David K. Keefer

Landslides caused by earthquakes

Data from 40 historical world-wide earthquakes were studied to determine the characteristics, geologic environments, and hazards of landslides caused by seismic events. This sample of 40 events was supplemented with intensity data from several hundred United States earthquakes to study relations between landslide distribution and seismic parameters. Fourteen types of landslides were identified in the earthquakes studied. The most abundant of these were rock falls, disrupted soil slides, and rock slides. The greatest losses of human life were due to rock avalanches, rapid soil flows, and rock falls. Correlations between magnitude (M) and landslide distribution show that the maximum area likely to be affected by landslides in a seismic event increases from approximately 0 at M ≅ 4.0 to 500,000 km2 at M = 9.2. Threshold magnitudes, minimum shaking intensities, and relations between M and distance from epicenter or fault rupture were used to define relative levels of shaking that trigger landslides in susceptible materials. Four types of internally disrupted landslides—rock falls, rock slides, soil falls, and disrupted soil slides—are initiated by the weakest shaking. More coherent, deeper-seated slides require stronger shaking; lateral spreads and flows require shaking that is stronger still; and the strongest shaking is probably required for very highly disrupted rock avalanches and soil avalanches. Each type of earthquake-induced landslide occurs in a particular suite of geologic environments. These range from overhanging slopes of well-indurated rock to slopes of less than 1° underlain by soft, unconsolidated sediments. Materials most susceptible to earthquake-induced landslides include weakly cemented rocks, more-indurated rocks with prominent or pervasive discontinuities, residual and colluvial sand, volcanic soils containing sensitive clay, loess, cemented soils, granular alluvium, granular deltaic deposits, and granular man-made fill. Few earthquake-induced landslides reactivate older landslides; most are in materials that have not previously failed.

Real-Time Landslide Warning During Heavy Rainfall

A real-time system for issuing warnings of landslides during major storms is being developed for the San Francisco Bay region, California. The system is based on empirical and theoretical relations between rainfall and landslide initiation, geologic determination of areas susceptible to landslides, real-time monitoring of a regional network of telemetering rain gages, and National Weather Service precipitation forecasts. This system was used to issue warnings during the storms of 12 to 21 February 1986, which produced 800 millimeters of rainfall in the region. Although analysis after the storms suggests that modifications and additional development are needed, the system successfully predicted the times of major landslide events. It could be used as a prototype for systems in other landslide-prone regions.

The importance of earthquake-induced landslides to long-term slope erosion and slope-failure hazards in seismically active regions

Abstract This paper describes a general method for determining the amount of earthquake-induced landsliding that occurs in a seismically active region over time; this determination can be used as a quantitative measure of the long-term hazard from seismically triggered landslides as well as a measure of the importance of this process to regional slope-erosion rates and landscape evolution. The method uses data from historical earthquakes to relate total volume of landslide material dislodged by an earthquake to the magnitude, M , and seismic moment, M0, of the earthquake. From worldwide data, a linear-regression relation between landslide volume, V, and M0 is determined as: V = M0/1018.9(± 0.13), where V is measured in m3 and M0 is in dyn-cm. To determine the amount of earthquake-generated landsliding over time, this relation is combined with data on seismic-moment release for a particular region, which may be derived from either earthquake-history or fault-slip data. The form of the M0−V relation allows the rate of production of earthquake-induced landslides over time to be determined from total rate of seismic-moment release without regard to the distribution of individual events, thus simplifying and generalizing the determination. Application of the method to twelve seismically active regions, with areas ranging from 13,275 to 2,308,000 km2, shows that erosion rates from earthquake-induced landslides vary significantly from region to region. Of the regions studied, the highest rates were determined for the island of Hawaii, New Zealand, western New Guinea, and the San Francisco Bay region of California. Significantly lower rates were determined for Iran, Tibet, the Sierra Nevada-Great Basin region of California, and central Japan (for the time period from 715 AD to the present). Intermediate rates were determined for Peru, southern California, onshore California, Turkey, and central Japan (for the time period from 1586 AD to the present). To determine the relative, long-term importance of seismically triggered landslides, these erosion rates are compared to erosion rates calculated for other slope processes and to rates calculated from fluvial sediment discharge. Comparisons with other slope processes indicate that earthquake-induced landslides are the predominant agents of slope erosion on the island of Hawaii, in the San Francisco Bay region, and in western New Guinea. For Hawaii, the San Francisco Bay region, and Sierra Nevada-Great Basin region of California, the erosion rates calculated for earthquake-induced landslides also exceed the regional erosion rates calculated from fluvial sediment discharge.

