Earthquake Landslides: Causes, Consequences, and Prevention

Earthquake-triggered landslides are among the deadliest secondary hazards associated with seismic events, sometimes killing more people than the ground shaking itself. The 2008 Wenchuan earthquake in China, for instance, set off roughly 200,000 landslides that swept about 20,000 people to their deaths. These slope failures are driven by a combination of seismic force, soil and rock properties, water content, and terrain shape, and their consequences can cascade far beyond the initial slide zone, damming rivers, burying roads, and altering ecosystems for decades.

What Makes a Slope Collapse During an Earthquake

When seismic waves pass through a hillside, they apply rapidly alternating forces to the soil and rock. Whether that slope actually fails depends on a few interacting factors. In loose, sandy, or silty soils, vibration can cause water pressure between soil grains to spike. That elevated pore-water pressure reduces friction between particles, and once friction drops far enough, the slope can liquefy and run out as a flow. Research on loess (a wind-deposited silt common across central China) has shown that rising pore-water pressure is the key driver that transforms gradual shear movement into a sudden, runaway flowslide.¹

But not every soil type behaves that way. In clay-rich soils with moderate to high plasticity, the mechanism is different. Because clay particles bind water more tightly, pore-water pressure may not rise much during shaking. Instead, the slope fails through accumulated shear displacement: the cyclic loading pushes the soil mass a little with each wave, and if those small displacements add up past a critical threshold, the slope gives way even without a big pressure spike.² This distinction matters for hazard planning: a clay hillside and a sandy hillside near the same fault can fail through entirely different processes, at different shaking intensities, and with very different runout behaviors.

The intensity of shaking itself is not uniform across a slope. Shaking table experiments have found that peak ground acceleration is amplified with elevation, meaning ridgetops and upper slopes experience stronger forces than the base. When seismic loading exceeds a critical range (roughly between 0.2 and 0.5 g), though, that amplification weakens as the slope begins filtering out higher-frequency vibrations.³ In practical terms, moderate earthquakes can cause disproportionately intense shaking at the crest of a slope, which is one reason landslide scars so often begin near the top of a hill.

Where Landslides Cluster and Why

Earthquake-triggered landslides are not randomly scattered across the landscape. Their distribution is shaped by topography, rock type, and the geometry of the fault that ruptured. Research comparing landslide inventories from several large earthquakes has shown that landslide density is controlled by the combined effect of seismic intensity and the local terrain’s inherent vulnerability to failure, principally its steepness and the type of rock or soil present.⁴

Fault style also plays a significant role. A study comparing seven landslide databases from earthquakes ranging from magnitude 5.6 to 7.9 found that dip-slip earthquakes (where the fault moves primarily up or down) triggered landslides at greater distances from the fault than oblique-slip events. Surface-rupturing earthquakes like Wenchuan, Chi-Chi, and Kashmir actually triggered landslides at shorter distances than expected for their magnitudes, likely because more of the seismic energy was released at the surface rather than radiating outward.⁵ In the Wenchuan event specifically, sections of the fault that moved primarily as a low-angle thrust produced far more landslides than steeper sections dominated by strike-slip motion.⁶

Topographic irregularities add another layer of complexity. Ridges, narrow valleys, and abrupt changes in slope can amplify seismic waves in ways that standard ground-motion models struggle to capture. Three-dimensional seismic simulations have confirmed that these amplification effects can trigger landslides in locations that predictive models would otherwise flag as stable.⁷ This is one of the reasons post-earthquake landslide maps so often turn up surprises.

The Legacy Effect

A common misconception is that once the shaking stops, the landslide risk returns to normal. It does not. Strong earthquakes weaken hillslopes in ways that persist for years, a phenomenon researchers call the earthquake legacy effect. Analysis of ground shaking and rock strength suggests that near-surface rock can lose as much as 75% of its strength when exposed to shaking above 0.9 g, and even moderate shaking can reduce strength by 60 to 75%.⁸ That weakened rock is then more vulnerable to subsequent triggers like heavy rain, aftershocks, or even normal seasonal processes.

Satellite-based monitoring using radar interferometry has tracked this effect in the field. In one study area, average hillslope deformation rates were still elevated above pre-earthquake levels roughly four and a half years after the main shock.⁹ For communities rebuilding after an earthquake, the legacy effect is a critical consideration: a slope that survived the main event may still fail during a monsoon season or moderate aftershock years later.

