A landslide is the movement of a mass of rock, soil, or debris down a slope under the influence of gravity. In broad scientific and hazard-management usage, the term includes falls, topples, slides, spreads, and flows, although a slide in the narrower mechanical sense moves along a recognizable failure surface. Landslides occur on mountainsides, coastal cliffs, riverbanks, artificial excavations, and submerged slopes. They range from small rockfalls to failures involving cubic kilometres of material, and from almost imperceptible movement to destructive motion lasting only seconds.
Landslides are major agents of landscape change and a widespread natural hazard. They transfer sediment from hillsides to valleys, reshape drainage networks, damage settlements and transport routes, and sometimes generate floods or destructive waves. Their occurrence reflects the interaction of geological structure, slope geometry, water, weathering, and external disturbances such as earthquakes or excavation. Understanding them requires distinguishing the conditions that make a slope susceptible, the processes that initiate failure, and the mechanisms that control movement after failure.
Classification and physical characteristics
Landslide classification combines the material involved with the type of movement. The influential classification published by David J. Varnes in 1978, revised with David M. Cruden in 1996, distinguishes falls, topples, slides, spreads, and flows. A further update by Oldrich Hungr, Serge Leroueil, and Luciano Picarelli in 2014 refined terminology to better represent material behaviour and movement mechanisms. These systems are related rather than interchangeable, and terminology can differ between engineering geology, geomorphology, and national mapping programmes.
A fall begins when material detaches from a steep face and descends mainly through the air, commonly bouncing or rolling after impact. A topple involves forward rotation about a point or axis below the centre of gravity of the displaced mass. Both are common where joints divide rock into blocks or columns. A slide involves displacement concentrated along one or more surfaces or relatively thin zones of intense shearing. Rotational slides move on curved surfaces and often produce backward-tilted blocks, whereas translational slides move along approximately planar surfaces such as bedding planes, joints, or contacts between soil and bedrock.
A spread involves lateral extension, commonly above a weaker layer that deforms or loses strength. Some spreads develop during soil liquefaction, when saturated, loose granular sediment experiences a large increase in pore-water pressure under shaking. A flow, by contrast, exhibits distributed deformation within the moving mass. A debris flow contains a mixture of water and sediment that moves with flow-like behaviour, often through a confined channel. Earthflows typically involve finer-grained material and may move intermittently for years. Rapid rock avalanches originate in large rock-slope failures and fragment extensively during motion.
Material categories also matter. Rock denotes relatively intact bedrock before movement, while soil is unconsolidated or weakly consolidated material. In the Varnes-derived scheme, earth is dominated by sand-sized and finer particles; debris contains a substantial proportion of coarser fragments. These are engineering classifications rather than ordinary-language distinctions. A landslide may change type as it travels: a rockslide can fragment into a rock avalanche, while a shallow soil slide can liquefy and become a debris flow.
A typical slide has a crown above the displaced material, a main scarp marking its upper boundary, lateral margins, and a toe at its lower end. The depleted source area and the accumulation zone need not coincide. Movement rates span many orders of magnitude: the Cruden–Varnes velocity scale extends from extremely slow movement below approximately 16 millimetres per year to extremely rapid movement above 5 metres per second. Size, speed, depth, and travel distance are separate characteristics; a very large landslide is not necessarily fast, and a small rockfall can still be lethal.
Slope mechanics and the development of failure
Slope stability depends on the balance between forces driving movement and the strength resisting it. The downslope component of a material’s weight provides much of the driving force. Resistance comes from friction, cohesion, rock bridges, root reinforcement, and geometric confinement. Geotechnical engineering commonly expresses this balance through a factor of safety, the ratio of available resisting strength to the strength required for equilibrium. A calculated value above one indicates stability under the assumptions of the model; a value below one indicates that those assumptions do not support equilibrium. It is not a direct probability of failure.
For many soils and rock discontinuities, shear strength is approximated by the Mohr–Coulomb relation:
where is shear strength at failure, is effective cohesion, is total normal stress, is pore-water pressure, and is the effective friction angle. The quantity is the effective normal stress. This equation explains why increased groundwater pressure can destabilize a slope without a corresponding increase in its steepness: water pressure reduces the effective force pressing grains or discontinuity surfaces together. Its application requires suitable drainage assumptions and material parameters; it does not describe every soil or rock behaviour.
Slope geometry controls both loading and confinement. Excavating the toe can remove support, while placing fill near the crest can increase driving stresses. In rock, the orientation and persistence of fractures may be more important than the strength of intact specimens. A strong limestone or granite slope can fail along weak, continuous joints. Planar sliding requires a discontinuity with a suitable orientation relative to the slope face; wedge sliding occurs where intersecting discontinuities define a removable block. Rotational failures are more typical of relatively homogeneous soils or highly fractured weak rock, although real slopes are often more complicated.
