At 4:11 in the morning on 4 September 2026, a seismograph on the Sunda Strait recorded a fourteen second event with a maximum amplitude of 33 millimeters. Small. The kind of reading that goes into a log and stays there. Three more followed before six that morning, each one longer and stronger than the last, and then the volcano went quiet for most of the day.
It started again at 10:03 that night and didn’t stop.
By half past midnight the ash column had reached fifteen kilometers, roughly fifty thousand feet, which is above the cruising altitude of every airliner in the region. The Darwin Volcanic Ash Advisory Center tracked the cloud moving west at about 74 kilometers an hour with a second layer at five and a half kilometers spreading south.
Soekarno-Hatta International, the second busiest airport in Southeast Asia, shut down. So did Halim, and Lampung, and eventually operations at Bali and Medan. The closures hit about 2,300 flights and 268,539 passengers across the four airports InJourney runs. Jakarta moved every level of school to remote learning. Fishing stopped.
The volcano doing this is called Anak Krakatau, child of Krakatoa, and it’s 157 meters tall. That’s shorter than the Washington Monument. It’s a stub of rock sitting in the flooded hole where a mountain used to be, and understanding why a stub can close an airport requires going back to what happened to the mountain.
What happened at Krakatoa in 1883?
The island of Krakatoa sat in the Sunda Strait, the channel between Java and Sumatra where the Indian-Australian plate grinds under the Eurasian one. Three cones had grown together to make one island: Rakata, Danan, and Perbuwatan. The whole structure stood on the rim of a much older caldera, formed when a previous volcano collapsed there, possibly in the sixth century.
Activity began on 20 May 1883. A German warship passing through reported a cloud of ash and dust seven miles high. For the next three months, ships and villagers on Java and Sumatra watched explosions that were loud but not alarming. In June a black cloud sat over the islands for five days. Pumice began turning up in the water. Local people treated the whole thing as a spectacle.
On the afternoon of 26 August the northern portion of the island destroyed itself. Four more explosions beginning at 5:30 the next morning finished the job. The largest, around ten in the morning on 27 August, is the loudest sound in recorded history. It was heard 3,000 miles away in Perth and near Rodrigues Island in the Indian Ocean. The pressure wave circled the planet seven times and was still detectable on barographs days later.
Somewhere between 18 and 25 cubic kilometers of rock went into the air, ash reached fifty miles up, and two thirds of the island collapsed into the emptied magma chamber. The official death toll is 36,417. Most of those people didn’t die from the explosion or the ash. They drowned.
The collapse pushed a wall of water up to 37 meters high across the strait, and it erased 165 coastal villages. The water carried a steamship called the Berouw nearly a mile inland on Sumatra and left it there with its entire crew dead. Pyroclastic surges crossed forty kilometers of open water and came ashore on Sumatra still hot enough to kill.
How does an eruption on one island cool the whole planet?
The number people quote is the rock volume, but the rock isn’t what changed the climate. Sulfur is.
An eruption that reaches the stratosphere injects sulfur dioxide above the weather. Below that altitude, rain scrubs the atmosphere clean in a week or two. Above it there’s no rain, so the sulfur dioxide reacts with water vapor to form a haze of sulfuric acid droplets that spreads around the planet and stays for years. Those droplets are small, bright, and very good at reflecting sunlight back to space before it reaches the ground.
Krakatoa put roughly twenty million tons of sulfur up there. Global average temperature dropped by something in the range of half a degree Celsius, with estimates running higher, and stayed down for around five years. Northern hemisphere summers cooled measurably. Sea levels fell, because the ocean contracted as it cooled. The same aerosol layer scattered light at sunset into reds and oranges so intense that fire brigades in New York were called out to nonexistent fires, and painters across Europe spent the next two years recording skies nobody had seen before.
The mechanism is what makes volcanoes a climate story as well as a geology story. It’s the same process that runs through the far larger event I wrote about in the Toba supereruption, where the sulfur load was large enough to put the entire species under pressure. Krakatoa is that mechanism at a scale small enough to have been documented by telegraph and newspaper instead of reconstructed from ice cores.
