Loess Is More: Field Investigation and Slope Stability Analysis of the Tanana 440 Landslide, Interior Alaska

Landslides are geologic hazards that threaten human lives, property, and infrastructure. In the United States alone, landslide damage exceeds $3 billion annually (Spiker and Gori, 2003; Burns, 2007). To mitigate these natural disasters, we must first know where they occurred in the past. Although this task sounds simple, it involves creating a landslide inventory through thorough and systematic mapping of landslides as geomorphic features (Burns and Madin, 2009; Slaughter et al., 2017).

As part of the field verification, we conducted detailed mapping and a subsurface investigation of one of the larger landslides, the Tanana 440 (T440) landslide (see Figure 2 for location). This dormant landslide is approximately 2-km long by 1-km wide, with a total areal extent of 1.9 km². It occurred on a 12.5° slope in loess (windblown silt) deposits overlying schist bedrock. We interpret the T440 as a flow slide in loess (Hungr et al., 2001) that occurred sometime during the late Pleistocene to mid-Holocene. In reviewing the literature, we found that only a few studies address the strength properties of silt, either through laboratory tests (e.g., Higgins and Fragaszy, 1988; Derbyshire et al., 1994; Harris et al., 2008; and Zhuang et al., 2018) or field tests (e.g., Lu et al., 2013). In this paper, we (1) present an overview of the landslide mapping effort in the FNSB, with a summary of general findings; (2) present the results of our fieldwork and laboratory testing for the T440 landslide, including direct shear tests for silt and a comparison of those results to field vane shear testing; and (3) explore potential triggers for this landslide event using slope stability modeling.

The FNSB is located in Interior Alaska, which has a continental sub-Arctic climate (Wendler and Shulski, 2009). Continental climates are characterized by large annual temperature fluctuations, which are reflected by the record high (35.6°C, June 1969) and low (−54.4°C, January 1934) temperatures recorded in Fairbanks (ACRC, n.d.(b)). The average annual snowfall and precipitation calculated for Fairbanks in a 30-year period (1981–2010) were 165 cm and 27.5 cm, respectively (ACRC, n.d.(a)). The mean annual air temperature in Fairbanks is –2.4°C, which is low enough to support a periglacial environment and permafrost (French, 2007). The FNSB is within the discontinuous permafrost zone, where 50–90 percent of the ground is underlain by permafrost (Jorgenson et al., 2008).

Fairbanks is located along the southern edge of the Yukon-Tanana Upland, which is characterized by rounded, even-topped ridges that rise 152–914 m above adjacent valleys and have gentle side slopes (Wahrhaftig, 1965). The bedrock of these uplands surrounding Fairbanks is part of the Yukon-Tanana Terrane unit of Upper Paleozoic and older metamorphic rocks (Newberry et al., 1996). This region is mainly underlain by the heterogeneous Fairbanks Schist unit, which includes a variety of metamorphic rocks (e.g., schist, marble, quartzite, and amphibolite) but is predominantly quartzite and quartz muscovite schist (Newberry et al., 1996).

Interior Alaska experienced only localized glaciation of small cirque glaciers in mountainous areas (Péwé, 1975; Briner et al., 2017) during the Pleistocene, and instead much of the region is blanketed by Quaternary loess. The wind-blown silt was transported mainly from the outwash plains of the Tanana River (as well as other fluvial sources) by southern katabatic winds and deposited during glacial cycles (Péwé, 1955, 1975). Herb tundra vegetation characterized glacial landscapes in Interior Alaska, and this lack of trees and shrubs may have affected loess entrapment rates (Muhs et al., 2003a) and reduced the overall contribution of root cohesion to slope stability. Loess deposition in Alaska began as far back as 3 million years ago (Westgate et al., 1990), but deposition around Fairbanks may have begun 2 million years ago (Begét et al., 2008). The loess around Fairbanks ranges in thickness from 0.3–24.4 m on hill tops to 3–30.5 m on middle slopes (Péwé, 1955). Retransported silt in the valley bottoms ranges from 30.5- to 91.5-m thick (Péwé, 1955).

