Because landslide regimes are likely to change in response to climate change in upcoming decades, the need for mechanistic understanding of landslide initiation and up-to-date landslide inventory data is greater than ever. We conducted surficial geologic mapping and compiled a comprehensive landslide inventory of the Denali National Park road corridor to identify geologic and geomorphic controls on landslide initiation in the Alaska Range. The supplemental geologic map refines and improves the resolution of mapping in the study area and adds emphasis on surficial units, distinguishing multiple glacial deposits, hillslope deposits, landslides, and alluvial units that were previously grouped. Results indicate that slope angle, lithology, and thawing ice-rich permafrost exert first-order controls on landslide occurrence. The majority (84%) of inventoried landslides are <0.01 km2 in area and occur most frequently on slopes with a bimodal distribution of slope angles with peaks at 18° and 28°. Of the 85 mapped landslides, a disproportionate number occurred in unconsolidated sediments and in felsic volcanic rocks. Weathering of feldspar within volcanic rocks and subsequent interactions with groundwater produced clay minerals that promote landslide initiation by impeding subsurface conductivity and reducing shear strength. Landslides also preferentially initiated within permafrost, where modeled mean decadal ground temperature is −0.2 ± 0.04 °C on average, and active layer thickness is ∼1 m. Landslides that initiated within permafrost occurred on slope angles ∼7° lower than landslides on seasonally thawed hillslopes. The bimodal distribution of slope angles indicates that there are two primary drivers of landslide failure within discontinuous permafrost zones: (1) atmospheric events (snowmelt or rainfall) that saturate the subsurface, as is commonly observed in temperate settings, and (2) shallow-angle landslides (<20° slopes) in permafrost demonstrate that permafrost and ice thaw are also important triggering mechanisms in the study region. Melting permafrost reduces substrate shear strength by lowering cohesion and friction along ice boundaries. Increased permafrost degradation associated with climate change brings heightened focus to low-angle slopes regionally as well as in high-latitude areas worldwide. Areas normally considered of low landslide potential will be more susceptible to shallow-angle landslides in the future. Our landslide inventory and analyses also suggest that landslides throughout the Alaska Range and similar climatic zones are most likely to occur where low-cohesion unconsolidated material is available or where alteration of volcanic rocks produces sufficient clay content to reduce rock and/or sediment strength. Permafrost thaw is likely to exacerbate slope instability in these materials and expand areas impacted by landslides.
1.1 Geohazards in a Changing Climate
Landslides pose a persistent hazard in high-latitude regions where permafrost is degrading rapidly (Huggel, 2009; Blais-Stevens et al., 2015b). Although models of forecasted permafrost loss are highly variable (Slater and Lawrence, 2013), regional models estimate that 40%–60% of permafrost by area will be lost by the end of the century (Pastick et al., 2015). Near Denali National Park (DNP), local monitoring suggests that permafrost temperatures are already near 0 °C (Osterkamp et al., 2009). Modeled permafrost response to climate change in DNP suggests that although 75% of the park was underlain by permafrost in the 1950s, only 1% of the park will be underlain by permafrost by the end of the twenty-first century (Panda et al., 2014). Changing precipitation patterns worldwide (Stoffel and Huggel, 2012; Intergovermental Panel on Climate Change, 2013), permafrost degradation, and the transition to freeze-thaw regimes will contribute to landsliding by increasing landslide frequency and magnitude (Patton et al., 2019). Changes to landslide regimes in Alaska and other high-latitude regions increase the uncertainty of landslide hazards assessments. As such, the need for mechanistic understanding of landslide initiation and up-to-date landslide inventory data is greater than ever.
The impact of climate change on geohazards in high-latitude regions has become a prominent research topic in the past decade (Geertsema et al., 2014; Capps et al., 2017; Coe et al., 2018; Higman et al., 2018), as land owners and managers seek to anticipate changes to hazard regimes. DNP is an informative study site to evaluate landslides in dynamic permafrost terrain with diverse lithologic, topographic, and elevation gradients. Insights from this study area improve understanding of the geomorphic processes that control landslide initiation at high latitudes.
