The rule, and it holds depressingly well across the record: a magnitude 7 or larger earthquake, shallow, close to a populated area, kills people in the thousands. Not because of the shaking, which kills almost nobody directly, but because buildings fall on the people inside them.
The exception arrived on Wednesday morning off the coast of Hualien, and the date to look back at in order to understand it is 21 September 1999.
What happened in 1999
The Chi-Chi earthquake struck central Taiwan at about 1.47 in the morning, magnitude 7.7. It killed around 2,400 people and injured roughly ten thousand, and it brought down something on the order of nine thousand buildings.
The post-event investigations found a pattern that is the signature failure of this kind of event. A great many of the collapses were soft-story failures: buildings whose ground floors had been opened up for shops, parking or lobbies, leaving fewer walls and columns at the level carrying the most load. Under lateral shaking that floor goes, and the structure above it comes down more or less intact onto whatever was in it.
Investigators also found buildings that departed from the codes then in force. That distinction matters, because a code failure and a compliance failure have different remedies and both were present.
What was done about it
This is the part that is worth sitting with, because it is slow, expensive and almost entirely invisible while it is happening.
Taiwan revised its building codes and tightened enforcement. It invested in seismic instrumentation and in an early warning capability. It ran sustained public education about what to do in the first thirty seconds. And, hardest of all, it began retrofitting existing buildings, which is the part every country finds difficult because the money is spent on structures that look fine.
None of that produces a measurable result in any given year. A country that spends two decades on seismic retrofitting and is not struck by an earthquake has spent the money and has nothing to show. The return arrives only in the form of an event that then fails to be as bad as it would have been, which is a counterfactual and therefore never quite believed.
What Wednesday turned that investment into
The 3 April earthquake was magnitude 7.4, shallow, just offshore from Hualien City, the strongest to strike the island in the twenty-five years since Chi-Chi. It struck shortly before eight in the morning, on a weekday, with people in schools, offices and transit.
At least eighteen people died. More than a thousand were injured.
Set that beside 1999 and the comparison has to be made carefully, because the two events are not identical. Chi-Chi was larger, it was inland rather than offshore, and the energy released by a 7.7 is several times that of a 7.4. The rupture was directly beneath a populated region rather than under water nearby. Nobody should present eighteen against 2,400 as a clean measurement of anything.
What can be said is narrower and still remarkable. Post-event assessments found that soft-story failures, the signature collapse mode of 1999, were markedly less common. That is the specific defect that was identified, legislated against and retrofitted, and it is the one that did not repeat.
Why the deaths that did occur are the wrong thing to look at
A discipline about what this shows: the deaths were substantially not building collapses. Rock falls killed hikers in Taroko Gorge, a truck driver at the mouth of a tunnel on the Suhua Highway, and a worker in a quarry, with others cut off in the mountains and airlifted out.
That distribution is itself the finding. Building codes address buildings. They do nothing about a cliff face above a road, and no amount of structural engineering removes the hazard of a mountainous coastline in a subduction zone. A country can solve the part of the problem that is made of concrete and steel and is left with the part that is made of geology.
So the correct reading is not that Taiwan has solved earthquakes. It is that it has substantially solved one failure mode, and the remaining deaths cluster in the modes it has not.
What an early warning system can and cannot do
Taiwan’s seismic network sends alerts to phones in the seconds between a rupture being detected and the shaking arriving, and the warning is worth being precise about because it is routinely oversold.
The physics allows it because the first waves to arrive carry little of the destructive energy and travel faster than the waves that do. Detect the first, calculate, transmit, and a location some distance from the epicenter gets a warning measured in seconds to tens of seconds.
Seconds are enough to stop a train, close a valve, halt a surgical instrument, and get a person under a table. They are not enough to evacuate a building, and evacuating during shaking is worse than staying put. So the warning saves lives in a specific and limited way, and a place very close to the epicenter, which is to say the place that gets hit hardest, gets the least warning of all.
Hualien was close to the epicenter. Whatever the network contributed on Wednesday, it contributed least where it was needed most, which is a permanent property of the technique rather than a shortcoming of this installation.
The part that does not transfer
It is tempting to end on a policy recommendation and it should be resisted, because the conditions that made this possible are unusual.
Taiwan had a catastrophe severe enough to produce sustained public demand, a state capable of enforcing codes against a construction industry, the money to retrofit, and twenty-five years without a change of course. Remove any one of those and the story ends differently. Plenty of countries have had the catastrophe and not the other three.
The honest general claim is smaller than the story invites: seismic building codes work, they are known to work, and the obstacle to applying them has never been the engineering.
What is standing
The photograph that went around the world this week is of an eight story mixed-use tower in Hualien City leaning at about sixty degrees, its lower floors gone, which is a soft-story failure and is exactly the thing that was supposed to have been designed out. One woman died in it. Twenty-three people were pulled out alive.
It is one building. Around it, in a city that took the full force of a magnitude 7.4 at the start of a working day, several thousand others are standing with people inside them, and there is no photograph of that, because a building that did what it was built to do does not look like anything at all.