Interior Alaska is a seismically active region, and the FNSB has a history of moderate- to large-magnitude earthquakes. The three main tectonic structural drivers in the Interior are dextral strike-slip faults (e.g., the Denali fault), seismic zones (e.g., Minto Flats fault zone), and thrust faults in the foothills of the Alaska Range (Tape et al., 2015; Koehler et al., 2018). The M 7.9 Denali fault earthquake in 2002 was the largest seismic event recorded in Interior Alaska and triggered thousands of landslides in a ∼13–19-km band along the rupture (Fuis et al., 2003). Although the 2002 tremblor and associated landslides occurred to the south of the study area, large earthquake events also were recorded in FNSB, including a M 7.4 event in 1912, M 6.3 and M 6.5 events in 1929, and M 7.3 events in 1937 and 1947 (St. Amand, 1948; Péwé, 1982). Multiple rockslides were reported along the Tanana River after the 1947 earthquake event (St. Amand, 1948).

Figure 2 includes the mapped landslides in the portion of the FNSB with lidar coverage. We mapped a total of 1,679 landslides and field-verified 51 landslides (Schwarber et al., in review). The landslide desktop mapping and fieldwork methodology is explained in detail by Schwarber et al. (in review). Schwarber et al. (2021) presented preliminary qualitative interpretations of the landslides and described the mapping effort. The mapped landslides vary in age, material, and movement type (Figure 3). Based on geomorphic expression, we interpreted that most of the mapped landslides were prehistoric (i.e., older than the approximately 100 years since the community of Fairbanks was established). We assumed that landslides with rougher surface expression are younger than those with relatively smoother surfaces (McCalpin, 1984). There are also historic and currently active landslides, verified by relative age relationships with existing roads or trails and field observations. We identified landslides of varying ages that occurred in bedrock and in surficial deposits. Landslides occurring in bedrock typically featured the characteristics of translational or rotational movement, but it was not always possible to distinguish the exact movement type due to a lack of geomorphological expression. The landslides that occurred in silty surface deposits, however, almost always exhibited the characteristics of flow slides, such as a concave head scarp, curved flanks, and a lobate toe (Varnes, 1984). In general, it was more difficult to gage the relative age of prehistoric landslides in silt due to erosion and its impact on geomorphological expression. These prehistoric flow slides typically only had the arcuate toe deposit remaining, with no clearly identifiable head scarps or flanks due to severe gullying. The landslides in bedrock, by contrast, typically had easily identifiable head scarps, flanks, and toe deposit regions.

We performed a suite of geotechnical laboratory tests on 69 samples collected from the test holes and stratigraphy sites (see Table 1 for a summary of the samples tested). All samples were classified using the Unified Soil Classification System (USCS) according to ASTM D2847 (ASTM, 2017a). We selected five organic samples from different boreholes for radiometric date testing (conducted by Beta Analytic, Inc.). Three of the five samples were rootlets (i.e., less than 10 mm long and less than 0.5 mm in diameter), along with one paleosol and one charcoal sample. The samples were selected based on their location on the landslide and depth in an effort to bracket the age of the landslide movement.

Table 1.

Laboratory testing conducted for the T440 landslide.

View largeDownload slide

View Large

A large portion of our laboratory testing program consisted of direct shear tests on silt samples. Most published direct shear testing results are for sandy or clayey soils, with only a few studies on silt or loess using other methods such as ring shear (e.g., Derbyshire et al., 1994; Zhuang et al., 2018). Typical shear rates varied from 0.02 mm/min to 2 mm/min for clayey to sandy soil, respectively, to achieve a shear strain rate of 5 percent per hour following ASTM D6528 (ASTM, 2017c). Based on published rates and the average grain-size distribution for the silt, we ran the tests at a shear speed of 0.5 mm/min. We tested both remolded soil and undisturbed samples collected from the stratigraphy locations. We tested the silt at a variety of moisture contents, including dry (∼0.3 percent), 5 percent, 10 percent, 15 percent, 20 percent, 25 percent, and saturated (∼37 percent) and at the in situ moisture content when sampled. We prepared all remolded samples to the average dry unit weight of the in situ soil as determined from volumetric measurements (which ranged from 11.9 to 12.7 kN/m³). All tests were duplicated for repeatability.

For the subsurface materials, we assumed silt overlying schist bedrock based on a published geologic map (Newberry et al., 1996) and well data of the local area (WELTS, 2005). The bedrock surface was placed parallel to the horizontal ground surface above and below the slope. Well data indicated a silt thickness of approximately 4.6 m at the top of the slope (WELTS, 2005), which is estimated to thicken to 100 m in the valley bottom (Péwé, 1955). From the surface reconstruction (Figures 4 and 5a), the surface slope angle (in silt) was estimated as 12.5°. The bedrock angle was estimated as 21.6° to account for the silt thicknesses previously described.