1.2 Study Area: Geologic Context
The geology of the Alaska Range is structurally complex with regional-scale faults and folds (Gilbert, 1979; Csejtey et al., 1982; Nokleberg et al., 1994; Wilson et al., 2015). The study region is located near the boundaries of multiple geologic terranes, including the Yukon-Tanana, Wrangellia, Farewell, and McKinley terranes (Csejtey et al., 1982; Ridgway et al., 2002; Dumoulin et al., 2018). Late Jurassic–Cretaceous collision and transpression of the Wrangellia island-arc composite terrane juxtaposed 3–5 km of marine strata (the Kahiltna assemblage) with the former North American continental margin (Yukon-Tanana terrane) (Ridgway et al., 2002).
The high-relief topography of the Alaska Range is young, driven by rapid exhumation along the Denali fault system beginning 5–6 Ma (exhumation >1 mm/yr) (Fitzgerald et al., 1993; Redfield and Fitzgerald, 1993) and the development of extensive valley glaciers during the last ice age and through the Pleistocene (Yeend, 1997). Quaternary dextral transpression along the Denali fault system continues to deform and exhume the high-relief topography of the Alaska Range (Haeussler et al., 2017). Seismicity in the Alaska Range contributes to landslide hazard: the McKinley strand of the Denali fault (south of the study area) is active at a slip rate of 7–12 mm/yr (Benowitz et al., 2011). In 2002, a Mw 7.9 earthquake that occurred on the Denali fault east of the study area triggered >1500 coseismic landslides (Gorum et al., 2014), and hundreds of local earthquakes have been recorded by U.S. Geological Survey (USGS) seismic arrays in recent decades. Detailed timing of landslide initiation relative to earthquake occurrence is not known in the study area.
Until recently, the major strand of the Denali fault system within the study region, the Hines Creek fault, was considered inactive (Wahrhaftig et al., 1975; Benowitz et al., 2011). Vertically offset alluvial fans and other Holocene deposits indicate that segments of this fault may still be active, with reverse motion along a north-dipping plane generating uplift at a rate of 0.7 mm/yr in the Pleistocene (Koehler et al., 2015). Major strike-slip movement on the Hines Creek fault took place prior to 95 Ma (Wahrhaftig et al., 1975; Csejtey et al., 1982). Based on the age of the units (Late Cretaceous–Quaternary), the primary lithologic units of the study area are not offset by movement along major terrane-bounding faults.
Our study area within the Alaska Range includes a representative transect that spans lithologic units and diverse topography and is an area that is of management concern to DNP. We focus on the 1-km-wide Denali Park road corridor from miles 33–66 (Fig. 1). The road corridor crosses the variable geology within the park, including bedrock exposures of multiple lithologies as well as diverse Quaternary sediment deposits. Furthermore, the Denali Park Road transects variable weathering products, map-scale faults, and gradients of elevation, slope, and permafrost temperature. We can therefore use our inventory to evaluate the influence of lithology, topography, permafrost, and other geomorphic controls on landslide occurrence in diverse high-latitude regions. As noted by Geertsema and Clague (2011), narrow hazards mapping zones may neglect the potential for landslides to initiate upslope and then enter the area of interest. Nevertheless, the map area defined in this study characterizes the initiation points of the majority of the landslides that have impacted the road corridor.
Primary lithologic units within the map area (Fig. 1) include Jurassic basalt and metabasalt of the Nikolai Formation, Cretaceous sedimentary rocks of the Cantwell Formation (sandstone and conglomerate), Paleogene volcanic rocks of the Teklanika Formation and Mount Galen Formation (basalt, andesite, rhyolite, and tuff/tuff breccia), and Quaternary sediments (glacial, alluvial, etc.) (Gilbert, 1979; Csejtey et al., 1992; Yeend, 1997). Soils in the study area are classified as inceptisols in the steeper terrain, spodosols in coarse alluvium and glacial deposits, and gelisols in gently sloping drift and alluvium (Clark and Duffy, 2006).
