Grimsvötn is heading for an eruption. There can be no doubt about that. Of course, it is always heading for an eruption. This volcano has ADHD. For Grimsvötn, more than a decade of brooding is unusual: normally it just throws it out. A misplaced snow flake can set it off. And it produces not only volcanic ejecta. In hindsight, it was unwise to grow a glacier on such a hot head. The ice melts and collects in a subglacial lake. Once enough water has collected, it lifts the constraining glacier and escapes. The jökulhaup that follows heads straight for the south coast, carrying big blocks of ice with it. The volume of water Grimsvötn generates exceeds the volume of its volcanic ejecta. This is the ultimate mountain of ice and fire, with a temper to match.
Grimnsvötn is heading for an eruption: no doubt about it. But science wants to know more. It wants to understand how the mountain works and why (and when) it erupts. That means making a model, and to validate a model requires making a prediction. If the prediction come true, the model is deemed to work – for now. If it fails, science goes back to the drawing board. And it is essential that the prediction is about the future. Predicting the past does not count: that is inside knowledge which already went into making the model and so can’t be used to validate it.
Grimsvötn will erupt. A year ago, Carl and I had a challenge, to predict when the next eruption would be. I used math and came up with January 2020 to June 2022. Carl used
magic sorry insider knowledge insight and knowledge and came up with December 2018 to December 2019. So far, nothing has happened, so we were both right. But when the activity increased during the past month, we decided to revisit our predictions. After all, a prediction is only a validation if you check whether it worked! We are not politicians: we cannot quietly ‘forget’ what we said last year. Science is brutally honest, and a wrong prediction is not a failure but an opportunity. It improves the models. Carl wrote his update last week and failed to change his prediction. It remains December 2018 to December 2019. So what about my prediction? Is it time to throw in the towel?
The closer to the eruption, the better the forecast is likely to be. I had a bit more time than Carl so decided to wait, and see whether Grimsvötn would do anything that helped my models. If that gave me an advantage over Carl, who am I to complain? Against
magic knowledge, I need all the help I can get! And indeed, things did happen.
Volcanoes are like metal fatigue. A minute crack forms, and over time it grows into a full scale failure. The final failure is certain. The speed at which the cracks grow it accelerates as the crack grows: it is self-reinforcing. But the ‘when’ of fatigue failure is not predictable because it depends on the chance stress events which first cause or enlarge the cracks. So it is with volcanoes. The crack can suddenly and unexpectedly grow. A year ago, Carl wrote: In fact, there could at any time come a rogue earthquake swarm, or a single M4 earthquake, that punches the value [of cumulative stress] fast into the eruptive ranges. It is though not as likely judging from previous behaviour of Grimsvötn. But this week, Grimsvötn did do just that. A single quake pushed up the plots to the danger area.
But before going into how this affects the predictions, first a bit of background. There is a lot of background.
Grimsvötn is Iceland’s most active volcano. It is also the best hidden one. Hiding an active volcano is not easy: an extended icecap helps, but eruptions leave signs even if the area is inaccessible. The fires above the glacier were a dead give-away that there was a volcano somewhere amidst the ice. The frequent jokölhaups were another.
And ‘frequent’ is the word. For the past 800 years, Grimsvötn has erupted on average every decade. Before that, it is was even more frequent. Most eruptions are small, around 0.1 km3, such as the 2004 eruption. But when the occasion is right, it can also go bigger, as it did in 2011 when it had its biggest burp for 140 years. And 2011 was by no means the worst it can do. The so-called Saksunarvatn ash is a 10,000 year old layer of tephra found across the Faroe islands, northern Europe and Greenland. The composition of the ash shows that it came from the Grimsvötn magma reservoir. In the Faroe islands, the ash is in places a staggering 45 centimeter deep. In addition to this large explosion, there are several smaller tephra layers close to the thick ash, showing that several smaller eruptions occured within about 500 years of the main event. Grimsvötn really threw a tantrum! The total tephra volume of the Saksunarvatn ash may be over 30 km3 although the precise value is very poorly known – so much ended up in the ocean. In comparison, even 2011 was just a hick-up.
