The collapse of Anak Krakatau

Anak Krakatau, in better days

The parent is famous. The shock waves of the eruption of Krakatoa in 1883 reverberated around the world – in the atmosphere, in the sea and in the news media. This was the first large eruption in the era of instant communication. The eruption itself was luckily on an island at some distance from human occupation which should have limited the damage. But the tsunami came swiftly and with little warning. It was devastating.

Volcanic tsunamis are not uncommon. Over the past 400 years, there have been some 130 documented tsunamis caused by volcanoes, most recently by Hunga Tonga in 2022. Around 80 volcanoes have caused tsunamis in that time, many of them more than once – including Krakatau.

The name of the parent, by the way, was a typo. The correct name was Krakatau (meaning crab). In Portuguese it is Krakatao. After the 1883 eruption, The Times newpaper misspelled this as Krakatoa and this name stuck. But the typographical naming error is used only for the parent volcano, which no longer exists, except as a deep hole in the sea floor. It is not used for the offspring.

44 years after the destructive explosion, the volcano re-emerged, at the location where Verbeek, the guru of Krakatau, said it would. This was the first child of Krakatau. It did not live long but quickly disappeared again. There were two more, before the current Anak Krakatau (‘child of Krakatau’) appeared in 1929, grew to 30 meters high, and remained above the waves. When it became famous, in 2018, it was 89 years old.

Anak Krakatau is a basaltic-andesitic volcano which, like any child, loves erupting. The surrounding region sits in a shallow sea between Java and Sumatra, on a plateau of continental crust. Below is the subducting oceanic crust from the Indian ocean, which is what drives the volcanic activity. But the activity is not solely due to that. Due to a bit of rotation between Sumatra and Java, the area of Krakatau has some extension and this allows for magma chambers to form. Anak Krakatau sits above multiple magma chambers, situated from the moho at 20 km to 4 km depth. The eruption rate indicates that the magma rises up easily, at least at the present time. The magma chambers are not all the same: the composition of the ejecta of Anak Krakatau can vary a bit, and have even included dacitic material.

The parent volcano was very different: it had not erupted for 200 years before the end: its magma did not rise up easily. The difference between parent and child suggests that whatever had been obstructing the conduit was removed in the explosion, and therefore the obstruction had been located in the upper kilometers of the crust.

Anak Krakatau used its rising magma well: after its emergence in 1929, it build up a 338-m tall cone in less than a century, growing at an average rate of 4 meters per year or 10 cm per week. The same rate must also have occurred between 1883 and 1929, to get from the sea floor depth of 250 meters to the surface in that time, but this of course had not been seen. There was a brief interruption of the growth around 1950 when the cone actually reduced in height. But in 1959, the cone was 150 meters tall and in 1981 it had grown to 200 meters. At this time there was a change, with eruptions moving a bit to the southwest and the lava temporarily becoming more evolved. In 2007, the volcano was measured at 305 meters high, and the eruptions were intensifying. By 2018 it was 338 meters.

That was to be the end.

Hidden danger

The danger of Anak Krakatau had been recognized. A paper by Giachette et al. in 2012 pointed out that Anak Krakatau had chosen the wrong place for its emergence. It is not our choice where we are born – we just have to try to make the best of it. Neither do we have responsibility for the faults of our parents, but we still may have to deal with the effects of those faults.

It was known already in 1929 that Anak Krakatau was growing on a slope, with the ocean being deeper towards the west. Under water, the cone slopes considerably steeper towards the west than to the east. By 1995 that had not changed: the volcano had grown but had not filled in the hole to the west.