For the silt, we applied the average saturated (γ_sat) and dry unit weights (γ_dry) and average cohesion (c) and internal friction angle (Φ) obtained from our direct shear testing of the wetted T440 loess. For the schist bedrock, we selected saturated and unsaturated unit weights from the average values of laboratory-tested schist (Özbek et al., 2018), cohesion from shear strength reported by Wyllie and Norrish (1996), and the average friction angle corresponding to foliation planes paralleling the slope angle from Gonzalez de Vallejo and Ferrer (2011). As no bedrock exposures are present in or near the T440 area, we chose this conservative value to simulate dip slope failure. Table 2 is a summary of the silt and bedrock properties.

Table 2.

Summary of the soil and rock properties used in the SLOPE/W modeling.

View largeDownload slide

View Large

Using the SLOPE/W module, we modeled four different groundwater cases, each with and without seismic loading (except Case 4), for a total of seven scenarios. For the first case, the water table was set at 124 m, the approximate elevation of the Tanana River, as determined from the lidar data (the Tanana River runs just south of the T440 area; see Figure 2). In the second case, the water table followed the schist bedrock surface down to the elevation of the Tanana River. The third case had the water table at the ground surface elevation to simulate completely saturated conditions. For the fourth case, we began with the Tanana River at the surface elevation to imitate flood conditions and incrementally rose the water table position from Case 2 conditions until slope failure occurred. Case 4 did not have a seismic component, as the purpose of this part of the analysis was to determine the effect of water table position on slope stability. As noted earlier, Interior Alaska is a seismically active region, with a history of M 7.0 earthquake events in the last 100 years (Péwé, 1982). Understanding this potential for seismic activity, we assessed the effect of earthquake events on slope stability and applied a horizontal peak ground acceleration of 0.31g to selected scenarios (Boore and Atkinson, 2007). We used the Spencer deterministic analysis as the main method for all iterations of the model conditions and compared the factor of safety (FS) results with those obtained using the Morgenstern-Price and Janbu Generalized methods. These methods were chosen because they satisfy both the moment and force equilibrium equations and account for both shear and normal interslice forces. We chose the piezometric line option to represent pore water pressure conditions and entry and exit with specified radius tangent lines as the slip surface method, which defined failure surface entry and exit points. Figure 5b is an example of the model environment for Case 1.

Three of the samples dated were rootlets collected from 2.55 m to 3.00 m below ground surface (bgs). We initially hypothesized that the rootlets collected near the toe were buried by the landslide event. The modern dates from these samples (Table 5) indicate that the root systems of modern trees penetrate the subsurface more deeply than expected, and thus these samples were not used for landslide dating.

Table 5.

Radiocarbon dating results.

View largeDownload slide

View Large

The remaining two samples yielded older dates. The oldest date (31,210–30,981 year Cal BP) was obtained from a paleosol at 2.37 m bgs in TH06, located on the distal slope of the inner toe. The other date (7,864–7,700 year Cal BP) was obtained from a charcoal layer at 1.65 m bgs in TH08, downslope of a gully incised through the outer toe.

The T440 landslide primarily consists of wind-blown silt with rare occurrences of sandy soils and with no bedrock exposure. We interpret the silt deposits as massive unstratified loess. The soil properties and our field observations suggest that the T440 landslide is currently unfrozen, although permafrost may exist at depth. Based on its geomorphology, we interpret the T440 landslide as being younger than other paleoslides in loess that we mapped in the western FNSB. For example, the T440 head scarp and flank regions are rougher and more well-defined than the loess landslides near Fairbanks, which are characterized by arcuate toe deposits and severe gullying in the head scarp regions (see Figure 3a as an example).

We do not have enough information to make a definitive conclusion on the timing of the T440 landslide with only two radiocarbon date results. We believe the young single landslide model is more likely based on our interpretation of the T440 geomorphology. It is improbable that a subsequent landslide event would overrun the inner toe resulting in a deposit with such a similar shape (i.e., the outer toe). Regardless, these dates indicate that the T440 landslide occurred in the late Pleistocene to mid-Holocene. Additionally, the T440 surface supports two dry, vegetated, flat-bottomed gully systems that run the full length of the landslide body, as well as two shorter gullies near the left flank (see Figure 4). None of these gullies demonstrate evidence of modern water flow, such as incision or sediment deposition over vegetation. These gullies most likely formed after the landslide occurred, during the late Pleistocene to mid-Holocene when the landscape was underlain by permafrost and characterized by herb tundra (Muhs et al., 2003a), with no trees to capture water or provide cohesion through their root systems.