Analysis of the seismic origin of landslides: Examples from the New Madrid seismic zone

By analyzing two landslides in the New Madrid seismic zone, we develop an approach for judging if a landslide or group of landslides of unknown origin was more likely to have formed as a result of earthquake shaking or in aseismic conditions. The two landslides analyzed are representative of two groups of land-slides that previous research on the geomorphology and regional distribution of landslides in this region indicates may have been triggered by the 1811-1812 New Madrid earthquakes. Slope-stability models of aseismic conditions show that neither landslide is likely to have formed aseismically even in unrealistically high ground-water conditions. Dynamic stability analysis using Newmarks method shows that both slides probably would have experienced large inertial displacements during earthquake shaking similar to that which occurred in 1811-1812; these displacements are large enough that catastrophic failure is highly probable. Thus, the stability analyses are consistent with other lines of evidence that these landslides formed as a result of strong earthquake shaking during the 1811-1812 earthquakes. Our analysis yields a general relationship between Newmark landslide displacement, earthquake shaking intensity, and the critical acceleration of a landslide. Using this relationship, we estimate the minimum shaking intensities required to trigger the types of landslides studied: an mb = 5.8 or M = 5.9 earthquake is the lower bound threshold at zero epicentral distance that could trigger catastrophic movement of typical block slides in the New Madrid seismic zone; for earth flows, mb = 5.4 or M = 5.3 is the threshold earthquake.

A 38,000-year record of floods and debris flows in the Ilo region of southern Peru and its relation to El Niño events and great earthquakes

Abstract Previous work throughout the Ilo region of south coastal Peru has documented the existence of flood and debris-flow deposits produced by two El Nino events evidently much more severe than any in recent history. These two events have been dated to ca. AD 1300–1400 and AD 1607–08. The Late Pleistocene to Holocene record of older sedimentary deposits in this region is dominated by flood and debris-flow deposits of similar scale. These older deposits have been described and dated from three coastal, alluvial-fan sites. These deposits, which are as old as 38 200 years, are dominated by massive debris-flow deposits, several tens of cm thick, typically composed of cobble- and boulder-sized clasts in a matrix of silty sand, with characteristics indicating generation by heavy rainfall in an arid environment. Twenty-two radiocarbon dates and a single infrared-stimulated luminescence date show that particularly severe El Nino events occurred throughout the Late Pleistocene and two of three divisions of the Holocene with significantly different frequencies. The period of greatest activity was during the Early Holocene when at least six such events took place during a period of ca. 3600 years, beginning near the end of the Younger Dryas ca. 12 000 years ago. One of these events produced a debris flow that may have caused abandonment of the Paleo-Indian site at Quebrada Tacahuay, one of the oldest on the Andean coast. No severe events took place during the Middle Holocene between ca. 8400 and 5300 years ago, when a wide variety of other paleoclimate proxy records indicate that the El Nino–Southern Oscillation regime was particularly weak. Since ca. 5300 years ago, four of these severe events have taken place. The Late Pleistocene sequence is constrained by only two dates, which indicate that at least ten severe events took place between ca. 38 200 and 12 900 years ago. Mechanisms probably responsible for generating these large-scale deposits include: (1) ‘Mega-Ninos’ that produced anomalously heavy rainfall along most or all of the central Andean coast; (2) El Ninos that occurred shortly after great earthquakes that produced large amounts of sediment; or (3) El Ninos that produced anomalously heavy local rainfall. The existence of these large-scale deposits in the Ilo region implies a level of hazard much higher than indicated by the historical record alone.

Landslides Triggered by the 2002 Denali Fault, Alaska, Earthquake and the Inferred Nature of the Strong Shaking

The 2002 M7.9 Denali fault, Alaska, earthquake triggered thousands of landslides, primarily rock falls and rock slides, that ranged in volume from rock falls of a few cubic meters to rock avalanches having volumes as great as 15×106 m3. The pattern of landsliding was unusual; the number of slides was less than expected for an earthquake of this magnitude, and the landslides were concentrated in a narrow zone 30-km wide that straddled the fault rupture over its entire 300-km length. The large rock avalanches all clustered along the western third of the rupture zone where acceleration levels and ground-shaking frequencies are thought to have been the highest. Inferences about near-field strong shaking characteristics drawn from the interpretation of the landslide distribution are consistent with results of recent inversion modeling that indicate high-frequency energy generation was greatest in the western part of the fault rupture zone and decreased markedly to the east.