Giachetti and collaborators saw this and were concerned. Their paper in 2012 calculated the consequences of a potential collapse of the mountain into the hole. The abstract gives all the essential information, in remarkable foresight:

Numerical modelling of a rapid, partial destabilization of Anak Krakatau Volcano (Indonesia) was performed in order to investigate the tsunami triggered by this event. Anak Krakatau, which is largely built on the steep NE wall of the 1883 Krakatau eruption caldera, is active on its SW side (towards the 1883 caldera), which makes the edifice quite unstable. A hypothetical 0.280 km3 flank collapse directed southwestwards would trigger an initial wave 43 m in height that would reach the islands of Sertung, Panjang and Rakata in less than 1 min, with amplitudes from 15 to 30 m. These waves would be potentially dangerous for the many small tourist boats circulating in, and around, the Krakatau Archipelago. The waves would then propagate in a radial manner from the impact region and across the Sunda Strait, at an average speed of 80– 110 km/h. The tsunami would reach the cities located on the western coast of Java (e.g. Merak, Anyer and Carita.) 35–45 min after the onset of collapse, with a maximum amplitude from 1.5 (Merak and Panimbang) to 3.4 m (Labuhan). As many industrial and tourist infrastructures are located close to the sea and at altitudes of less than 10 m, these waves present a non-negligible risk. Owing to numerous reflections inside the Krakatau Archipelago, the waves would even affect Bandar Lampung (Sumatra, c. 900 000 inhabitants) after more than 1 h, with a maximum amplitude of 0.3 m. The waves produced would be far smaller than those occurring during the 1883 Krakatau eruption (c. 15 m) and a rapid detection of the collapse by the volcano observatory, together with an efficient alert system on the coast, would possibly prevent this hypothetical event from being deadly.

The research was perceptive and timely, and it gave time to prepare. But inexplicably, it was published behind a paywall! Perhaps for that reason it did not attract the attention it deserved. Geology does not do open science as well as some other fields of research. It is build on commercial interests, as much of geology is connected to mining. The publishers used the paper for income, rather than helped to disseminate the research as widely as possible. And so the paper was less read than it could have been, and nothing was done. The people who needed to know perhaps never saw the paper. The prediction came true, but it was unexpected.

Volcano-induced tsunamis are extremely dangerous. There is often little or no warning. They can come from submarine eruptions (as in Hunga Tonga), caldera collapse, pyroclastic flows entering the water (which happened in the Tambora eruption) or flank collapses. Ritter Island, in 1888, generated 10 meter high waves hundreds of kilometers away. Hunga Tonga caused two fatalities on the far side of the Pacific ocean, in South America. Krakatau’s tsunamis in 1883 killed over 30,000 people. (Around 1000 died from pyroclastic flows reaching the coasts, and the 3000 people who died on Sebesi (whether there were no survivors) died of a combination of tsunami, pyroclastic flows and 1 meter of ashfall.) The collapse of Mount Unzen in 1792 killed 15,000 – and it was caused by an earthquake, not an eruption. Volcanoes don’t have to erupt to cause a tsunami: the instability of a growing cone can be sufficient in itself.

Anak Krakatau had already given a warning with a small tsunami in 1981. And there were stronger warnings. INSAR measurements showed that after the 2008 eruptions, the southwest slope of Anak Krakatau subsided by 18 cm. No, the 2012 paper should have been widely distributed. Hiding it behind a paywall was problematic.

Rhymes and reasons

How was it that Anak Krakata, growing up on the heart of old Krakatau island, found itself on a slope?

Let’s go aback to the parent. Krakatau is a cyclic volcano which starts basaltic (as did Anak Krakatau) but evolves to dacitic magma over a cycle. This change is unusual for an Indonesian volcano. The cycle is terminated with a large eruption, and the cycle restarts. Anak Krakatau is at an early phase in the cycle.

Before the 1883 eruption, there were four islands in the group. Sertung and Panjang were remnants of a much older phase, seemingly delineating an old crater wall. These islands still exist. There was another small island in between them, called Polish Hat. It was destroyed in 1883.

The main island of Krakatau was much larger, and carried the volcanic activity. It is indicated by solid line in the image. It contained a number of separate volcanos: Rakata, the largest and oldest, extinct at the time, Danang, a double peak, and Perbuatan. Perbuatan erupted in 1680 and again in 1883. Danang probably also erupted in 1883. There were a further series of active vents along the line between Danang and Perbuatan. Note that the west coast side of the island was not explored in 1883 as it was too dangerous, so it is not known whether there was activity there as well – nor do we know where exactly this coast was located at the time of the final eruption.

