meaning they're LPs or low frequency events. So based on all these earthquake families, there's three phases that you can identify based on changing seismicity. The first is what starts at the beginning here in mid-October and would have started before then. But starting in mid-October, there's all of these earthquake families here happening simultaneously, all these VT earthquakes that are happening simultaneously and have a large number of earthquakes per family. These are all the earthquakes that are happening over on the flanks and between Batur and Agung. The interesting thing is that around November 8th and November 12th, most of these families, by November 12th, have ceased. There's no more activity on the faults that are causing the earthquakes. But there's new families of earthquakes that start to occur between November 8th and 12th. These are the earthquakes that were a lot closer to the summit of Agung. Some of these are the earthquakes that we identified as the early LPs, even though they've got higher frequency indexes. These are all proximal VTs and LFs happening around November 8th and 12th, coincident with when most of the seismicity proximal to that dike that intruded has stopped. Then on the 25th of November, there's another major change in the seismicity and the repeating earthquake families. It starts more or less with that family of approximately 20 large LPs that I showed you earlier during the eruption overview. It starts with those 20 LPs, more or less coincident with the onset of lava fusion. Then after that, there's a mix of these proximal VTs and low-frequency earthquakes. Another thing to note is during this phase where we mostly have these VTs out on the flanks, most of the earthquakes happening during this time belonged to one of these repeating families. About 60 percent of them are repeating during this time. By November 8th, which is that last and start date that showed inflation, and the first time that we saw a tremor, by November 8th through to 12th, most of these earthquakes are now orphans. Only about a third of the earthquakes are repeaters. Then starting with the large low-frequency earthquakes on the 25th, only about 20% of the earthquakes are repeaters. The percentage of these repeating earthquakes just keeps decreasing. As we go into this phase, we see that a lot of these families last a lot longer than families earlier on. A lot more of these families are long-lived. This is just a schematic of that same figure before and I just put it into an interpretation to go along with the conceptual model we have for unrest and eruption at Agung. This first phase of earthquakes where most of them are repeaters, this is what I call the intrusion phase where we have a ton of simultaneous earthquake families. We're interpreting this as repetitive failure on all of these small, tiny faults proximal to the dike. It makes sense with repeated failures and rapid reloading of shear stress on all of these faults. Even though a majority of these earthquakes are repeating, there's still a large percentage that they aren't. There's still a large percentage of these orphans, which I think just speaks to the diversity of faulting orientations in the shallow crust. So you have this one source that just keeps reloading shear stress on these faults, but there's still a ton of faults that are slipping and failing and then not failing again. So there's still a large amount of orphaned seismicity during this time. Then during the transition phase, these earthquakes are associated with the ascension of magma from the dike up towards the summit. Then by the time we get to the magmatic phase, after the lava has extruded to the surface, the repeating earthquake families here behave very similarly to any other volcanic system that's extruding magma. They behave very similarly to repeating earthquakes observed at Soufriere Hills and other places, there's a mix of VTs, low-frequency earthquakes, the families tend to last a long time, but most of the earthquakes are actually orphans. So in this case, at Agung, about 80% of them are orphans, probably from destructive sources happening up in the shallow system. The implications for monitoring and what tools do we want for monitoring RSAM and earthquake counts have been really useful for a long time and always will be useful. Classifying earthquakes into VTs, LPs, tremor, all that stuff is also very important, but it could also be time consuming and quite challenging during an eruption crisis like at Agung. For me, retrospectively analyzing earthquakes and organizing them into their respective families makes it easier way to make sense of what seismicity is occurring at the volcano and what processes it represents. You only need one station. You don't need to be able to do earthquake locations. It's just based on STA-LTA algorithms and then cross-correlation. That can be problematic and it can be challenging, but it's also very simple. It's fairly easy to set up in real-time if you're already familiar with the techniques and the software and everything like that. That's all I have for you today. I think most of the people participating here aren't seismologists. If you have questions about the rest of the eruption, I'll do my best to answer those and certainly if you have any questions about the repeating earthquakes or anything that I've focused on specifically, I can answer those. Thank you.