Influence of Surface-Normal Ground Acceleration on the Initiation of the Jih-Feng-Erh-Shan Landslide during the 1999 Chi-Chi, Taiwan, Earthquake

The 1999 Chi-Chi, Taiwan, earthquake triggered numerous landslides throughout a large area in the Central Range, to the east, southeast, and south of the fault rupture. Among them are two large rock avalanches, at Tsaoling and at Jih- Feng-Erh-Shan. At Jih-Feng-Erh-Shan, the entire thickness (30-50 m) of the Mio- cene Changhukeng Shale over an area of 1 km 2 slid down its bedding plane for a distance of about 1 km. Initial movement of the landslide was nearly purely trans- lational. We investigate the effect of surface-normal acceleration on the initiation of the Jih-Feng-Erh-Shan landslide using a block slide model. We show that this ac- celeration, currently not considered by dynamic slope-stability analysis methods, significantly influences the initiation of the landslide.

Statistical analysis of factors affecting landslide distribution in the new Madrid seismic zone, Tennessee and Kentucky

Abstract More than 220 large landslides along the bluffs bordering the Mississippi alluvial plain between Cairo, Ill., and Memphis, Tenn., are analyzed by discriminant analysis and multiple linear regression to determine the relative effects of slope height and steepness, stratigraphic variation, slope aspect, and proximity to the hypocenters of the 1811-12 New Madrid, Mo., earthquakes on the distribution of these landslides. Three types of landslides are analyzed: (1) old, coherent slumps and block slides, which have eroded and revegetated features and no active analogs in the area; (2) old earth flows, which are also eroded and revegetated; and (3) young rotational slumps, which are present only along near-river bluffs, and which are the only young, active landslides in the area. Discriminant analysis shows that only one characteristic differs significantly between bluffs with and without young rotational slumps: failed bluffs tend to have sand and clay at their base, which may render them more susceptible to fluvial erosion. Bluffs having old coherent slides are significantly higher, steeper, and closer to the hypocenters of the 1811-12 earthquakes than bluffs without these slides. Bluffs having old earth flows are likewise higher and closer to the earthquake hypocenters. Multiple regression analysis indicates that the distribution of young rotational slumps is affected most strongly by slope steepness: about one-third of the variation in the distribution is explained by variations in slope steepness. The distribution of old coherent slides and earth flows is affected most strongly by slope height, but the proximity to the hypocenters of the 1811-12 earthquakes also significantly affects the distribution. The results of the statistical analyses indicate that the only recently active landsliding in the area is along actively eroding river banks, where rotational slumps formed as bluffs are undercut by the river. The analyses further indicate that the old coherent slides and earth flows in the area are spatially related to the 1811-12 earthquake hypocenters and were thus probably triggered by those earthquakes. These results are consistent with findings of other recent investigations of landslides in the area that presented field, historical, and analytical evidence to demonstrate that old landslides in the area formed during the 1811-12 New Madrid earthquakes. Results of the multiple linear regression can also be used to approximate the relative susceptibility of the bluffs in the study area to seismically induced landsliding.

Rock Avalanches Caused by Earthquakes: Source Characteristics

Study of a worldwide sample of historical earthquakes showed that slopes most susceptible to catastrophic rock avalanches were higher than 150 meters and steeper than 25 degrees. The slopes were undercut by fluvial or glacial erosion, were composed of intensely fractured rock, and exhibited at least one other indicator of low strength or potential instability.

The Borah Peak, Idaho Earthquake of October 28, 1983—Liquefaction

The most pervasive and damaging effects of liquefaction generated by the 1983 Borah Peak, Idaho earthquake occurred in the Big Lost River and Thousand Springs Valleys above Mackay Reservoir. Less severe effects occurred in the Big Lost River Valley south of Mackay Reservoir and in the Pahsimeroi Valley. Nearly all of the liquefaction effects developed in floodplain alluvium of late Holocene age. However, the sediment that liquefied beneath the alluvial fans on the east side of the Thousand Springs Valley was deposited in late Pleistocene time. The distance from the fault to the farthest effect of liquefaction was unusually short for an M S = 7.2 event. The distribution of liquefaction effects were consistent, however, with the distribution of MMI intensity and estimated peak ground motion parameters, both of which attenuated more rapidly than is generally expected for an earthquake of this type and magnitude.