Grimsvötn also mixes up eruption styles. Most often it does explosions from the summit or along small fissures near the summit, but on occasion it also feeds effusive eruptions from a much more extended rift system, 90 kilometers in length. Mount Laki formed from Grimsvötn magma. And in 1783, the rift again gave way and the largest Icelandic eruption since Eldgja devastated the country. It is called Laki, although Laki mountain itself did not take part. After the Laki eruption, the listed population of Iceland had decreased by 25%. This volcano packs a punch.
In spite of its hidden nature, the Icelandic people were well aware of this invisible volcano. The ice-covered lake of the Grimsvötn caldera was known very early on. It may have been more accessible in the past, before the major glacier growth of the little ice age overran the local farms.
Grímsvötn is first mentioned in an Icelandic letter dating from around 1600:
`… fire from within the lake […] that is called Grímsvötn in our language, flinging out, higher than the highest mountains and with tremendous power and destruction, huge amounts of pumice and gravel that is catapulted and dispersed to the most far-flung parts of the country, damaging meadows and terrifying the inhabitants’
A description of an eruption of Katla in 1625 mentions Grimsvötn in passing:
‘… there is also a common rumour that to the north of Glómagnúpssandur , next to Skeiðarárjökull, there is a lake called Grímsvötn, which flows with fire, ice and water like this glacier [i.e. Kötlujökull] and what so often happens is that the fiery eruption first bursts forth from the centre of the aforesaid Grímsvötn, from which flames burn like a pyre or bonfire’
So it was clear that Grimsvötn was known and had been visited. The Atlas Danicus (1684) contains an account of a volcanic eruption in Grímsvötn and Grímsfjall in the so-called Gríms Vatna Jökull glacier. Maps which for the first time showed the caldera at the right location appeared in 1730. But as the glaciers advanced, the stories became less clear, and began to sound almost mythical: ‘Men have seen 15 separate fires burning on the surface of the lake; what the cause of this can be, people find hard to say, except perhaps that it is caused by an overabundance of sulphur deep in the Earth, for whenever folk ride by these glaciers there is the most powerful odour of sulphur there could ever be’.
After Laki, it seems that the memory of Grimsvötn was completely lost, even though the jökulhaups and eruptions continued. Everyone who knew the stories had either died or moved away during this disaster. In June 1903, pastor Magnús Bjarnason wrote about an on-going jökulhaup, showing how little was left known. He attributes the eruption to the southern edge of the glacier:
[the jökulhaup] occurred at the same time as an eruption in Skeiðarárjökull (?) [nearer Þórðarhyrna] which often happens, to a greater or lesser extent, when there is an outburst in Skeiðará river. Flames arose north and west of the so-called Grænafjall mountain, which lies to the north of Eystrafjall, on Thursday 28 May, upon which the floodwaters increased as the glacier melted, searching for channels with such power that they burst open the front of the outlet glacier and surged forwards with blocks of ice that, according to the post bearer, were 60 feet high, and over a wide area there were others on top of them another 25– 35 feet thick. [..] The most terrifying jökulhlaup was on Thursday and Friday night, when the eruption had begun and was at its peak. Water spurted up in high columns through cracks opening up in the glacier and all the ground and houses in Skaftafell and Svínafell in Öræfi shook so much in an earthquake that windows were broken in Skaftafell, and thuds and bangs were heard as far east as Hornafjörður, for there was a westerly wind. The evening and night sky were lit up by the volcanic flames that stretched high into the sky and from its smoke came flashes of lightning that brightened the land and rivers all the way out to Álftaver, and this was later than ten o’clock at night. A cloud of smoke lay over the Öræfi district, concealing the mountain tops above Sandfell and Hof in black billowing waves with flashes of fire and lightning streaking from them above the farmsteads. Even the earth itself shook when the glacier was cracking open and the explosions louder than the firing of any canon. All of this combined was so terrifying that the folk of Sandfell (Pastor Ólafur himself had recently just left to the west with two children) fled eastwards to Fagurhólsmýri. Fortunately, there was no consequent ash fall, and the cloud and eruption were much milder the next day, indeed the jökulhlaup itself was beginning to abate.
