The Earthquake Made a Waterfall — Then the River Moved It 770 Meters Upstream

Earthquake shaking lasts minutes. A river can spend decades responding to it. On the Hakka River in Wajima’s Monzen district, the 2024 Noto Peninsula earthquake abruptly lifted the coast, creating a steep reach near the river mouth that behaved like a waterfall. Flood by flood, that break in slope began retreating inland.

A research team led by Takayuki Takahashi of Tohoku University’s International Research Institute of Disaster Science monitored roughly the lowest kilometer of the Hakka River using UAV lidar, drone photogrammetry, post-earthquake synthetic-aperture radar imagery and Geospatial Information Authority of Japan aerial data. Their paper was published September 14 in Communications Earth & Environment. ^{[1] [2]}

Within the first year, they documented as much as 2 meters of vertical riverbed incision and 770 meters of upstream migration by the main knickpoint — the sharp break in a river’s longitudinal profile that can appear as a rapid or waterfall. The university says this is the first high-resolution observation of post-earthquake river-terrace formation on this timescale. ^{[1] [2]}

From the river’s point of view, the sea suddenly fell about 3.5 meters

River profiles are controlled by a base level, usually the elevation of the water body into which the river drains. If the land rises while the ocean stays where it is, the river experiences the same relative problem as if sea level had suddenly dropped.

At the Hakka River mouth, the earthquake produced a relative base-level fall of about 3.5 meters, according to the study. Seafloor that had been below water was uplifted and exposed, extending the river toward a new shoreline. The old, gentler channel profile was suddenly too high above the new outlet, creating a steep transition zone. ^[1]

The scale of the tectonic change is independently well established. GSI’s emergency survey measured as much as 4.10 meters of uplift in western Noto. Near Kaiso in Monzen, Geological Survey of Japan field teams measured uplift of 3.8 to 3.9 meters using the former upper limit of attached marine organisms on harbor structures as a shoreline marker. ^{[3] [7]}

The first sharp break appeared within hours

The team distinguished several knickpoints rather than treating the river as one moving waterfall. One feature, labeled KP-1, was already visible in a SAR image acquired at 11:10 p.m. on January 1 — less than seven hours after the main earthquake. The study suggests tsunami run-up may also have influenced the formation and position of this feature. ^{[1] [9]}

The major migrating feature was KP-2. Assuming it began near the pre-earthquake mouth, it had retreated roughly 350 meters upstream by early September 2024. Then the extraordinary Oku-Noto rainfall of September 20–22 pushed it another roughly 420 meters. Total retreat reached 770 meters. ^[1]

The movement was not smooth. The researchers calculate typical mean retreat rates of roughly 1 to 3 meters per day over the first year, with interval rates reaching about 8.4 meters per day. Movement accelerated when river stage rose and slowed or stopped between floods. ^[1]

The earthquake created the step; floods did much of the cutting

Heavy rain struck Noto on September 20–22, 2024. The Japan Meteorological Agency issued an emergency heavy-rain warning for parts of the peninsula on September 21 and later documented the event as a major disaster-producing weather episode. ^[6]

In the Hakka River, the paper reports a peak discharge around 100 cubic meters per second on September 21. That flood rapidly extended incision upstream, removed terrace material near bridges and widened portions of the lower channel to more than 60 meters. ^[1]

The sequence matters. The earthquake and the flood were not merely two separate disasters that happened in the same year. The earthquake changed the river’s geometry; the later flood supplied the hydraulic energy that exploited that new geometry.

An artificial weir stopped the retreat — for a time

During the September flood, KP-2 reached an artificial weir roughly one kilometer upstream of the post-earthquake mouth. The structure temporarily arrested further upstream migration. The researchers argue that without it, the knickpoint would likely have moved farther during that event. ^[1]

But stopping a knickpoint is not the same as ending the adjustment. Heavy rain in August 2025 scoured beside the weir, while another knickpoint, KP-3, migrated farther upstream. The paper includes terrain observations through October 29, 2025 and warns that if artificial controls are undermined or bypassed, incision can resume. ^[1]

That has direct engineering significance. Bridges, revetments and weirs are fixed structures embedded in a channel whose bed elevation is not fixed. Lower the surrounding bed by one or two meters and foundations can become exposed, bank protection can be undermined and water can attack structures from a different elevation or direction. The paper reports that bank-structure failures and scouring around bridge piers have already been observed on the Hakka River. ^[1]

The river begins to armor itself

A river does not necessarily keep cutting downward at the same rate. As flowing water preferentially removes finer sediment, larger gravel can be left behind on the bed. Geomorphologists call this armoring.