Failure commonly develops progressively rather than everywhere at once. Small zones reach peak strength, deform, and weaken, transferring stress to neighbouring material. Clay-rich shear zones may lose considerable strength as mineral particles become aligned. An old landslide can therefore reactivate along an established surface at stresses below those required to initiate a new failure. Time-dependent deformation, chemical alteration, opening of cracks, and repeated wetting and drying can gradually change the stability margin.
Movement after failure introduces additional processes. Saturated loose sediment may contract during shearing, raising pore pressure and reducing resistance. Fragmentation changes grain sizes and energy dissipation, while entrainment adds material from the travel path. Some large rock avalanches travel unusually long distances relative to their descent. Proposed explanations include evolving granular mechanics, elevated pore pressures where water is present, and localized reductions in basal resistance. No single mechanism adequately describes every highly mobile landslide.
Water, earthquakes, and human disturbance
Landslide investigations distinguish predisposing conditions, which create susceptibility, from triggers, which initiate a particular movement episode. Weak rock, steep terrain, adverse bedding, and old failure surfaces may persist for centuries. A storm, earthquake, or excavation can then cause failure by changing stresses or material strength. The distinction is not absolute: rainfall can gradually weaken a slope over a season and also provide the final trigger during an intense burst.
Rainfall acts through several pathways. Infiltration can raise groundwater levels, produce perched water above less permeable layers, and increase pressure within fractures. In initially unsaturated soils, wetting reduces the stabilizing effect of matric suction. Added water also increases weight, although changes in pressure and suction are often more important than added mass alone. Shallow landslides may respond within minutes or hours, while deep-seated failures can lag rainfall by weeks or months because water must move through thick or poorly permeable material. Consequently, storm intensity, storm duration, and antecedent wetness all influence failure.
Regional rainfall regimes shape seasonal patterns. Monsoon rainfall generates repeated periods of high landslide activity across parts of South and East Asia, while tropical cyclones can trigger thousands of failures during individual storms. Snowmelt and rain falling on snow may supply substantial water even without exceptional rainfall totals. River and wave erosion remove support at slope toes. Rapid lowering of a reservoir can also be destabilizing: external water support decreases while pore pressures inside a slowly draining slope remain high.
Earthquakes impose rapidly changing inertial forces and can damage rock masses, increase pore pressures, and dislodge unstable blocks. The distribution of earthquake-triggered landslides depends on shaking intensity, topography, geology, and distance from the rupture, not simply earthquake magnitude. The 1999 Jiji earthquake in Taiwan triggered numerous failures, including the large Jiufengershan landslide. Strong earthquakes also leave fractured rock and abundant loose sediment that can sustain elevated landslide and debris-flow activity during subsequent rainy seasons.
Human activity modifies both loading and drainage. Road cuts can steepen slopes and expose discontinuities, while poorly placed spoil may load unstable ground. Leaking pipes, concentrated runoff, irrigation, and reservoir operations can alter groundwater conditions. Deforestation may reduce root reinforcement after roots decay and change interception and transpiration, but its effects vary with soil depth, vegetation, and rainfall. Mining, quarrying, and construction can combine several mechanisms. Attribution therefore requires examination of the actual failure process rather than assuming that either intense rain or nearby development alone explains an event.
Global distribution and landscape evolution
Landslides occur wherever sufficient relief and susceptible materials coincide. They are especially prominent in active mountain belts, deeply incised plateaus, volcanic regions, glaciated valleys, and steep coastal terrain. Plate tectonics continually generates relief and fractured rock in ranges such as the Himalayas, Andes, and Southern Alps of New Zealand. Rivers then incise valleys, creating steep hillsides from which material is removed by slope failures. Even regions with modest relief can experience serious landslides where weak clays, eroding riverbanks, or artificial cuts provide suitable conditions.
Geology strongly influences the regional pattern. Alternating permeable and impermeable strata can concentrate groundwater at contacts. Bedding that dips toward a valley may provide potential sliding surfaces. Volcanic deposits can be loose, porous, or altered to weak clay-rich material, while the collapse of volcanic flanks can mobilize enormous volumes. In formerly glaciated regions, valley sides may remain oversteepened after ice retreat. The removal of glacier support, long-term fracture development, and changing groundwater conditions can contribute to failures long after deglaciation.