Cascading Hazards on Land

Landslides set off by earthquakes frequently cause secondary disasters that can be just as destructive as the slides themselves. One of the most dangerous is the formation of landslide dams, where debris blocks a river and creates a rapidly filling lake. The Tangjiashan landslide dam, created by the 2008 Wenchuan earthquake, impounded a lake holding an estimated 300 million cubic meters of water. The potential for a catastrophic breach threatened more than 2.5 million people in downstream towns and the city of Mianyang, 85 km away. Authorities had to evacuate parts of the city until an artificial spillway was cut to drain the lake in a controlled manner.¹⁰

Even after such immediate crises are resolved, the sediment dumped into river channels keeps causing problems for a long time. The Min Jiang River, which drains part of the Wenchuan earthquake zone, was still carrying dramatically elevated sediment loads a full decade after the event. Researchers found roughly a sixfold increase in total riverine sediment flux sustained over that period, with the river having exported only about 9% of the total landslide sediment mass after ten years. Bedload (the coarser material that bounces along the river bottom) increased by roughly 27 times compared to pre-earthquake levels and accounted for about 65% of the post-earthquake sediment export.¹¹ That ongoing sediment pulse raises flood risk, buries farmland along river margins, and can clog reservoirs and water-supply infrastructure downstream.

Mountain river systems respond to this overloading by fundamentally reorganizing. Channels widen, gravel bars shift, and the rivers develop armor layers of coarser sediment on their beds. Observations from detailed terrain surveys have shown how these changes modulate the delivery of sediment to lower reaches, sometimes storing huge volumes temporarily and then releasing them during high-flow events.¹² The practical upshot is that flood and aggradation hazards in populated valleys downstream of earthquake-damaged mountains may persist for decades.

Submarine Landslides and Tsunamis

Landslides triggered by earthquakes are not confined to mountains. Underwater slopes, particularly along continental margins and volcanic islands, can fail during seismic events with devastating consequences. In 1998, a seabed slump off northern Papua New Guinea generated a tsunami up to 15 meters high that killed more than 2,200 people. That event changed the scientific understanding of tsunami generation and proved that submarine landslides can produce devastating waves through failure mechanisms very different from classical earthquake-generated tsunamis.¹³ Some of the most destructive historical tsunamis, including those at Messina in 1908, Puerto Rico in 1928, and Japan in 2011, involved a combination of earthquake and landslide mechanisms.

The 2006 Pingtung earthquake doublet off southern Taiwan illustrated how these hazard chains unfold in sequence. The initial earthquakes triggered regional tsunamis and submarine landslides, which in turn generated turbidity currents that severed telecommunication cables across the Kaoping Canyon and Manila Trench. Wave analysis suggested that landslide-triggered tsunami waves may have arrived in two episodes: one set produced by slides triggered simultaneously with the earthquake doublet, and a second set possibly caused by the largest aftershock 14 hours later triggering additional slides near the canyon head.¹⁴ The cable damage alone caused millions of dollars in losses and disrupted communications across East Asia for weeks.

Human Toll and Infrastructure Damage

The human cost of earthquake-triggered landslides is staggering. The Wenchuan earthquake’s roughly 200,000 landslides killed about 20,000 people and created more than 10,000 potential ongoing hazard sites.¹⁵ In the 2018 Hokkaido Eastern Iburi earthquake in Japan, a much smaller event at magnitude 6.6, at least 9,295 landslides were triggered across an area of about 9.4 square kilometers, killing 36 people and causing losses of roughly 153 billion yen.

Roads are among the most vulnerable pieces of infrastructure. After the 2022 Luding earthquake in China (magnitude 6.8), more than 95% of highway damage consisted of roadbed failures: cracks, collapse, burial under debris, and washout. Over 66% of that roadbed damage was classified as medium severity or higher.¹⁶ Damaged roads are not just an economic loss; they cut off access for rescue teams and delay evacuations. In mountainous regions where a single highway may be the only route into a valley, a landslide blocking that road can strand entire communities for days.

Environmental Aftermath

The ecological footprint of earthquake landslides extends well beyond the visible scars on hillsides. Vegetation stripped away by landslides represents a sudden loss of carbon that the ecosystem had been accumulating for years or decades. In the Wenchuan earthquake zone, vegetation net carbon dropped by about 138 grams of carbon per square meter in the month the earthquake struck, a decline of roughly 17% compared to the average for that month in the seven preceding years.¹⁷ Recovery of vegetation carbon across different landslide areas took more than a decade, and some scarred slopes remained carbon sinks long after the event.

The landslides also eroded massive quantities of organic carbon from the soil. The Wenchuan event stripped away an estimated 5.5 teragrams of organic carbon, amounting to roughly 2 to 6 percent of the global annual total of terrestrial carbon erosion. The majority of this carbon was eroded within just two river basins (the Minjiang and Tuojiang), and it was deposited widely along hillslopes and in river channels, awaiting further transport or chemical transformation.¹⁸ Whether that carbon ultimately gets buried (removing it from the atmosphere for geological time) or oxidized (releasing it as carbon dioxide) has implications for the global carbon cycle, though the balance remains an active area of research.

