The harbor empties before the water arrives.
Boats out, gates shut, crates left where somebody stopped working.
Then the sea pulls back through the mouth in a smooth brown sheet, lifting mud off the bottom, and the tide line on the wall sits dripping and dark a foot above the surface.
Nobody on the quay is reading an instrument. They are acting on a sentence that arrived on a phone.
That sentence was written and sent while the ground was still moving.
The first warning does not come from the sea
This is the part that surprises people who assume seafloor sensors do it.
They do not, at least not first.
The agency locates the earthquake and estimates its size from seismometers on land, in seconds, and then does something that looks almost like cheating. It looks the answer up.
Roughly a hundred thousand tsunami scenarios sit simulated and stored, matched to a coast chopped into 66 forecast blocks.
Find the epicenter, find the magnitude, pull the nearest precomputed case, publish the height band for each block.
No wave has to be observed for that to work, which is the whole design, because a wave close to shore arrives with no useful lead and one far out takes too long to reach a sensor.
The port the water reached first
Kuji sits on the northern Sanriku coast, a town of about 34,000 around a fishing and oil storage harbor.
Its defenses are unfinished, and that is not a scandal. It is scale.
The bay mouth breakwater is designed to run 2.4 miles in two arms, and after three decades it is still short.
The seawall behind it stands 26 feet above sea level, built to the size of the 1933 tsunami rather than the larger one of 1896.
In 2011 the water here reached 89 feet locally and pushed two and a half miles inland, destroying 444 houses.
Against that, a gauge reading in single digits of feet is a quiet day, which is exactly what the system is hardest to judge.
What the instruments recorded
The rupture happened at 16:52 and 59 seconds, about 12 miles down, on the plate boundary roughly 62 miles east of Miyako.
Magnitude is where the early reporting went wrong. The preliminary 7.4 was revised to 7.7 on the agency’s own scale, while the moment magnitude computed elsewhere stayed at 7.4, because those are two different measures rather than a correction.
The warning went out at 16:55, about two minutes later, for Iwate and the central Hokkaido Pacific coast. Aomori was upgraded at 17:08. Everything was downgraded at 20:15 and cancelled at 23:45.
The refined gauge figures, published three weeks later, came in below the first night’s.
Kuji peaked at a little over two and a half feet, or about 31 inches. Miyako reached 13 inches, Hachinohe under 11, Sendai under 10.
Ten people were hurt, two seriously.
Where the 41 minutes belongs
The figure gets attached to the wrong event.
Forty one minutes is when the water at Kuji peaked. The first wave got there at 17:26, which is 33 minutes, and the first wave to reach any gauge anywhere hit Miyako at 17:10, seventeen minutes after the rupture.
That distinction is the case for the precomputed approach. Seventeen minutes is all the lead time the nearest town ever had.
Two more things get overstated. The seafloor cable network is 150 stations across six segments and about 3,420 miles of cable, not hundreds, and the agency says plainly that the first warning still comes from the land network while the seafloor data feeds a separate confirming bulletin.
And the deep water signal was one buoy, 715 miles away, reading under a third of an inch, from which an American research model of the source was inverted, as one deep water buoy record shows. Japan’s own source model came from onshore satellite positioning instead.
Infrastructure that is only tested by the rare case tends to carry gaps like that, whether it is a tsunami warning chain, the steel nobody had drawn on a tunnel plan, or the labor behind 36 ports.
Why the forecast was allowed to be too high
The announced band tops out just under 10 feet and starts a little over three, and the observed maximum was under three.
That gap is deliberate. There are five bands, and a warning names the upper bound of its height class rather than a best guess.
The reason is 2011. The first bulletin then was based on magnitude 7.9 and predicted 20 feet for Miyagi against a final 9.0 and a runup measured at 131 feet.
The fix adopted afterward was to stop publishing a number at all when the magnitude is uncertain and large, and to say huge or high instead until the real size is known.
So an over prediction here is the system working as rebuilt.
The cost of that choice is the thing worth watching. Every band that lands high spends a little of the credibility the next one has to borrow.
Read the whole thing?
Get the week's signal, not the noise
Our sharpest reporting on energy, climate and nature — free, once a week.