Landslides are not merely isolated disturbances superimposed on an otherwise stable landscape. They are integral to geomorphology, controlling hillslope shape and the delivery of sediment to channels. In some rapidly eroding mountains, large failures account for a substantial fraction of total sediment production. A single earthquake or storm can generate a pulse of loose material that takes years or decades to move through a drainage basin. Sediment may be stored temporarily on slopes, in tributaries, or behind natural dams rather than immediately reaching the main river.
The ecological effects vary across the landslide footprint. Source areas can lose soil and vegetation, exposing bedrock or nutrient-poor substrate. Depositional areas may bury existing habitats while creating new surfaces, ponds, or channel margins. Ecological succession on these disturbed surfaces depends on substrate stability, moisture, seed supply, and climate. Repeated landsliding can maintain a mosaic of vegetation ages and physical habitats within a forest, although frequent disturbance may also prevent soil development and persistent tree cover.
Landslides also affect the movement of carbon. They can expose organic matter to decomposition, bury it in deposits, and transfer soil and woody debris into rivers. The net carbon effect depends on what is eroded, how rapidly material is transported, and whether organic carbon is oxidized or preserved. Old landslide deposits, displaced terraces, buried vegetation, and scarps provide evidence of past landscape change. Dating such features helps distinguish a recently triggered failure from the reactivation of a slope that has moved repeatedly over thousands of years.
Runout, natural dams, and underwater failures
The area damaged by a landslide can extend far beyond its source. Runout describes movement through the transport and deposition zones and is governed by volume, material properties, water content, terrain, and interaction with obstacles. A fall may follow a series of ballistic trajectories, whereas a debris flow can remain channelized for kilometres. Large rock avalanches may cross valleys and climb opposing slopes. Changes in path geometry can cause material to spread, pile up, or divide into separate branches, making the source area alone an inadequate guide to the impact zone.
A simple measure of apparent mobility is the ratio , where is vertical descent and is horizontal travel distance measured using a specified convention. Lower values indicate greater travel relative to fall height. This ratio is useful for comparisons and preliminary assessments, but it does not directly measure a material’s friction coefficient. The source and deposit geometry, fragmentation, entrainment, and path curvature all affect it. Modern runout models instead represent motion using particle methods, depth-averaged flow equations, or other numerical approaches, with parameters constrained where possible by observed events.
When displaced material blocks a valley, it forms a landslide dam. Impounded water may inundate settlements upstream, while erosion or collapse of the barrier can release a destructive flood downstream. Longevity varies enormously. Some dams fail within hours; others persist for centuries. Stability depends on the dam’s geometry, internal structure, grain-size distribution, inflow, and spillway evolution. The 1911 Usoi landslide in the Pamir Mountains blocked the Murghab River and created Lake Sarez. In 2008, the Wenchuan earthquake in China produced numerous landslide dams, including the Tangjiashan barrier, which required major emergency management.
Landslides entering reservoirs, lakes, or fjords can generate tsunamis or other large displacement waves. Wave size depends on slide speed, volume, geometry, impact angle, water depth, and basin shape. At Lituya Bay, Alaska, in 1958, an earthquake-triggered rockslide generated a wave whose maximum documented run-up reached about 524 metres on the opposite slope. This was a highly localized run-up in a confined inlet, not the height of a wave travelling across an open ocean. Reservoir margins and steep-sided lakes can therefore present hazards even where oceanic tsunamis are absent.
Submarine landslides occur on continental margins, delta fronts, volcanic island flanks, and lake beds. Weak sediment layers, rapid deposition, elevated pore pressures, and earthquakes can contribute to failure. Water changes effective stresses and flow dynamics, and displaced sediment may evolve into a turbidity current. The Storegga Slide off Norway, about 8,200 years ago, displaced thousands of cubic kilometres of sediment and generated a tsunami recorded around parts of the North Atlantic. Such events show that relatively gentle underwater slopes can fail on scales rarely matched by terrestrial landslides.
Observation, mapping, and warning systems
Landslide investigation usually begins with an inventory: a spatial record of known failures, ideally including their type, date, activity, and boundaries. Inventories can be compiled from field surveys, historical records, aerial photographs, and satellite imagery. Their completeness varies with vegetation cover, image resolution, preservation, and reporting effort. A map of visible scars therefore records both actual landslide occurrence and the limits of observation. Separating landslides triggered by a single earthquake or storm from older background failures is particularly important for understanding triggering conditions.