Climate is highly variable over the large area of the park, particularly across the range divide that separates the northern and southern regions. Current climate in northern DNP is typical of an interior Alaska landscape, with low annual precipitation (average 38 cm), cold winters (average 5.8 °F), and mild summers (average 53 °F) (Denali National Park and Preserve Weather and Climate, National Park Service, https://www.nps.gov/dena/learn/nature/climate.htm). In the northern portion of the park, low annual precipitation results in minimal average snow cover in winter (46 cm in March in the northern part of the park) (Clark and Duffy, 2006).
Global permafrost models indicate that the study area in DNP is underlain by spatially discontinuous permafrost, meaning that permafrost exists only where local conditions are suitable (Gruber, 2012). At a local scale, the distribution of permafrost is controlled by surface material, microtopography, snow thickness, and aspect (Hasler et al., 2015). Regional permafrost models estimate that 75% of DNP was underlain by near-surface permafrost in the 1950s, and 50% of DNP was underlain by near-surface permafrost at the beginning of this century (Panda et al., 2014). Permafrost is most prevalent at high elevations and in the low-relief terrain in the northern portion of the park. In many areas of the park, permafrost includes abundant ground ice as small crystals, lenses, or seams of ice (Clark and Duffy, 2006; Yocum et al., 2006). In areas underlain by permafrost, perched water and saturated conditions are common in summer months due to the impedance of groundwater flow (Clark and Duffy, 2006; Walvoord and Kurylyk, 2016). Periglacial landforms, including polygonal ground, gelifluction lobes, and frost hummocks (palsas) are present throughout the study area (Clark and Duffy, 2006; Yocum et al., 2006).
2.1 Surficial Geologic Mapping and Landslide Inventory
We conducted field mapping of surficial geologic units at 1:24,000 scale in the 1-km-wide corridor along the Denali Park Road from miles 33–66 (Fig. 1, Plate S11), building from unit designations in the existing literature, including work by Decker (1975), Gilbert (1979), Csejtey et al. (1992), and Yeend (1997). We mapped all portions of the map area in the field in June 2017 and June 2018. During field excursions and in post-processing, we used derivations of the 2015 5-m interferometric synthetic aperture radar (IFSAR) digital elevation model (DEM) and 0.5-m 2014 DigitalGlobe imagery to inform contact placement. Notably, our investigation refines and improves the resolution of mapping in the study area and adds emphasis on surficial units, distinguishing multiple glacial deposits, hillslope deposits, landslides, and alluvial units that were previously grouped in more general unit designations.
Surficial mapping efforts included a comprehensive inventory of recent landslides that occurred within the study area before August 2018. Landslide inventory data consist of a mapped area, the initiation site (designated near the crown of the landslide), and a general failure style. We identified and delineated landslides based on preserved landslide morphology, vegetation contrasts, and/or exposed soil (Galli et al., 2008; Guzzetti et al., 2012; Burns and Mickelson, 2016). With the exception of landslides that were inaccessible due to wildlife closures, we observed all inventoried landslides on the ground to confirm landslide style. Categories of style include rockslides, earthflows, debris avalanches, debris flows, debris slides, and rotational slides (Varnes, 1978; Cruden and Varnes, 1996; Hungr et al., 2014). The inventory also includes active layer detachments, a type of translational landslide that occurs in thawed soils overlying permafrost (Lewkowicz and Harris, 2005a). We mapped very old landslide and colluvial deposits that do not have preserved landslide morphology as Quaternary Mass Wasting (Qmw) (Table 1).
We used landslide initiation points to identify key geomorphic characteristics of each landslide for comparison with the entire study area, including underlying lithology, modeled permafrost presence and depth, elevation, slope aspect, slope angle, and hillslope curvature. We derived morphometric data from the 5-m IFSAR DEM produced in 2015. To avoid noise from microtopography, we bilinearly resampled aspect data to 20 m resolution using the sequence suggested by Grohmann (2015). Curvature was calculated using the ArcGIS planform curvature tool, described in the ESRI online tool reference (ESRI, Curvature Function, ArcGIS for Desktop http://desktop.arcgis.com/en/arcmap/10.3/manage-data/raster-and-images/curvature-function.htm). Notably, curvature is highly dependent on scale, and curvature values calculated at a coarse scale do not characterize very small geomorphic features and curvature variability (Deng et al., 2007) that predict landslide initiation (Chudý et al., 2019).