How do we know so much about an eruption from 1883?
Because a Dutch geologist named Rogier Verbeek was living in Java when it happened and went to work immediately.
Verbeek had already surveyed the region’s geology. After the eruption he traveled the affected coasts, collected eyewitness accounts, measured deposits, and observed the destruction himself. His report ran to 550 pages, and the government of the Dutch East Indies published it in 1885. It’s the reason the timeline is as precise as it is, and it’s one of the founding documents of modern volcanology.
The 1883 event also happened to be the first global disaster of the telegraph age. Barometers in Europe recorded the pressure wave. Tide gauges as far away as the English Channel picked up the water disturbance. Ships’ logs across half the world recorded ashfall, floating pumice, and the sunsets. The eruption produced a dataset by accident, and it’s a decent lesson in why contemporary observation beats reconstruction. Somebody wrote it down while it was happening.
I spent five decades on an amateur passion for geology before I ever wrote a book that needed it, and the thing that keeps pulling me back is exactly this. The rock record is a set of inferences. The 1883 record is a set of observations. When you can put the two against each other, you find out how good the inferences are.
What is Anak Krakatau and where did it come from?
The 1883 collapse left a caldera about 250 meters deep, with only a remnant of Rakata standing above water. Danan and Perbuwatan were gone.
The vent didn’t close. In 1927 a new cone broke the surface inside the caldera and was promptly destroyed by the sea. It came back. It kept coming back, and by the 2010s Anak Krakatau stood 338 meters above the water and was growing at a rate people could measure year to year. It became a tourist destination. Boats ran day trips from the Java coast so visitors could climb a live volcano.
It’s a basaltic-andesitic cone, which matters. The 1883 magma was silica-rich and viscous, the kind that traps gas until the whole system fails at once. Anak Krakatau erupts a gassier, runnier magma in near-continuous Strombolian bursts, throwing incandescent rock in fountains instead of accumulating pressure for years. That’s a real difference in behavior, not a reassurance. A cone that erupts constantly also builds fast, and a cone that builds fast on the unstable rim of a submarine caldera builds something that can fall over.
What happened in December 2018?
On the evening of 22 December 2018, at about 20:55 local time, the southwestern flank of Anak Krakatau slid into the sea.
The landslide was small by geological standards, under 0.2 cubic kilometers, roughly 150 to 180 million cubic meters of rock. It removed about two thirds of the island’s above-water volume and cut the cone from 338 meters to about 110. The material dropped into the deep caldera floor southwest of the volcano and displaced an enormous quantity of water in a few seconds.
Waves reached 85 meters close in and 13 meters on the coasts of Java and Sumatra. At least 437 people died, more than 14,000 were injured, and around 33,000 were displaced. It was the deadliest volcanic tsunami since 1883.
The part that should stay with anybody thinking about warning systems is this. Indonesia has a tsunami warning network. It’s built to detect earthquakes. A landslide doesn’t produce the sharp, short-period seismic waves that trigger it, so the network stayed silent. Later analysis showed the collapse was detectable in long-period seismic data within about eight minutes, which is enough time to be useful, but nobody was reading for it. The wave arrived on a Saturday night at a coast full of holiday visitors, with no siren and no phone alert.
Researchers had modeled a flank collapse at Anak Krakatau and published the tsunami scenario years earlier. The hazard was known. It wasn’t systematically monitored.
What is happening at Anak Krakatau right now?
The current sequence didn’t start in September. Indonesia’s Geological Agency raised the alert from Level II to Level III, the second highest on a four level scale, on 2 July 2026, after sulfur dioxide emissions and satellite thermal anomalies appeared through June and crater hotspots showed up on the tenth. A three kilometer exclusion zone went in around the crater and has stayed there.
Activity built through August. In the monitoring period from 22 to 31 August the agency recorded eruption columns of 50 to 400 meters above the crater, sulfur dioxide output between 715 and 2,922 tonnes a day, 174 eruption earthquakes and 714 emission events. The evaluation described shallow magma dynamics as very high.
