On the shores of the Indian Ocean in 2004, the water went out and people followed it. Holidaymakers who had come for the warm sea walked across ground that had been seabed a moment before and gathered shells and small fish. Mikhail Nosov, an oceanologist who studies how tsunamis are born, tells it with a dry edge. After that point, he says, running is pointless, because you will not make it.
The sea had not gone away. The retreat was the first half of the wave. The same night killed about 230,000 people around the Indian Ocean. Japan's tsunami in 2011 killed something under 19,000, more than ten times fewer. Nosov credits a warning system that worked and a population that knew what to do. Knowing what a retreating sea means is part of that.
So why does the water leave? Sometimes a tsunami arrives with a rise first, and sometimes with a drop. Researchers tried to explain the difference in many ways, especially in the decades when there were few computer models and they had to work it out on paper. Nosov's answer is plain, and he is practically sure of it.
Picture a subduction zone, where one plate dives beneath another and the sea floor lies over the seam. When it ruptures, the seabed does not move as one slab. The strip closer to land drops. The strip farther out toward the ocean is thrown up.
In subduction zones, tsunami sources are, as a rule, arranged so that the subsidence zone lies closer to the shore and the uplift zone lies farther out to sea.
В зонах субдукции очаги цунами устроены, как правило, именно таким образом, что ближе к берегу находится зона опускания, а мористия находится зона поднятия.
— Mikhail Nosov, Osnova, 38:22
Now follow the water. Over the sunken strip the sea has somewhere to fall, so it falls, and the surface there sags. That sag travels to the coast as a low wave, and the beach seems to widen. Over the raised strip the sea is shoved upward, and the push is usually the stronger one. That hump runs in behind the first. Low wave, then high wave, in that order, because the sunken ground lies nearer the shore.
That also settles the oddity of a wave arriving in two halves. A big source still lies offshore. Nosov puts a number on how much water a big one can move: the 2011 Tohoku earthquake displaced about 100 cubic kilometres.
Records from other earthquakes show the pattern. A study in Frontiers in Earth Science in April 2026 went back to tide gauges from the 1945 Makran subduction earthquake and found the sea level dropping suddenly before the waves arrived. The same work puts about a metre of uplift at the town of Ormara and a metre of subsidence at Pasni. One fault raised one stretch of coast and lowered another.
Nosov is careful about how far he takes it. Landslides might add something, he says, and so might nonlinear effects in the water. You can invent many such ideas, and he calls them fantasies. Then he comes back to what he has seen. Whenever he looks at the sources of subduction earthquakes that produced tsunamis, they look like this, and his own experience tells him this is the main story.
The next time you see footage of a beach emptying, you can read it. The earthquake has already pushed the seabed out at sea upward. The empty sand is the trough in front of that water, and the sea is not leaving. It has dropped into a hollow the ground made, and the rest of it is on its way back.