Permafrost in the Fairbanks region last formed during the Wisconsin glaciation beginning around 150,000 year BP for Alaska (Péwé, 1977). An Arctic climate with windy and arid conditions persisted during the Last Glacial Maximum from 30,700 to 15,700 year BP, and the landscape was characterized by herb tundra with minimal ground roughness (Muhs et al., 2003a; Finkenbinder et al., 2014). Factors that favored loess production (e.g., strong winds) were outweighed by factors that hindered loess deposition (e.g., herb tundra) (Muhs et al., 2003b). These conditions supported periods of paleosol pedogenesis, as reduced surface roughness was insufficient to trap loess. The paleosol that we sampled and dated at ∼31,000 year BP from the T440 landslide may correspond to this period of soil formation. It is also likely that the T440 landslide slope was perennially frozen (permafrost) during this period. Starting around 15,700 year BP, the climate shifted to warmer, wetter, and less windy conditions and herb tundra declined (Finkenbinder et al., 2014). The Holocene Thermal Maximum occurred 11,000 to 9,000 year BP in Alaska (Kaufman et al., 2004). The higher-than-current temperatures during that period (Kaufman et al., 2004) may have created favorable conditions for permafrost thaw. The early Holocene period from 9,400 to 8,700 year BP saw considerable changes as the climate transitioned to stable Holocene conditions, with increases in temperature and precipitation and further decline in herb tundra (Finkenbinder et al., 2014). There was a warming interval during the early to middle Holocene from 8,000 to 5,000 year BP that lowered the permafrost table and melted ice wedges (Péwé, 1977). Alder and spruce also appeared and increased during this period (Finkenbinder et al., 2014), indicating the shift from herb tundra to forest. The T440 charcoal dated at ∼8,000 year Cal BP is consistent with the timing and appearance of forest vegetation.

All modeled scenarios with a FS < 1.0 demonstrated surficial failure in the loess. This modeled failure shape and depth agrees with the T440 slope failure A-A′ profile and the lack of bedrock observed during fieldwork. One possible trigger was thawing permafrost. Results from the slope stability analysis indicate that a high water table at or near the surface (see Figure 12e–g and Table 6) results in slope failure. The dry gullies and dry soils intercepted while drilling and in stratigraphic sections suggest that a fully saturated condition is not probable for current conditions. As permafrost thawed and remained at depth, however, it would have formed an impermeable layer. This, coupled with the pore water pressures modeled in Scenario 7, may have triggered the T440 landslide. The timing of the young, single landslide model fits well with mid-Holocene warming when permafrost started to degrade.

Our slope stability analysis results also suggest that a strong seismic event is a possible trigger or contributing factor for the T440 landslide (see Figure 12b, d, and f and Table 6). Future seismic events in this region may cause new landslides, as evidenced by rockslides triggered by a powerful earthquake event in the 20th century (St. Amand, 1948).

As part of this study, we present direct shear results for silt with negligible clay and sand content. Published grain-size distributions of Alaska loess indicate the soil typically has sand and clay contents of under 5 percent (Muhs and Budahn, 2006), which agrees with our results. Only a handful of studies have analyzed the strength of silt (e.g., Higgins and Fragaszy, 1988; Derbyshire et al., 1994; Harris et al., 2008; and Zhuang et al., 2018), and the tested soils were silty sand or clayey loess. Test methods varied from ring shear tests (Derbyshire et al., 1994; Zhuang et al., 2018) to consolidated-undrained and unsaturated-consolidated-undrained triaxial tests (Higgins and Fragaszy, 1988). Our work, however, consisted of a systematic program of direct shear testing of non-plastic loess at varying moisture contents. These results represent a novel contribution to the body of literature. The lack of clay content is a distinguishing feature of loess found in Interior Alaska. It also has a low calcite content, which distinguishes it from highly calcareous North American, European, and Chinese loess (Muhs and Budahn, 2006). Our average cohesion value (3.0 kPa) agrees with the values obtained by Higgins and Fragaszy (1998) and Harris et al. (2008) of 2.1 kPa and 3.5 kPa, respectively. Our average friction angle (23.1°) is lower than other recorded values of 34.5° (Higgins and Fragaszy, 1998) and 28°–36° (Harris et al., 2008) of silty soils. The friction angle dropped when the silt was wetted, but then remained consistent with varying moisture content, demonstrating that loess loses shear strength once wetted (Derbyshire et al., 1994). Cohesion demonstrated more scatter around the average value with varying moisture content, dropping to 0 kPa at the maximum water content (37.2 percent).