Remote sensing has greatly expanded observation. Optical imagery reveals fresh scars, disrupted vegetation, and displaced terrain, while repeated images can measure surface movement by feature tracking. Lidar provides detailed topography and, where sufficient laser returns reach the ground, can reveal scarps and hummocky deposits beneath forest. Interferometric synthetic aperture radar measures changes in radar distance between repeated satellite observations. It is especially useful for slow deformation, but dense vegetation, changing surface conditions, steep terrain, and rapid movement can cause decorrelation or ambiguous measurements.
Ground instruments provide complementary information. Survey benchmarks and satellite-positioning receivers measure displacement at selected points. Borehole inclinometers identify deformation at depth, while piezometers measure groundwater pressure. Crack meters, extensometers, ground-based radar, and automated total stations can track movement at short intervals. Rain gauges and soil-moisture sensors help connect displacement to hydrological conditions. An effective monitoring layout follows a geological interpretation of the slope; large numbers of instruments do not compensate for monitoring the wrong block or missing the controlling groundwater system.
Regional warning systems often use rainfall thresholds, expressed as combinations of intensity, duration, and antecedent conditions. Exceeding a threshold indicates that conditions resemble those associated with earlier landslides; it does not mean every susceptible slope will fail. Physically based systems estimate infiltration and stability, while statistical systems relate past occurrences to environmental predictors. Satellite rainfall estimates extend coverage but may miss local extremes in mountainous terrain. Machine learning can identify complex associations, although performance depends on representative training data and appropriate tests using independent places or time periods.
Site-specific warnings may use accelerating displacement, rising pore pressure, or several indicators together. Some failing slopes display acceleration patterns that help constrain a possible failure time, but others move episodically, accelerate without collapse, or fail with little detectable precursor. Warning effectiveness also depends on communications, defined responsibilities, and the time needed to act. Evaluation therefore includes false alarms, missed events, lead time, and whether messages reach exposed populations. A technically accurate forecast has limited protective value if roads, communications, or evacuation arrangements fail during the same storm.
Hazard assessment, exposure, and risk reduction
Landslide susceptibility, hazard, and risk answer different questions. Susceptibility concerns where failures are more likely given terrain and environmental conditions. Hazard adds the probability of occurrence within a specified period and often the likely magnitude or intensity. Risk incorporates exposed people, buildings, services, and their vulnerability to damage. A highly susceptible remote slope may pose little immediate societal risk, whereas a smaller unstable slope above a densely occupied road can present substantial risk. These distinctions determine what data and modelling are appropriate.
Susceptibility mapping commonly uses slope angle, geology, land cover, drainage, and landslide inventories. Statistical models estimate relationships between mapped failures and explanatory variables; physically based models represent stress balance and hydrology. Both have limitations. Inventories may omit small or old failures, geological maps may be too coarse, and mapped land cover may postdate the event being modelled. Validation requires more than a good fit to the same landslides used to construct the model. Transfer to another region or rainfall regime can fail when the underlying controlling processes differ.
Structural stabilization seeks to alter forces or strengthen resisting elements. Surface channels and subsurface drains can reduce water entry or pore pressure. Regrading decreases driving forces, and appropriately designed toe support can increase resistance. Retaining structures, soil nails, rock bolts, anchors, and piles transfer loads into more stable ground. Rockfall nets, catch ditches, barriers, and protective galleries mainly intercept or redirect moving material rather than stabilize the entire source. Each approach requires knowledge of the expected failure mechanism, potential volume, groundwater conditions, and forces.
Land-use measures reduce exposure through development controls, setbacks, and avoidance of source or runout zones. Urban planning must consider areas below and beyond unstable slopes, not just the slopes themselves. Road alignment and drainage design are especially consequential in steep terrain because roads can both create failures and become isolated by them. Vegetation-based measures can control surface erosion and reinforce shallow soil, but cannot be assumed to stabilize deep bedrock failures. Ecological restoration and engineered works address different parts of the problem and may be combined where their functions are understood.
Residual risk remains after treatment. Drains can clog, protective structures have finite capacities, and an event larger than the design scenario can exceed protection. Maintenance and renewed assessment are therefore part of the intervention rather than optional additions. Economic appraisal must also account for indirect losses, including interrupted transport, water supply, power, and commerce. Because many landslides occur during storms or earthquakes that damage several systems at once, assessing individual structures in isolation can underestimate the consequences of network-wide disruption.
Historical disasters and the development of knowledge
Historical landslides illustrate the diversity of failure mechanisms and the difficulty of inferring safety from past stability. On 2 September 1806, the Goldau rockslide in Switzerland destroyed settlements beneath the Rossberg and killed more than 450 people. Weak layers and geological structure helped control the failure, while prolonged wet conditions contributed to instability. The disaster became an important European example of catastrophic rock-slope movement and drew attention to the relationship between stratigraphy, groundwater, and slope orientation.