In 1919, two Swedish students, Hakon Wadell and Erik Ygberg, set out to explore the possible causes of the jökulhlaups that frequently emerged from beneath Skeiðarárjökull, unaware that they were retracing the past. They found the glacier very difficult to traverse: they had four horses pulling the sledge, but this was hard going because of a 10-cm thick cover of tephra, which had come from the large eruption of Katla. Suddenly, the front horse stopped, refusing to budge or go any further forward no matter how hard the students beat it. A slight break in the fog appeared, and they found they were standing on the edge of a precipitous cliff. The horse had sensed the up-draught at Grímsfjall! ‘When we came to the head of the horse, there appeared below us a crater comparable to the side of hell in size, though even that would have been expanding in these recent and most difficult times.’ The crater welcomed them with ‘… the rumbling of ice rocks that continually crashed from the cliff walls, hundreds of metres high, down into the crater’s basin where they melted in warm, emerald green water.’ Typical students, they toasted their discovery with cognac and christened the caldera Svíagígur (‘Swedish crater’), a name that did not catch on.
They continued to cross Vatnajökull but had to abandon their horses due to the conditions. The went back afterwards and two of the horses were recovered; the other two and a dog are presumed to have died. The atrocious conditions of the ice and the weather left their mark: Ygberg never fully recovered his health after this trip. And they failed to convince the world that there really was a major volcano hidden in the icecap, something they both reportedly felt bitter about in later life. Who believes students?
Grimsvötn erupted again in 1934, and this time people were able to travel to the caldera during the the eruption. It left no doubt that this was the origin. The travel was as difficult as it had been in 1919: the thick layer of ash made sledging and skiing impossible so the explorers had to walk for 20 kilometers towards the eruption plume. There may have been some health and safety violations: ‘Clumps of slag, pieces of lava and chunks of the crater bed lay in heaps and mounds, in some places as if raked together like haystacks, and the nearer we reached the edge of the crater, the more grandiose everything became.’ And more dangerous: they reported that plumes of steam, yellow, green and steel-coloured, billowed up through a gaping wound in the ice sheet while black pillars of smoke were flung into the air from the pillar lava. Clouds of steaming sulphur seared their lungs as cinders rained down onto the glacier and hissing and sucking sounds were heard all around them as well as a loud rumbling from the deep. Their observations confirmed that the two Swedish students, 15 years earlier, indeed had discovered Iceland’s most active volcano.
The icecap that covers Grimsvötn is Europe’s largest glacier (by volume; by area it is the second largest). Snow and ice have a way of smoothing over the cracks and calderas. This glacier indeed covers a multitude of sins.
Remove the glacier and no fewer than five volcanic centres appear: Öræfajökull, Bárðarbunga, Kverkfjöll, Grímsvötn, and Hamarinn. They sit on a 700-meter tall massif, bisected by a deep valley. The valley runs runs south-southwest to north-northeast, parallel to the main spreading axis. The main volcanic activity is in the ridge west of this valley, where Grimsvötn has pride of place.
The glacier around Grimsvötn contains a detailed record of previous eruptions, in the form of layers of ash. This has been used to derive the eruption history. Between the years 1200 and 1800, there are 66 separate ash layers, each from a local eruption within the icecap. Five of these layers indicate that two different volcanoes were erupting in the same year, so the total number of eruptions in these 600 years was 71. Eruptions on the fissures outside the glacier will not be included. Adding the 15 more recent eruptions bring us to 86 eruptions between 1200 and 2000. Two thirds of the ash layers come from Grimsvötn: it erupts more than twice as often as Bardarbunga. In fact, Grimsvötn is responsible for over a third of all eruptions in Iceland. Not bad for a volcano only officially discovered in 1919!
Going back in time, the frequency of the ash layers further increases. The peak activity was between 1000 and 2000 years ago. This was a time when the glaciers were less extended, and perhaps this made eruptions easier.
Recent eruption were in 1903, 1922, 1934, 1954, 1938, 1983, 1996, 1998, 2004, 2011. In spite of the high frequency, there were no eruptions in Vatnajökull between 1938 and 1983, 45 years of quiescence. It now seems unbelievable. But this is part of a pattern. The ash layers show that times of frequent eruptions were interspersed with times of less frequent ones: these two phases alternate every 50-80 years. Around 1950 was the depth of a period of infrequent activity. Large eruptions, such as the 2011 one and 1873, seem to occur about every 140 year, i.e. once in a combined high/low frequency period.