2.2 Landslide Distribution in Permafrost
Using the landslide inventory described above, we compared the distribution of landslides within the map area and the distribution of modeled permafrost. Panda et al. (2014) modeled mean decadal ground temperature (°C) and active layer thickness (m) within the national park using the GIPL 1.0 model (Shiklomanov et al., 2007; Jafarov et al., 2013; Luo et al., 2014) at a spatial resolution of 28 m, although some input data sets are of coarser resolution. The model is a spatially distributed equilibrium model for calculating active layer thickness and mean decadal ground temperature at the bottom of the active layer. It uses Fourier temperature wave propagation through a material with phase transitions (ice to water) to estimate thermal offset (temperature gradient) of the active layer or seasonally frozen layer. Panda et al. (2014) calculated permafrost characteristics using climate data, ecotype, soil characteristics, and snow characteristics to represent permafrost conditions during the 2001–2010 decade. Negative active layer thickness values indicate that permafrost is not present, and instead describes seasonal frost thickness. Reported accuracy of the model used in this study is ±0.2–0.4 °C for the mean ground temperatures and ±0.1–0.3 m for the active layer thickness. The model has not been extensively tested with ground-based measurements but rather characterizes patterns of permafrost distribution within the park using best available knowledge. Field-based identification of permafrost features in the study area aligns with the patterns of modeled permafrost (Yocum et al., 2006; this study). Depth to ice (measured with a manual probe) and surface temperature near Stony Pass, Eielson Bluffs, and east of the Toklat River corroborate the model (Patton et al., 2018). Individual values are nevertheless intended to be used in a comparative context in this study, rather than as absolute measurements of permafrost characteristics. Using the modeled permafrost extent developed for the previous decade (2001–2010), we characterized both the ground surface temperature and active layer thickness (or seasonal frost thickness where permafrost is not present) at landslide sites and within the map area.
2.3 Clay Sample Collection and Analysis
Clay minerals influence the geotechnical properties of slope materials where in situ weathering or hydrothermal alteration of bedrock promotes their development (Ikari and Kopf, 2011). We characterized the mineralogical composition of clay-rich sediment in landslides to evaluate the mechanism driving lithologic control of landslide initiation. We collected 13 clay samples from mapped landslides where a significant volume of clay material was exposed in the scarp or within the landslide deposit (Fig. 1). We collected samples from all landslides we investigated in person that exposed clay-rich material in the scarp or deposit (∼80% or more clay as determined in the field). This sampling strategy ensures that samples represent the clay weathering history in units that are susceptible to landsliding and clay formation. We then analyzed clay for mineral assemblage and chemical composition using a TerraSpec Halo multispectral mineral analyzer and mineral identification software.
3.1 Surficial Geologic Mapping
Surficial geologic mapping throughout the study area refined contacts and map resolution along the DNP road corridor (Plate S1 [footnote 1]; Figs. 2 and 3). In comparison with previous small-scale mapping efforts, this work distinguishes multiple units that were previously unmapped, grouped in more general units, or mapped in less detail. For example, we identify multiple glacial, alluvial, and hillslope deposits, including delineation of the currently active braid plain (Qbp), colluvium (Qcol), pediment (Qp), and relict landslide deposits (Qmw) (Table 1) that were previously mapped together as alluvium and/or colluvium (Yeend, 1997). Furthermore, the geologic map differentiates exposures of the Mount Galen volcanics and Teklanika volcanic units by lithology, including basalt, andesite, rhyolite, tuff, and tuff breccia where surface exposure of bedrock is adequate. Mapped lithologies are described in summary in Table 1. The fault mapped near Eielson Bluffs (Fig. 2) cuts across topography and Quaternary sediments with a vertical topographic offset of ∼20 m. The complete 1:24,000 map is available as Plate S1 in the Supplemental Material (footnote 1).