Two previous studies demonstrated that vane shear strength and direct shear strength results were nearly identical for clay soils (Lefebvre et al., 1988; Hirabayashi et al., 2016). To our knowledge, no studies exist that compare direct shear and field vane shear measurements of silt. Unlike the clay soils, our direct shear strength results for silt did not demonstrate such a strong correlation to field vane shear test results. Our results indicated that the vane shear strength most closely matched the residual shear strength from laboratory tests (see Figure 8b). This is most likely due to the structure of the loess being disturbed and soil grains, particularly platy micas, being realigned during the shearing process. Although the overall correlation of the data in Figure 8b is low (R² = 0.20), we attribute these differences to the natural variation with depth in moisture content and structure of the in situ soils as compared to the controlled moisture content and remodeled structure in the laboratory samples.

We interpret the T440 landslide as an inactive flow slide in loess that occurred during the late Pleistocene to mid-Holocene based on an analysis of stratigraphy, soil testing, and geomorphology. Two possible interpretations, based on radiocarbon dating, of how the T440 landslide occurred are (1) as one event younger than ∼8,000 years with portions moving as competent blocks or (2) as two events that formed the inner (prior to ∼31,000 years ago) and outer toe deposits (after ∼8,000 years ago). Additional dating is required to determine the landslide age conclusively. Thawing permafrost during early- to mid-Holocene warming may have been a possible trigger. Slope stability modeling indicates that seismic activity also could have been a trigger for this landslide, as every model with an applied seismic load resulted in slope failure.

We provide a new research contribution through the direct shear testing of silty soils. Our work systematically tested non-plastic silt with variation in moisture content and recorded the soil strength parameters of internal friction angle and cohesion. The average values of the internal friction angle and cohesion were 23.1° and 3.0 kPa, respectively. Our results are consistent with loess losing shear strength once wetted. These values can be used for engineering design purposes for Interior Alaska silt for any gravimetric water content over 5 percent.

Here, we present the results of laboratory and field testing of the strength of silt to support this specific landslide analysis. It was not the purpose of our study to validate these methods against each other. Our results, however, indicate that a comprehensive study of the strength properties of silt, including the repeatability of field vane shear testing in loess deposits, would be beneficial. The field vane shear test is simple and inexpensive to perform; validating its use in silt against laboratory testing may provide another tool to practicing engineers in Interior Alaska. Additionally, we recommend future work exploring the effect of increasing normal stress and soil density on field vane shear testing of silt.

The landslide inventory map was not the focus of this paper; however, prior to conducting the comprehensive analysis to produce the inventory, we were unaware of the presence of the T440 landslide. Therefore, the landslide inventory represents a much-needed resource that can be used to determine which landslides in the FNSB represent potential risks to infrastructure. Future work should include determining the absolute and relative ages of other landslides in the FNSB. In addition to collecting more organic samples from flow slides, bedrock exposures on rockslides can be radiometrically dated using cosmogenic nuclide methods (Ivy-Ochs and Kober, 2008), and tephra samples should be analyzed to determine their source and age. Identifying clusters of landslides that occurred at the same time could indicate if they were triggered by one major or a series of smaller seismic events or other possible triggers. Suggested fieldwork includes trenching landslide bodies and coring local long-lived lakes to search for seismically disturbed sediments or flood deposits to identify evidence of possible trigger mechanisms. Thermal modeling with paleoclimate scenarios can be used to determine conditions necessary for thawing permafrost underlying south-facing slopes.

This research was supported by the U.S. Geological Survey, National Cooperative Geologic Mapping Program, under USGS award number G20AC00130. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the U.S. Government. We extend our thanks to the staff of the Alaska Division of Geological & Geophysical Surveys for their support during fieldwork and help with formatting the maps, and we are grateful to the homeowners who allowed us access to their private properties during the T440 fieldwork.