Eruption volumes are not well known because so much is intercepted by the ice, but are in the range 0.01 to >0.5 km3. The 1996 Gjalp eruption produced 0.45 km3. Over the past 1000 years, Grimsvötn has produced an estimated 15km3 of lava and14km3 of tephra, or 21km3 DRE in total (remembering that tephra is three times less voluminous as DRE). Of course, the lava output is almost entirely due to Laki!
The minerals that are found in the ashes trace the conditions under which the magma crystallized. Certain minerals form at certain pressures and temperatures. The ones found in the ashes formed at a depth of 15-17 km, and that is where the main Grimsvötn magma chamber appears to be. The crystals in the Laki eruption formed at the same depth and a temperature as those found from the 1823 Grimsvötn eruption, further strengthening the hypothesis that Laki lava came from Grimsvötn (other evidence is the lava composition and the fact that Grimsvötn itself erupted at the same time as Laki). The summit eruptions themselves come from much shallower magma, fed from the deeper magma storage. The most recent eruptions have minerals from a wider range of temperatures and pressures, more so than during the 19th century. The model for this is a series of dikes and sills, of different depths, fed from the 15-km reservoir, where the efficiency with which magma moves from the deeper reservoir to the sills has been decreasing with time. The 15-km deep reservoir is likely fed from a neutral buoyancy layer at the mantle-crust boundary, 30 km deep.
Holuhraun had a more primitive magma, but with similar crystallization depth as Grimsvötn. The eastern volcanic region may well have a series of such magma reservoirs at 15+-5 km deep, which stores magma for a shorter time for Bardarbunga, and longer for Grimsvötn. Laki was fed from such a reservoir, either directly or indirectly, where the data shows that the same reservoir also feeds Grimsvötn.
A deep connection between Bardarbunga and Grimsvötn seems possible. A paper in the last year pointed out that in the year before Holuhraun, Grimsvötn stopped inflating. It implied that magma was being diverted. The amount of magma involved was rather small (0.015 km3), but it indicated a connection. Either magma was transferred from Grimsvötn to Bardarbunga, or there was a pressure imbalance in a deep common reservoir (30 kilometer deep). The latter seems more likely. As Grimsvötn 2011 showed some evidence for a magma recharge of the deep reservoir, it is possible that the same recharge also was the ultimate cause of the Holuhraun eruption.
The minerals found in the ash indicate the temperature at which crystalization occurred, i.e. the main storage chamber. This shows an interesting effect: the Grimsvötn magma system appears to be slowly cooling. The temperature of the Laki magma (1783) was 1140 C. The composition of the Saksunarvatn ash indicates the same temperature. But eruptions in 1823 and 1873 indicate magma temperatures of 1130 C, and the 2004 and 2011 eruptions gave 1110 C. A paper by Hadadi et al. (2017) find a cooling rate of around 0.1C per year over the past 200 years. The 2011 eruption was the first to show evidence for FeTi particles, a lower temperature mineral: previously the magma had been too hot for this mineral to form.
So how does an eruption proceed? A description of events leading up to the 2004 eruption by the IMO gives an idea what we might expect. Earthquake activity began to increase in the middle of 2003. Tremor was first seen at Grimsfjall in August 2004, each burst lasting 30 minutes. In October, the earthquakes increased in strength from mostly M1 toM2.
The IMO report notes that in October 2004, the GPS station at Skrokkalda (SKRO) started rising, by 40 mm in three weeks, and moved west. This was seen as evidence for strong inflation. But in hindsight (a popular alternative to insight), it is not so clear. SKRO is 56 km from the eruption site, and the the magma reservoir supply would have had to be enormous to give such an effect. Inflation is commonly seen around the glacier in late autumn, due to the increasing weight of the snow. Looking at the SKRO signals now, its motion was probably nothing to do with Grimsvötn.