3.2 Landslide Inventory
Landslides occur throughout the study area, although clusters of landslides occur in areas of relatively greater relief, including (from west to east) Eielson Bluffs, Stony Pass, the Toklat and East Fork River valleys, and Igloo Canyon (Fig. 1). A total of 90 modern landslides were mapped in the study area, with 85 initiating within the map area. The majority of landslides in the study area are small in total surface area (Fig. 4); median landslide area is 0.20 km2; 84% of inventoried landslides are less than 0.01 km2; and 94% of inventoried landslides are less than 0.05 km2. The distribution of landslide areas roughly corresponds to that expected for landslide inventories (Malamud et al., 2004). Planform curvature is similar between the landslide population and the map area, with mean curvature values near zero (0.07 ± 0.09 and 0.02 ± 0.0005, respectively, ± standard error), indicating a similar distribution of concave, convex, and planar hillslopes in both data sets at the scale of the 5-m DEM. This analysis does not characterize very small geomorphic features and curvature variability that may influence landslide initiation (Deng et al., 2007; Chudý et al., 2019).
Landslides initiate on high-elevation slopes of all aspects (Figs. 5 and 6). The study area is biased toward south- and east-facing hillslopes because the park road follows the south-facing hillslopes of the Toklat, Stony Creek, and Thoroughfare river valleys (Fig. 6). Landslide distributions are generally similar, following the same trend of bias toward south-facing slopes with few landslides on east-facing hillslopes. There is a slight bias toward landslide occurrence on north- and northwest-facing slopes relative to the map area as a whole. The distribution of slope angle of landslide initiation is bimodal, with peaks in landslide occurrence at ∼18° and 28° (Fig. 5B). The Hartigans’ dip test of unimodality (Hartigan and Hartigan, 1985) demonstrates the significance of the multimodality of landslide slopes, with a p-value of 0.03 (Fig. 5C). Calculation of this dip-test statistic does not depend on bin size.
Landslides occurred throughout the study area following multiple failure styles and in almost all of the mapped units (Fig. 5C). A spatially disproportionate majority of inventoried landslides (51 of 85) initiated in unconsolidated sediments (Fig. 5C; Table 2), including various glacial deposits (Qg), relict landslide deposits (Qmw), and colluvium (Qcol) (Fig. 5C). Landslides in bedrock occurred predominantly in felsic volcanic rocks of the Teklanika volcanics (Tt) and Mount Galen volcanics (Tmg). Landslides in those units also account for a disproportionately large number of landslides (18 of 85) relative to the map area underlain by felsic volcanics (Fig. 5C). The majority of landslides are rotational or translational debris slides, active layer detachments, and debris avalanches (Fig. 7; Table 2). Landslides in Pleistocene glacial sediments (Qg), relict landslide deposits (Qmw), and colluvium (Qcol) occur dominantly as rotational or translational debris slides or as active layer detachments within thawing permafrost. Landslides in the felsic volcanic rocks of the Teklanika volcanics (Tt) and Mount Galen volcanics (Tmg) occur dominantly as debris or rock slides or debris avalanches.
3.3 Landslide Distribution in Permafrost
As modeled by Panda et al. (2014), permafrost extended across 62% of the map area in the previous decade, with a mean surface temperature of −0.09 ± 0.001 °C and a median active layer thickness of 0.94 m in the study area (range −1.6–1.9 m). Landslides mapped in this study occurred preferentially in terrain underlain by permafrost, as modeled by Panda et al. (2014) (Figs. 8 and 9); 81% of landslides initiated in permafrost, with a mean surface temperature of −0.23 ± 0.04 °C and a median active layer thickness of 0.96 m (range −1.4–1.9 m) at initiation sites. A disproportionate majority of landslides occurred where modeled active layer thickness is ∼1 m (Fig. 9). Slope angles of landslides that occur within permafrost terrain are lower than slope angles in landslides where seasonal thaw occurs (Fig. 10). Median slope angle of landslides in permafrost is 20.1°, compared to a median 27.1° slope angle of landslides on slopes where complete seasonal thaw occurs.