Detecting and Mapping Landslides Rapidly

Speed is critical after an earthquake. Knowing where landslides have occurred helps rescue teams prioritize, engineers assess dam breach risk, and authorities plan evacuations. Traditionally, landslide inventories were assembled from aerial photographs and field surveys, a process that could take weeks or months. Satellite radar has changed that equation. Synthetic aperture radar (SAR) penetrates clouds, works at night, and is acquired on regular schedules worldwide. Using freely available Sentinel-1 radar data in cloud-based platforms, researchers can generate landslide density heatmaps within days of a triggering event.¹⁹

Machine learning is accelerating this process further. One approach uses a progressive training strategy: analysts start with a small, incomplete inventory (as little as 100 square meters of mapped landslides) drawn from the first available radar image, then iteratively improve the map as cloud-free optical images become available. For the 2018 Hokkaido earthquake, this method produced a preliminary landslide inventory within a single day, with only a modest performance penalty compared to traditional, labor-intensive mapping.²⁰

For longer-term monitoring, radar interferometry can detect slow-moving landslides that were reactivated or accelerated by an earthquake but did not fail catastrophically during the shaking. In the Jiuzhaigou area after a 2017 earthquake, InSAR analysis identified 16 slow-moving landslides, including historical slides that had been dormant and were remobilized by the event. Researchers categorized three distinct patterns: acceleration of already-moving landslides, reactivation of dormant ones, and remobilization of new earthquake-triggered debris.²¹ This kind of ongoing surveillance is essential for managing the legacy effect discussed earlier.

Engineering Slope Stabilization

Where slopes threaten homes, roads, or critical infrastructure, engineers have several tools to reduce landslide risk during earthquakes. Soil nailing is one of the most common. Steel bars are drilled into the slope face and grouted in place, effectively stitching the unstable soil to more competent material beneath. Shaking table experiments have confirmed that the seismic response of soil-nailed walls depends heavily on nail length and the character of the shaking. Conventional methods for predicting the earth pressures these walls must resist during an earthquake tend to overestimate the forces and predict the wrong location for the resultant pressure, meaning designs based on those older formulas are unnecessarily conservative in some cases.²²

Three-dimensional analysis of nailed slopes has advanced design practice by accounting for the fact that real slopes are not infinite two-dimensional slices. Both the tensile failure of nails (snapping) and pull-out failure (nails sliding out of the ground) are considered in newer approaches, and stability charts have been developed that allow engineers to quickly evaluate the safety factor of a reinforced slope under seismic loading without running complex computer models for every case.²³ Beyond soil nailing, other common measures include retaining walls, rock bolts, drainage systems to lower pore-water pressure (addressing one of the key triggering mechanisms), and reshaping slopes to reduce their steepness.

Risk Assessment and Land-Use Planning

Engineering individual slopes is expensive and only practical for specific high-value sites. For broader protection, communities need landslide susceptibility maps that guide where people build in the first place. Modern susceptibility assessments combine terrain data like slope angle, elevation, curvature, and distance to rivers and faults with geological maps and records of past landslides. Machine learning classifiers trained on these factors can predict which areas are most likely to fail in a future earthquake. Testing multiple algorithms against real landslide inventories has shown that ensemble methods (specifically the “extra trees” classifier) tend to outperform other approaches in accuracy.²⁴

Susceptibility maps tell you where landslides are likely; hazard maps go further by estimating how often they might happen. The shift from qualitative zonation (“high/medium/low risk”) toward quantitative approaches that assign annual probabilities to potential landslides has been a major trend in recent decades, driven by the need for risk management frameworks that can weigh costs and benefits of mitigation.²⁵ In practice, this means a planning authority can say not just “this hillside is dangerous” but “there is roughly a 1-in-100-year chance of a landslide reaching this proposed housing development,” which gives decision-makers something concrete to work with.

Scaling relationships also help with regional planning. The total area and volume of landslides triggered by an earthquake follow predictable patterns tied to the earthquake’s magnitude and source characteristics. Models built on these relationships can predict the total landslide volume for a given earthquake scenario to within a factor of two in most cases.²⁶ That kind of estimate is invaluable for pre-positioning emergency resources and planning evacuation routes in seismically active mountain regions.

Reading the Past to Forecast the Future

One underappreciated tool in earthquake landslide assessment is the geological record itself. Ancient landslides whose scars and slip surfaces are still preserved in the landscape can be back-analyzed to estimate how strong the ground shaking must have been to trigger them. Researchers studying prehistoric landslides around the Sea of Galilee applied this technique and found that the shaking needed to cause those failures ranged between 0.15 and 0.5 g, which is consistent with what seismologists independently estimate for large earthquakes along the Dead Sea transform fault.²⁷ In regions where the written historical record is short but the geological record is long, old landslides become a kind of seismograph, filling in gaps about what past earthquakes were capable of and, by extension, what future ones might do to the landscape.

This approach also helps calibrate the size-frequency distributions that underpin hazard maps. The observation that landslide sizes follow a power-law distribution for events above a certain threshold has been confirmed across multiple earthquake inventories and rainfall-triggered events alike.²⁸ Knowing the shape of that distribution lets planners estimate not just how many landslides a future earthquake might trigger but roughly how large the biggest ones could be, which is the number that drives dam-breach scenarios, road closures, and evacuation planning.