On 30 October 2004,a jökulhaup occurred. This is in itself not unusual. The autumn snows can trigger them. For instance, the jökulhaup in 2010 occurred on almost the same date, 31 October. But it is quite possible that the increasing heat accelerated the snow and ice melt, filling the lake. However, the ice above the lake floats, and therefore melting it does not change the pressure on the surrounding ice. For that to happen, either ice from the surrounding glacier needs to flow down and fill the basin (a slow process which happens after every eruption), or ice and snow further away need to melt and flow towards the caldera lake. This requires activity on the short fissures around Grimsvötn, and in fact several times a jökulhaup came after an eruption north of Grimsvötn, causing melt water to flow into the caldera. This happened in 1861, 1867, 1892 and 1938. A summit eruption should not trigger a jökulhaup because it melts already floating ice. However, it is possible that a jökulhaup can trigger a summit eruption. It quickly removes a lot of weight from the caldera, and the lower pressure can initiate an eruption.
Two days after the jökulhaup, on November 2, the mountain erupted and Grimsvötn had again made its existence known to the world.
Now let’s get back to the topic of this post. Can we make a prediction as to when the next eruption is likely to happen? How well did it follow the predictions from a year ago? Earthquake activity has significantly increased since last year. Tremor has not yet been seen: even Friday’s (23 November) M3.2 event (unusually strong for the caldera) was purely tectonic in nature. That would suggest that we are at least a month away from the next eruption, if it follows the 2004 path, and it could still be years.
I modelled the earthquake activity with an equation also used for fatigue failure:
Here, tc is the time when things actually break, measured in days since the previous eruption. A larger tc means a slower build-up to the eruption. The factor k is a scaling constant, and M is the total cumulative moment, as plotted daily by IMO. The equation is normally used for number of events rather than moment (strength), and this version basically assumes that all individual events (earthquakes) are of similar size. That makes it rather difficult to deal with much a single much stronger earthquake, such as the recent M3.2. Should it count as a single quake, should it count for its strength, or should it be deleted from the data?
Now I quote myself from last year: The plot shows the results obtained with this equation. Green is for data leading up to the 2004 eruption, blue the 2011 eruption, and red for the current cycle. The data has been read off from the IMO plots and there may be some inaccuracies. I first fitted the equation to the data leading up to the 2004 eruption. The curve fits the data well if I take the numbers tc=2200 days and k=1.2. When I fit the data leading up to the 2011 eruption in the same way, it yields tc=2550 days, and the same value for k. Interesting, the curve suggests that the 2011 eruption happened slightly ahead of schedule, by some 100 days.
We can also now fit the current cycle in the same way. There are two problems with this: the cycle isn’t far enough yet to give a unique solution, and the fit is quite dependent on the single earthquake around day 1850. Regarding the first problem, the range of dates for which the fit is made to work is quite small. That makes is susceptible to noise, random fluctuations in the earthquakes. I went for a fit which works best towards the later dates, where the random noise becomes (hopefully) less important. For the second problem, I have plotted the cycle both with this jump, and without it (the lower red line at the end). Using the same value of k as found in the previous two cycles, that gives me two possible values for tc (noting that both are still quite uncertain!), tc=3200 days including the jump, and tc=4000 days excluding it. Say 8.5 to 11 years since the previous eruption. But be aware that the current cycle is still in an early phase and the fit could easily change – it is not yet well determined.
I can take the outcome as the possible range of dates for the time of the next eruption. Starting from June 2011, the early date corresponds to January 2020, and the later date is June 2022. There is a good chance the eruption may occur between these dates.
Looking back at what I wrote a year ago, I missed one aspect. Perhaps the 2011 eruption being ahead of schedule was because of the jökulhaup? Was the 2011 eruption triggered rather than purely spontaneous?
But let’s get back to the current events. How does it look like now?
Over the past year, initially the progression was slow. There was continuing earthquake activity, but it was not particularly strong and the total cumulative moment rose slower than predicted. But over the past few weeks, the mountain has kicked into action. Friday’s M3 quake has brought to total seismic moment to within a factor of two or eruption values.