From Deplus et al. 1995

Anak Krakatau grew up on the line between old Danang and Perbuatan, where Verbeek had predicted Krakatau would re-appear. It appears to be using the same magma path as the parent.

The original Rakata volcano, Panjang and Sertung islands may trace the outline of an ancient caldera. The large eruption thought responsible for this has been dated to 60,000 years ago based on drill data. However, the caldera may also have formed in a series of caldera-forming eruptions of which the one 60,000 years ago was the most recent one, in which cases the caldera may have build up over time.

Rakata already existed in the previous cycle, but it was largely destroyed 60,000 years ago. A new cone grew up on the remnant, reaching 800 meter high before activity moved to the new andesitic cones of Danang and Perbuatan. Half of Rataka survived the explosion in 1883 which removed almost all of the rest of the island. It seems to have collapsed along the old caldera fault, with the part situated outside of the fault remain standing. (The ‘B’ in the image shows the location of a rock pinnacle that survived in 1883, called Bootmans Rock. It seems to no longer exist.)

Source: Duplus et al. 1995

Bathymetry has revealed the shape and size of the caldera that formed in the 1883 euption. It lies between Rakata and Sertung . The hole covers much of the old island, but not all. The location of old Perbuatan and Danang is near the edge of the hole: the main caldera is centred a bit further west, out at sea. The cause of this offset is not clear. Were the Krakatau volcanoes actually located on an old caldera rim? It is not the same direction as the one determining Rakata’s collapse. Perhaps there was a complex structure with overlapping ancient calderas, each with their own ring fault. Once at fault, always at fault.

This meant that when Anak Krakatau started growing, it did so near the edge of the 1883 caldera, not at the centre. This is why it became build on a slope. This slope was the side of the caldera! It was not a safe place to grow up. This is why the 2012 paper warned -unheard- of the danger of collapse.

The year of living dangerously

The 2012 paper was based on a height of the volcano of 300 meters. By 2018 it was higher than that. There had been a series of intensifying eruptions which caused growth. The figure below shows the strong eruption sequence starting around 2008 and lasting until 2013. After 5 years of quiet time (a normal interruption for Anak Krakatau), July 2018 saw a new outbreak, now at higher intensity. The eruption peaked in September, reaching phases with time averaged eruption rates of 3 m3/s. It calmed down a bit after that, apart from a brief phase in mid November. The erupted volume from July to December was 0.025 km3. Much of this ended up on the southern slopes. The scene was now set for the events of 22 December.

Eruptions of Anak Krakatau. Source: Walter et al. 2019

In hindsight, there had been trouble brewing before these eruptions. From January 2018, the southwest and southern side had been moving 4 mm per month westward, while at the same time the slope was subsiding. This movement accelerated from July when the eruption restarted. But it was not noticed at the time.

On 22 Dec 2018, the eruption resumed, although the activity was relatively minor compared to the intense eruptions of September. Hot tephra was deposited on the slopes. The activity ended after about 6 hours. An hour later, there was an earthquake or explosion. Seismic signals indicate that the collapse started two minutes after this. And two minutes later, half the mountain had avalanched into the sea. Four minutes from the initial, small earthquake, Anak Krakatau was gone.

Was the collapse caused by the eruption or by the earthquake? The eruption had ended an hour earlier, so could not have directly cause the collapse. In fact, the ejecta from the collapse show that there was no hot magma involved. But perhaps it is a moot question. The eruptions of 2018 had left the mountain unstable, even if they had not provided the trigger. The cone had started sagging already a year before, and perhaps even as much as a decade. Every eruption contributed its bit. Was the final eruption on Dec 22, if a minor one, the proverbial drop which caused the bucket to overflow? Or was the collapse already set in stone? Had the fast growth in an unstable location made the outcome inevitable?