Then came 4 September. Continuous eruption, described by the Geological Agency as resembling a lava fountain capable of producing lava flows, with the plume glowing bright red and rumbling audible at the Pasauran observation post across the water. People reported booming and vibration across West Java, Banten, and Lampung. Satellite instruments measured about 0.2 teragrams of sulfur dioxide injected into the upper troposphere within a day.
The nearest settlement is more than sixteen kilometers away, so nobody lives in the danger zone. The damage is being done by ash at distance: aviation, respiratory health, schools, and fishing. Indonesia’s disaster agency has gone as far as planning cloud seeding to bring rain down and wash ash out of the air.
Will Anak Krakatau cause another tsunami?
This is the question everybody asks, and the answer has three parts.
First, what the authorities are saying. The Geological Agency states that the volcano’s growth since the 2018 collapse remains limited and there’s currently no potential for a collapse large enough to trigger a tsunami. That’s a statement about the present shape of the cone. It lost two thirds of its volume seven years ago and hasn’t rebuilt the mass that fell over.
Second, what they’re careful not to say. Lana Saria, who heads the agency, put it plainly in a briefing: the repeated eruptions indicate a continued supply of magma and gas beneath the volcano, and officials cannot predict whether a larger eruption will occur. She also noted they evaluate the volcano on all monitoring data, not on the number of eruptions or the height of the ash column. That’s a scientist declining to be cornered into a forecast, and it’s the correct answer.
Third, the thing 2018 taught. The collapse that killed 437 people came from a landslide of less than a fifth of a cubic kilometer. That’s small. Small enough that a cone which has been rebuilding for seven years could produce something comparable without first becoming the mountain it was in 2018. The 2018 study that modeled it concluded that these lateral collapses can happen without precursory signals and are an efficient and unpredictable tsunami source.
So: no immediate threat on current data, a monitored volcano, a documented mechanism, and a coastline with more than twenty million people living within a hundred kilometers. All four of those are true at once, and any account that gives you one of them is selling something.
What Krakatoa teaches about writing disaster
I’ve written about disaster from inside it. When Hurricane Irma came through, and again for Milton, the thing I kept noticing was the gap between the aggregate and the experience. The aggregate says a Category 5 with a certain wind field. The experience is a specific noise at three in the morning and a decision about whether the bathroom is better than the hallway.
Krakatoa has the same gap and it’s wider, because the numbers are so large that they stop meaning anything. Thirty six thousand people isn’t a quantity a reader can feel. One steamship a mile inland with its crew still aboard is.
The craft lesson generalizes past volcanoes. When you write catastrophe, the scale figures are the setup and the single observed detail is the payoff. Verbeek understood this in 1885 without anybody explaining it to him. His report is full of measurements, and the reason people still read it’s the eyewitness accounts he collected: what the sky looked like, what the sound did to people, the man who watched the water leave the harbor before it came back.
That’s also why Family on Ice follows one family through a volcanic winter instead of narrating a species-level catastrophe. A reader can hold a family. Nobody can hold a population estimate.
What this should mean to anyone reading it now
Anak Krakatau isn’t going to do what its parent did. The magma is different, the edifice is a fraction of the size, and 1883 was the failure of a system that had been loading for two centuries. What Anak Krakatau can do is close airports, put ash in the lungs of a city of ten million, and drop a piece of itself into the water without warning.
None of that is a reason for panic and all of it is a reason for the unglamorous work. Indonesia’s warning network couldn’t see the 2018 landslide because it was designed for a different kind of event. That’s the most transferable lesson here and it’s nothing to do with volcanoes. The failure mode is almost never the thing you prepared for. It’s the thing next to it, arriving through a channel your instruments don’t watch.
I wrote a whole book on that idea, which is Real World Survival, and the argument is the same at every scale: prepare for the shape of trouble instead of the specific trouble, because the specific one will surprise you. If you want the broader collection, the Prepared, Not Paranoid hub gathers it, and the Disaster Recovery Hub covers the same reasoning applied to systems instead of households.
The volcano is still going as I write this. The seismographs are still recording. Somebody at the Pasauran post is listening to a mountain across the water, the way Verbeek’s eyewitnesses did, and writing down what they hear.