Now let’s compare the progression with the predictions from last year. The plots below show the two figures, one with the large quake at day 1850 included and one with it subtracted, together with the two original fits. (If you wonder why the data doesn’t look as crisp as in the IMO version, I do not have the actual data so had to digitize the plot to get the values. This procedure is not perfect for a step function. The ‘big quake’ took place some distance from Grimsvötn and its relation to it is unclear, so there is a case for discounting it. Last week’s M3, on the other hand was within the caldera and is harder to ignore.)
On the left is the ‘fast’ fit. The fit is not great with the data continuously falling below the projections since the time the fit was made. The burst of quakes of last week has brought the curves closer, but it is not a convincing fit. On the right is the ‘slow’ fit. Here the fit is better, with the two curves staying closer together. The M3 earthquake has brought the data from below to above the curve, but it remains not far off.
It is interesting that in both cases, the data stays well below the model until mid 2014, after which it sharply increases. This change coincides with the onset of the Bardarbunga eruption. This was interpreted above as a possible diversion of magma. The plots here show that Grimsvötn recovered once Bardarbunga erupted. This shows that the issue was pressure, not magma transfer. As magma rose into Bardarbunga, it sucked magma in from a wider area and reduced the pressure underneath Grimsvötn. Magma does not flow from one reservoir to the next: the communication is through a deeper region.
So far, no reason to change the predictions of an eruption around 2021. But the fits are not great. Can we do better? There are two things we can change. First, during 2018, the activity at Grimsvötn was very low. In a way, this looks similar to the pre-Bardarbunga episode. Is it possible that Grimsvötn was affected by the magma rising into Öræfajökull? If I discard this phase, and just fit the recovery last week, it gives me the left plot below. Now the eruption is predicted for early 2021.
The second try is by assuming that the 2011 eruption removed so much stress that the onset of the build-up to the next eruption did not start immediately. Picture the Grimsvötn magma chamber as being surrounded by en elastic band (a bit like stretchable trousers). The build-up to the next eruption begins when the elastic band begins to expand. After the eruption, the band was larger than the remaining chamber, and so there was no stretch – compare your loose-fitting trousers after a crash diet. After some time, enough
weight magma has been added that the band begins to stretch. This is the moment that the clock begins to count. This situation with a no-stretch-phase is depicted in the plot on the right. I get a decent fit with an 800-day delay. The fit is not bad, and based on appearance (‘goodness of fit’) it seems better than the others. It has much less of a dip prior to the Bardarbunga eruption. However, the 2018 quiet months are still notably discrepant. In this fit, the eruption is predicted for mid-2020.
So there are still a range of possibilities. The next few months will tell us what will happen. All the models here predict that the flurry of earthquakes of last week won’t continue at that rate. There will be a decline again in activity, although perhaps not to the very low levels of last summer. If the rate continues at the high rate, then it is clear that the models here do not work well, and an earlier eruption can be on the cards. Otherwise, I find that mid 2020 to late 2021 is the most likely window.
One final point. In addition to the earthquakes, the local GOS (GFUM) showed sharp inflation over the past month. It looks convincing. But is it?
The main problem here is that there the area has a lot of snow in autumn, and snow and ice on GPS systems introduces errors in the output. GPS system on the icecap often show large variations in their readings this time of the year. The plot below shows a multi-year GFUM curve. The jump seen in recent weeks is exceptionally large. But it happens at a time of the year when this GPS is normally out of action, so there is not a lot of comparison data. So again it is too early to tell. We will have to wait for spring to see whether the inflation is real, or is caused by the onset of winter.
And that is where we are. We have a model, and we have a prediction. It suggests that we still have two to three years to wait. But somehow, I doubt that volcanoes feel any obligation to predictions. They do their own scheduling. They are complex systems. Perhaps what we need is not so much physics, but a study of volcano psychology. Grimsvötn still has ADHD.
Albert, November 2018
This post made heavy use of the book “The Glaciers of Iceland: A Historical, Cultural and Scientific Overview” by Helgi Björnsson, published by Atlantic Press in 2017 and translated into English by Julian Meldon D’Arcy. It has a wealth of information. I used it in particular for the historical background, and parts of the post follow the book very closely as I could not find a better way to phrase it.
Fun activity: how many
Icelandic volcanoes black dots can you see simultaneously?