The collapse itself was not directly seen by anyone. It is however revealed by the seismic signals. These show that the initial earthquake or explosion was equivalent to an M2, so a small event. The signature was different from normal earthquakes, with a long-lasting high frequency component. After this initial event there were two minute of seismic silence, before the landslide started. The slide itself lasted 90 seconds, with a total energy equivalent to an M5.3 earthquake. It was also detected as an atmospheric infrasound wave, but this lasted a bit shorter, around 1 minute. This probably means that after 60 seconds the landslide entered the sea, and was no longer disturbing the atmosphere. The landslide was followed by a 5-minute phase of strong volcanic explosions, caused by the sudden removal of the weight of the cone. Over the following hours, volcanic tremors continued, but the seismic signature differed from those in eruptions before the collapse. They had the appearance of steam-driven explosions, caused by the magma in the conduit which was now at the surface. These explosions, including the 5 minute phase of intense explosions, also released a large sulfur cloud, something the eruption before the collapse had not done, and produced significant tephra that covered the remains of the volcano.

From Walter et al. 2019

The landslide had a ‘slide angle’ of 12 degrees, a bit less than the 18 degrees of the downslope in the sea. Clearly, there had been a plane of weakness within Anak Krakatau. This may have been caused by the southwest migration of the volcano 40 years before which meant it had started growing on the ejecta of the previous phase, rather than adding to the summit. Later we’ll see another past event that may be more likely as the cause of the fault. Even though the fault was buried, there was still a contact plane, not far above sea level. This may have been the ultimate weakness in Anak Krakatau: you can bury the past but it still affects you.

The physics of the event was not addressed in the paper by Walter et al. Imagine a rock sitting on a sloping surface. Gravity tries to pull it down the slope, with a force that depends on the weight of the rock and the angle of the slope. Friction keeps it in its place. Stationary friction, when objects are not moving, is quite high, so this is an effective way of keeping things in their place even on a slope. But the friction of a moving object is much lower. You will know this from experience: pulling something from a stationary position requires more force than what is needed to keep it moving afterwards. That means that once the rock begins to move under the force of gravity, it finds itself unable to stop. I have discussed this effect before, in the collapse of Kazbek.

The slow sliding of the slope of Anak Krakatau in the year(s) before the collapse was different. This was most likely deformation (shear) of the cone or of the region around the connecting layer. The static connection remained locked in place during this time. Earthquakes also act in this way. Imagine two tectonic plates moving past each other in a transform fault (the San Andreas come to mind.) The plates are moving past each other, smoothly and continuously. They have to: there is no way one fault can stop an entire plate! But while the plates move, the connecting plane is stuck. This makes the plate close to the fault line deform, in a shearing pattern. This deformation puts more stress on the fault plane, until it suddenly gives way, and the deformed zone can catch up with the rest of the plate.

Here is an example from Turkey, taken from the post on the North Anatolian Fault. The wall was built 70 years before the photo was taken. The fault, which is stuck, is just behind the man, at the location where the wall bends. You will understand that originally the wall was straight. The plate on which the photographer stands moves to the left at 1cm/yr, or 70 cm over the time the wall was built. At the bottom of the picture, the wall has moved by this much. The fault hasn’t moved at all. (One day, it will.) The region in between is where the shear deformation takes place. The angle of the wall shows this deformation in action. Once the fault gives way, this region will jump left to catch up with the moving plate: this is the earthquake. And now the wall will consists of two straight sections, offset by a meter or more, depending on when the earthquake happens. The property protected by the wall will now have a strange shape, but this is what happens when you build on a fault. In a thousand years time, the two parts of the property may have become unconnected. This has happened even to an entire volcano.

Underneath Anak Krakatau, out of sight, the same thing was happening. The connecting plane (the decollement) was stuck, but the mountain above was deforming down the slope. Until, on 22 December 2018, the mountain became unstuck. This was the earthquake which happened two minutes before the slide. In a normal earthquake, the rocks are stiff and initial movement after the failure of the fault is fast. The energy stored in the deformed rocks creates the force that causes the movement. As the rocks move, the deformation lessens and the force becomes smaller. Finally (within seconds) it is less than the friction force for moving objects. At this time movement stops. Now moving friction is replaced by the (higher) static friction, and the fault is locked in place. It will remain so until the next failure.

