Theme
Language
The 1920 Haiyuan Earthquake: When the Loess Hills Collapsed
ENJAES

On the evening of 16 December 1920, at about 19:05 local time (12:05 UTC), a major earthquake struck Haiyuan County, in what was then Gansu Province and is now the Ningxia Hui Autonomous Region of north-central China. USGS, in its Today in Earthquake History summary, lists it as M7.8 on the Gansu–Ningxia border, with about 200,000 deaths — one of the world’s most destructive earthquakes on record. Later Chinese estimates put the toll higher, at around 273,000; the figures in this article follow USGS where sources disagree, and hedge where they do.

The shaking reached intensity XII — the maximum on the Modified Mercalli scale — in the Lijunbu–Haiyuan–Ganyanchi area, where USGS describes total destruction. That single number explains much of what follows. Intensity XII is not “very strong” in the ordinary sense; it is the top of a scale built from observed effects, the point at which the built and natural landscape itself fails.

What Happened

USGS counts more than 73,000 people killed in Haiyuan County alone, and more than 30,000 in Guyuan County. Nearly all the houses collapsed in the cities of Longde and Huining, in the USGS summary, and damage rated VI to X occurred across seven provinces and regions, including the major cities of Lanzhou, Taiyuan, Xi’an, Xining and Yinchuan. The quake was felt from the Yellow Sea to Qinghai Province, and from Inner Mongolia south to central Sichuan — a felt area that covers much of northern and central China.

The ground itself moved in ways people could still read years later. About 200 km (125 miles) of surface faulting was seen, USGS notes, from Lijunbu through Ganyanchi to Jingtai. Large numbers of landslides and ground cracks spread through the epicentral area. Some rivers were dammed by sliding ground, others changed course. Far away, the energy was still measurable in water: seiches — standing waves sloshing in enclosed water — were observed in two lakes and three fjords in western Norway.

Why the Loess Made It Worse

Haiyuan sits in loess country. Loess is wind-blown silt, deposited over long ages into thick, soft hillsides and terraces. Dry, it can stand in near-vertical cliffs and be carved into dwellings; shaken hard, it can lose its strength and flow. That combination — homes carved into, or built from, a material that collapses under strong shaking — is central to why a quake here killed on such a scale, even in a region that was not densely urban.

USGS gives one stark example: a landslide buried the village of Sujiahe in Xiji County. It was not an isolated slide. Across the epicentral area, slopes failed, cave dwellings in loess cliffs collapsed, and adobe and earth-built houses disintegrated. The lesson travels far beyond Ningxia: in earthquake country, the ground under a building is part of the building’s safety, and a secondary hazard — here, collapsing loess and landslides — can rival the shaking itself. For how ground failure works elsewhere, see Liquefaction, a different mechanism with a similar moral: soft ground amplifies disaster.

A Winter Disaster, in a Divided Country

The timing compounded everything. The quake struck just after dark in mid-December, at the start of a hard winter. Survivors faced sub-zero nights without houses, in a remote region whose roads were cut by landslides and ground cracks. China in 1920 was in its warlord period, politically fragmented, and organised relief was slow to reach a distant, devastated county. Many deaths counted in the larger later estimates are attributed not only to the shaking and slides themselves, but to exposure and hardship in the weeks that followed — one reason the toll is so hard to pin to a single number, and why this article keeps the USGS figure in front and the higher later estimate clearly labelled as an estimate.

Aftershocks added to the misery and the uncertainty. Contemporary and later accounts describe aftershocks continuing for years, a pattern that is plausible after a rupture of this size and length, and that kept survivors from trusting damaged structures long after the mainshock. Whether any single aftershock count is exact matters less than the lived fact: for people in Haiyuan, the earthquake did not end on 16 December.

The Fault Behind It

The earthquake is associated with the Haiyuan Fault, a major left-lateral strike-slip fault along the northeastern edge of the Tibetan Plateau. The plateau is being squeezed and pushed outward by the continuing collision of India with Asia; faults like Haiyuan take up part of that motion horizontally, sliding past rather than thrusting up. When a long, locked segment of such a fault breaks at once — here, on the order of 200 km of surface rupture in the USGS summary — the result is a great continental earthquake, far from any ocean trench. For the basic mechanics, see Fault Types Explained and Tectonic Plates Explained.

That setting is also why Haiyuan still matters to seismologists. Continental strike-slip faults run through populated regions on several continents, often with long quiet intervals between large ruptures. Haiyuan is a case study in what the end of such an interval can look like when vulnerable housing sits on unstable ground: the magnitude starts the disaster, but the landscape and the buildings decide its size.

Reading the Numbers Honestly

Older great earthquakes rarely carry one clean set of figures, and Haiyuan is no exception. You will see magnitudes from 7.8 to 8.5 in different sources, depending on the scale and the era of the estimate. USGS’s history summary uses 7.8; its catalogue tradition has also been described elsewhere as 8.3, and Chinese sources have used 8.5. This article uses the USGS history figure — M7.8, about 200,000 deaths — as its anchor, and notes the later Chinese estimate of around 273,000 deaths as exactly that: a later estimate, not a settled count. The intensity, by contrast, is not in dispute in the USGS account: XII, the maximum, with total destruction in the worst-hit area.

That honesty is not pedantry. Haiyuan’s real claim on attention does not need the largest number in any range. Even on the most conservative USGS figures, it is among the deadliest earthquakes ever recorded, in a place most readers have never heard of — which is itself part of the story. Remote, rural, and politically overlooked disasters kill at scale too, and they are the easiest for the rest of the world to forget.

The Haiyuan checklist: 16 December 1920, about 19:05 local time · M7.8 (USGS history summary) · intensity XII, the maximum, in the Lijunbu–Haiyuan–Ganyanchi area · about 200,000 deaths (USGS); later Chinese estimates around 273,000 · about 200 km of surface faulting · seiches in Norway. Magnitude started it; loess, housing, and winter decided it.

The Bottom Line

The 1920 Haiyuan earthquake is the disaster that happens when three things line up: a great continental fault rupture, ground that turns into a hazard of its own, and homes with no resistance to either. USGS counts more than 73,000 dead in Haiyuan County alone, in a region of farms and small cities — proof that an earthquake does not need a megacity to become one of history’s deadliest. A century on, the practical lesson is unchanged and general: know your ground, respect secondary hazards like landslides, and build for the shaking your fault can actually deliver. Our guides to What to Do During an Earthquake and What to Do Before an Earthquake cover the personal side of that lesson, wherever you live.

← Ancash 1970 Liquefaction →