That coarser surface acts like protective skin over more erodible sediment below. The Hakka River team argues that armoring has begun to inhibit further vertical incision in parts of the channel. Once that happens, erosive energy can be redirected toward the banks, helping explain why the post-knickpoint channel became wider and more laterally active. ^[1]

This is one reason a simple extrapolation would be wrong. A feature that retreated 770 meters in year one is not expected to repeat that exact distance every year. Channel slope, sediment strength, artificial structures, discharge and the evolving armored bed all change as the feature moves inland.

A river terrace can be born in a year

River terraces are often read as records of long geological history: older floodplains left stranded above a river after the channel cuts downward. Many classic terraces preserve thousands or tens of thousands of years of tectonic and climatic change.

The Hakka River compresses that process into human time. As the channel incised after the earthquake, pieces of the former floodplain were abandoned above the new bed. The researchers say they were able to document the development of a new fluvial terrace within a year. ^{[1] [2]}

Along the nearby coast, another kind of terrace tells a much older story. Geological Survey of Japan researchers had already identified three low marine terraces thought to have formed since roughly 6,000 years ago, implying repeated episodes of coastal uplift. They describe the 2024 uplift around Kaiso as effectively creating a fourth step. ^{[7] [8]}

The same earthquake therefore made a new step in two landscapes at once: a marine terrace at the coast and a fluvial terrace along the river.

How far upstream can the adjustment travel?

The paper’s before-and-after profiles suggest that more than one meter of uplift affected roughly the lower 12 kilometers of the Hakka River. If the river ultimately works back toward a graded profile similar in shape to the pre-earthquake one, incision may have to propagate more than 10 kilometers farther upstream from the area already observed. ^[1]

The researchers estimate that even if the first-year retreat rate were sustained, adjustment would take more than a decade. In reality it may take longer because retreat generally slows upstream and because multiple artificial structures interrupt the channel. ^[1]

Heavy-rain frequency is another controlling variable. Much of the first-year knickpoint movement happened during high-flow events. That means the future rate of geomorphic change will depend not only on tectonics and riverbed materials, but also on how often intense rainfall returns.

Reconstruction is happening on a river that is still changing

The science intersects directly with reconstruction policy. Ishikawa Prefecture’s emergency flood-control program for Oku-Noto includes the Hakka River among nine rivers where levees, revetments and related functions are being strengthened alongside disaster restoration. The Hakka basin plan also calls for channel excavation, vegetation removal, dredging, driftwood removal and other flood-control work. ^{[10] [12]}

The prefecture has also revised flood-inundation maps for the Hakka River and other rivers on Noto’s outer coast specifically because earthquake-induced terrain change altered the basis on which flood risk had previously been calculated. ^[11]

That is where a geomorphology paper becomes an engineering problem. A design based on a single post-earthquake survey may become outdated if the bed continues to lower or widen after the next major rainfall. “Restoration” cannot mean simply rebuilding to the pre-earthquake geometry when the river itself is no longer trying to return to that geometry.

The earthquake ended. The geomorphic event did not.

The magnitude-7.6 Noto Peninsula earthquake struck at 4:10 p.m. on January 1, 2024 at a depth of 16 kilometers and produced maximum seismic intensity 7. GSI found uplift approaching four meters in western Noto. ^{[4] [5]}

Those numbers were known quickly. What could not be known on January 2 was what a river would do with the new elevation difference.

The Hakka River provides the sequence now: the ground rises; a break in slope forms; rain arrives; the break retreats inland; the channel cuts down; former floodplain becomes terrace; coarse gravel begins to armor the bed; erosion shifts sideways; a weir arrests the process; later rain starts finding a way around it.

The earthquake was an instant. The river’s response is a continuing event.