On a slope, it is different. The rocks in this particular case are much less strong. (Volcanoes may look impressive, but they are made of pretty weak stuff and are build on sand.) This made the initial movement much less forceful and smoother. The earthquake caused by the ‘unsticking’ was therefore a relatively small one, but longer lasting. But the cone above the now unstuck plane was still subject to the force of gravity, and moving friction was insufficient to keep it in place. The cone started to move, slowly at first. It started in one place, and the moving rock was being held in place (somewhat) by the mass ahead of it that hadn’t failed yet. This moved the excess weight to that part of the stuck plane, which was now also pushed over the edge and failed. Two minutes later, the entire slope was moving and gravity had free reign. In the end, gravity always wins. It just has to wait. And 90 seconds later, half the mountain was lying on the sea floor. This displaced a considerable amount of ocean water, and this displacement triggered the tsunami. The wave reached the shores of Java and Sumatra 45 minutes later, fully unexpected as no one was aware of or had seen what had just happened. It would be days before radar images would show a disbelieving audience that an entire volcano was gone. Again.

The slide produced large blocks, up to 80 meters tall, which now lie in the sea floor. They did not travel far: the debris is within 1.5 km of the shore.

Walter et al. estimate the volume of the collapse as 0.1 to 0.2 km3. Hunt et al, a year later, measured the sea floor debris and derived a volume of 0.175 km3. This is a little less than what was assumed in the Giachetti et al. warning paper from 2012 (0.25 km3, but it was not far off. The collapse event largely happened as had been predicted.

Past and future

It turns out, this was not the first collapse of Anak Krakatau. The mountain had also failed in June 1949. The photo below was taken in 1950, a year after this collapse. As in 2028, half the mountain had been lost. This had been a much smaller event as the mountain was a lot smaller at that time, and it had happened with little fanfare and without a recorded tsunami. It temporarily reduced the height of the mountain. The rebuilding entrained the plane of failure into the new cone: the cone of Anak Krakatau was build on this surface. This may well be the origin of the decollement which failed in 2018.

Source: Hunt et al. 2021

After the 1949 collapse the cone was replaced by a lake, having dropped to below sea level. It took a decade for the lake to be replaced by a cone.

Source: Zen, 1964

There may have been another event as well in the history of Anak Krakatau. There was a small tsunami observed at Rakata, with a height of 1-2 meters, on October 19-20, 1981. This may have to do with another landslide of Anak Krakatau, albeit small compared to 2018 and 1949. 1981 was the year when Anak Krakatau briefly erupted dacite, and when the eruption site moved a bit to the southwest. Wast this related to a small collapse? (The recent surveys of the caldera floor showed blocks that had not come from Anak Krakatau, and probably came from a collapse of the side of Rataka, sometime after 1883. So it is possible this Rakata tsunami was self-induced. But the fact it happened while Anak Krakatau was going through changes points the finger there.)

The collapse of 2018 took the mountain back to its height after the 1949 collapse. It was measured at 126 or 152 meters (both numbers appear in the literature). The crater is much lower, and as after 1949, initially was a lake. The lake did not last long. The cone rebuild itself remarkably quickly. Below is a youtube video from an eruption in July 2023, where the cone is nicely building up. On the most recent images, the cone is already close to the level of the old rim, although it has not yet filled in the entire hole. The eruptions are faster than before and the rebuilding is faster than it was after 1949.

But the seeds of future failure are still there. The collapse has not filled in the crater and the caldera edge remains as steep as it was before. It may take 100 collapses before this has changed and the caldera floor has been raised to safer height. So far there have been two collapses. And Anak Krakatau is rebuilding itself with vigour. The island is now larger than it was before 2018. The first collapse was 20 years after its emergence above sea, or 66 years after its onset on the seafloor. The second collapse was 70 years later. (We could call it the 70-year itch.) What happened twice (or perhaps thrice) will happen again. It is a matter of when, not if. And it may well be in this century.