On 9 October 1963, approximately 260 million cubic metres of rock moved from Monte Toc into the Vajont reservoir in northeastern Italy. The displaced water overtopped the dam and devastated communities downstream, killing about 1,900 people. The dam itself largely remained standing. Geological structure, weak layers, an ancient landslide geometry, groundwater, and reservoir-level changes were important to the slope’s behaviour. Monitoring had documented movement before the final collapse, making Vajont a central case in the study of progressive failure, reservoir-slope interaction, and the consequences of underestimating landslide-generated waves.
The 1970 disaster at Yungay, Peru, followed the 31 May Ancash earthquake. A rock-and-ice avalanche from Huascarán developed into a highly mobile mass that overwhelmed Yungay and nearby areas, killing many thousands. It demonstrated how a mountain-source failure could affect populated terrain far from the initial detachment zone. In contrast, the 18 May 1980 sector collapse of Mount St. Helens in the United States was closely connected with volcanic deformation and eruptive activity. Its debris avalanche, with a volume of roughly 2.5 cubic kilometres, provided unusually detailed observations of a giant volcanic landslide.
More recent events have refined understanding of rainfall-driven failures. On 22 March 2014, the Oso landslide in Washington State, United States, killed 43 people when a rapidly moving mass crossed the North Fork Stillaguamish River and struck a residential area. The failure involved glacial sediments within a landscape containing older landslides. Its mobility and complex movement prompted detailed research into stratigraphy, groundwater, material behaviour, and runout. The case also highlighted how historical landslide deposits can identify a hazardous setting without precisely predicting the timing or extent of the next event.
Scientific development has combined such case studies with mechanics, mapping, and experimentation. Twentieth-century soil mechanics, including work associated with Karl Terzaghi, established effective-stress concepts central to understanding water-driven instability. Engineering-geological classifications made inventories more comparable, while aerial photography and later satellite observations expanded regional analysis. Large-scale experiments and computational models now investigate fragmentation, entrainment, and flow transitions. Nevertheless, field reconstruction remains essential: a numerical model can reproduce a deposit for different combinations of parameters, so agreement with the final outline alone does not establish that the underlying mechanism has been identified.
Climate change and unresolved scientific questions
Climate change can alter landslide activity through precipitation, snow and ice, vegetation, and hydrology. More intense rainfall can increase the likelihood of shallow failures where storms exceed local infiltration and stability thresholds. However, a change in extreme rainfall does not translate directly into the same proportional change in landslides. Soil thickness, drainage, vegetation, sediment availability, and the season in which rain occurs all matter. Projecting future landslide occurrence therefore requires linking climate information to slope processes rather than applying a universal multiplier.
High mountains are affected by glacier retreat and warming of permafrost. Ice loss can change support at valley sides, while thawing ice-filled fractures alters the mechanical and hydraulic properties of rock masses. Warm periods may permit water to enter fractures previously sealed by ice, and changing freeze–thaw conditions can promote crack growth. The response depends on rock temperature, fracture geometry, water availability, and the history of glacier retreat. Individual failures cannot generally be attributed to warming solely because they occur in a warming mountain range.
Vegetation and disturbance introduce further interactions. Drought can alter plant cover, soil cracking, and the hydrological response to later storms. Wildfire removes protective vegetation and can change soil properties, increasing runoff and erosion. Some post-fire debris flows begin through intense runoff entraining sediment rather than through a discrete landslide, an important distinction when selecting models and warning thresholds. Over longer periods, vegetation recovery changes root reinforcement and water use. Climate, land management, and disturbance history thus jointly influence slope behaviour.
Several fundamental prediction problems remain. The timing of deep-seated failure depends on subsurface structures and groundwater systems that are difficult to observe continuously. The transition from slow sliding to catastrophic motion is not inevitable, and apparently similar slopes can behave differently. Large, rare failures provide few well-instrumented examples, limiting statistical inference. Runout prediction remains sensitive to uncertain material behaviour, erosion of the travel path, and the amount of water incorporated during motion. These uncertainties affect both the mapped impact area and estimates of destructive intensity.
Global comparisons are also constrained by uneven records. Fatal events are more likely to be documented than small failures in remote terrain, and earthquake-triggered inventories may be far more complete than historical rainfall inventories. Changes in reporting, satellite coverage, and settlement can mimic changes in occurrence or losses. Research increasingly combines repeat topographic surveys, continuous deformation measurements, hydrological monitoring, and event-based inventories to separate these effects. Distinguishing changes in the physical frequency of landslides from changes in exposure and observation remains essential to interpreting long-term trends.