3.4 Clay Composition and Weathering History
Clay samples collected from modern landslide deposits (Fig. 11) include primarily 2:1 smectite minerals (montmorillonite and beidellite) and vermiculite (Table 3). Almost all of the minerals present in clay samples are typical of hydrothermal alteration of volcanic rocks and subsequent interaction with groundwater. The presence of montmorillonite and beidellite (smectite minerals) in most samples indicates formation of clay minerals in alkaline conditions with persistent groundwater (Anthony et al., 2001). A common weathering product of tuff and ash, montmorillonite (samples EF1, EF9, PR1, E16, I8, C1, and EF5) may form within glassy felsic flows or minor ash beds along bedding (flow) contacts, and vermiculite (samples EF1 and E16) is common at the contact between felsic and mafic rocks (Anthony et al., 2001). Ashy flow contacts (Frothingham and Capps, 2018) and mafic-felsic contacts (Gilbert, 1979; this study) are both typical of the Teklanika volcanics in the study area. The presence of rectorite and halloysite (samples BC1, I4, HP2, and I9) is consistent with clay minerals that form due to the weathering of feldspars (especially potassic feldspars) (Anthony et al., 2001).
4.1 Geologic Mapping and Landslide Inventory
The surficial geologic map produced in this study improves the resolution of available geologic data along the DNP road corridor, providing a valuable resource for understanding landslide controls in broad areas of diverse lithology and areas of permafrost degradation. The extensive landslide inventory conducted within the map area provides a census of all modern landslides that occurred before fieldwork was completed in August 2018. Characterizing this population is therefore representative of the map area and typifies spatial patterns of landslide initiation in the northern Alaska Range and other high-latitude mountain ranges with similar climate, topography, and diverse underlying geology. Comprehensive information about landslide location, surface area, and type at a known time (August 2018) provides a baseline for future landslide inventories and time-series analysis. Such monitoring will provide valuable insight into the many mechanisms by which climate change influences landslide initiation (Gariano and Guzzetti, 2016; Patton et al., 2019).
Landslides occurred throughout the study area in almost all of the mapped units (Fig. 5C), except those that are inherently characterized by very shallow slope angles (e.g., alluvial sediment and pediment). Poorly sorted, unconsolidated sediments on hillslopes (Qg, Qmw, and Qcol) accounted for the vast majority of landslides in the study area (51 of 85). Landslides in bedrock occurred predominantly in feldspar-rich volcanic rocks (18 of 85) of the Teklanika volcanics (Tt) and Mount Galen volcanics (Tmg). This pattern suggests that landslides throughout the Alaska Range and similar climatic zones are most likely to occur where low-cohesion unconsolidated material is available or where alteration of volcanic rocks produces sufficient clay content to reduce rock and/or soil strength, as discussed below (Behnsen and Faulkner, 2012; Torrance, 2014; Isobe and Torii, 2016). Notably, many of the unconsolidated units in the study area contain high concentrations of volcanic detritus and clay minerals.
Within the DNP road corridor, landslide distribution also indicates several important topographic controls on slope failure. Landslides occurred on all slope aspects throughout the study area, primarily at relatively high elevations (>1050 m) where topographic relief is greatest (Figs. 5 and 6). The majority (84%) of landslides within the study area are less than 0.01 km2 in area, indicating that small, frequent hazards are a significant source of concern in the region. Larger landslides, though present within the map area and beyond, are likely infrequent (Wolman and Miller, 1960; Guthrie and Evans, 2007).