As I write this, Ruang is having it second set of spasms. This too is a volcano with tsunamic history. In 1871, it caused the second largest known volcano-tsunami in Indonesia, with a run-up height of 25 meters. This is why people are worried about its eruption. Remember the 130 tsunamis from 80 volcanoes? Mathematics tells you that this makes volcanoes repeat-offenders. What it does once, it will do again. And unusually, in the case of Anak Krakatau we know we only have decades to prepare – not centuries.

And this is why scientific papers should never be published behind paywalls.

Albert, April 2024

More about Krakatau

If you would like to know more about Krakatau, there have been many posts here:

The rise and fall of Anak Krakatau

Prelude to Krakatau. I

Prelude to Krakatau. II

Prelude to Krakatau. III

Krakatoa: a blast from the past

Krakatoa skies: when the Sun turned blue


Walter, T.R., Haghshenas Haghighi, M., Schneider, F.M. et al.: Complex hazard cascade culminating in the Anak Krakatau sector collapse. Nat Commun 10, 4339 (2019).

Giachette, T.,Paris, R., Kelfoun K., Ontowirjo, B.: Tsunami hazard related to a flank collapse of Anak Krakatau Volcano, Sunda Strait, Indonesia. Published in Natural Hazards in the Asia–Pacific Region: Recent Advances and Emerging Concepts. Geological Society, London, Special Publications, 361, 79–90 (2012)

Christine Deplus, C., Bonvalot, S., Dahrin, D., et al..: Inner structure of the Krakatau volcanic complex (Indonesia) from gravity and bathymetry data, Journal of Volcanology and Geothermal Research, Volume 64, 1995, Pages 23-52,

Hunt, J.E., Tappin, D.R., Watt, S.F.L. et al. Submarine landslide megablocks show half of Anak Krakatau island failed on December 22nd, 2018. Nat Commun 12, 2827 (2021).

Kyra, S. Cutler, S. Watt, M., Amber L. et al. Downward-propagating eruption following vent unloading implies no direct magmatic trigger for the 2018 lateral collapse of Anak Krakatau,Earth and Planetary Science Letters, Volume 578, 2022, 117332,

Zen, M.T., Hadikusumo, D. Recent changes in the Anak-Krakatau volcano. Bull Volcanol 27, 259–268 (1964).

Mutaqin B., Lavigne F., Hadmoko D., Ngalawani M.,:  

Volcanic Eruption-Induced Tsunami in Indonesia: A Review 2019 IOP Conf. Ser.: Earth Environ. Sci. 256 012023 (2019).

275 thoughts on “The collapse of Anak Krakatau

  1. The active cone is steaming a bit and an area to the west at 6:34am both seem to be a slight increase

    • Today I’d assume that Sunday is the eruption day … three days after the end of the activity on the cone. If the activity there had continued, the eruption day would be around 16th May.

      • I don’t think we support dynamic voting – let me check the VC constitution and ask our fully independent supreme court. If we do, Thomas should probably have first pick on changing dates

  2. Thank you, Albert, this was an informative read and I learned much about the history and physics of past and future collapse events at AK. Regarding the 2012 publication paywall, I doubt that it precluded Indonesian scientists from having knowledge of and access to the article, and I fully expect that the implications were known to various Indonesian agencies years before the 2018 event. The Geological Society is not an obscure organization or journal, and the fourth author works for the Indonesian government. Nevertheless, I appreciate your point that paywalls are an impediment to disseminating and accessing information, and I certainly agree that some scientific information and publications have immediate and important implications for both the professional audience and for laypersons.

    • The transfer from information from scientific domains to political ones is not automatic – nor should it be as the rule is that scientists advice but do not decide. In this case, the paper will have been known to specialist scientists but not to anyone else. Even my university does not subscribe to all Geological Society journals and again the rule is not to cite what you can’t read (for obvious reasons). In fact, the dominant way of this kind of information to get to the politicians is by the media and the media would not have open access. So I think it was a significant barrier. It could have been overcome if the authors had cried wolf and told some media that the end of Indonesia was certain, but these authors were not like that.

      • Yup, “the end of Indonesia was certain” would be the likely media chyron.

Comments are closed.