Multiple landslide styles occur in the study area, including debris flows, rockslides, rotational or translational debris slides, active layer detachments, etc. The abundance of multiple failure types emphasizes the importance of considering diverse triggering mechanisms for landslides in high-latitude environments. Similarly, the bimodal distribution of slope angles in the study area indicates that there are two populations of landslides within discontinuous permafrost zones. On steep hillslopes (≥20°), landslides occur according to well-established patterns, particularly when rainfall or snowmelt increases soil saturation beyond threshold limits (e.g., Borga et al., 2014) and/or seismic activity reduces cohesion (e.g., Kargel et al., 2016). Landslides that occur on shallow hillslopes require substantially lower shear strength due to low cohesion or friction (Milledge et al., 2014). Many of the shallow-angle landslides mapped in DNP are active layer detachments (11 of the landslides inventoried) in unconsolidated material; these detachments typically occur where melting of ice lenses at the base of the active layer reduce shear strength and allow landslides to occur on very shallow (10°–20°) hillslopes (Lewkowicz and Harris, 2005b; Lewkowicz, 2007; Blais-Stevens et al., 2015b). As discussed below, permafrost and permafrost thaw increase the susceptibility of shallow hillslopes to landsliding.
Assuming that the map area is representative of other high-latitude regions, landsliding in discontinuous permafrost may be separated into two categories: atmospherically driven versus those driven by ice and permafrost thaw. Reviews of landslide response to climate change predict that increased rainfall intensity will increase landslide frequency in upcoming decades (Stoffel et al., 2014; Gariano and Guzzetti, 2016) and that ice-driven landslides will be more frequent in upcoming decades (Patton et al., 2019), resulting in frequent small landslides. Due to the typically small size and relatively slow flow speeds of active layer detachments (Lewkowicz, 2007), these types of landslides often result in infrastructure damage but do not typically pose serious hazards to human safety on a large scale. Loss of permafrost, which may allow more deep-seated landslides to initiate (Harris et al., 2009; Keiler et al., 2010), is predicted to increase the magnitude of landslides in this and other regions experiencing permafrost thaw (Patton et al., 2019). The combined effect of permafrost loss and rainfall-driven landsliding on steep slopes is likely to result in larger and more hazardous landslides in high-latitude mountains (Geertsema et al., 2006; Coe et al., 2018) such as the Alaska Range, posing significant concerns for human safety and infrastructure stability.
4.2 Landslide Distribution in Permafrost
A disproportionate number of landslides (81%) initiated in permafrost terrain relative to the 62% of the map area underlain by permafrost (mean decadal surface temperature <0 °C). This discrepancy demonstrates the influence of (ice-rich) permafrost in landslide initiation. Furthermore, median slope angles where landslides occur are ∼7° lower in permafrost terrain, demonstrating the ability for ice-rich permafrost to facilitate landslide initiation on shallow hillslopes due to perched groundwater and elevated pore pressure in the active layer (Walvoord and Kurylyk, 2016) and increases in pore pressure during ice melt (Patton et al., 2019). Furthermore, low cohesion and friction along permafrost boundaries can also reduce shear strength of a slope and facilitate landslide initiation (Haeberli et al., 1997; Huggel, 2009). The susceptibility of permafrost slopes to landsliding suggests that the bimodal distribution of slope angles observed at landslide sites (Fig. 5) results from a population of shallow-angle landslides (∼18°) that occur where permafrost and/or permafrost thaw reduce shear strength.
Landslides in the study area typically occur where depth to permafrost is 0.8–1.1 m, which is consistent with active layer detachments observed on Ellesmere Island, Canada, where landslides slid along the thaw boundary at 0.6–1.2 m depth (Lewkowicz, 2007). Parker et al. (2016) estimated that 20°–25° slopes in the Southern Appalachians would be characterized by a minimum critical soil depth of ∼0.5–1 m of soil and/or colluvium necessary for landsliding to occur under saturated conditions. Some, but not all, landslides observed in our study are primarily within or above the 20°–25° range (Fig. 10). The occurrence of multiple ∼1 m depth landslides on slope angles <20°, however, indicates that landslide initiation in permafrost regions is controlled by reduced thresholds of low-cohesion (i.e., thawed) soil and/or colluvium depth relative to temperate climates. We postulate that, even on shallow slopes (10°–20°), an 0.8–1 m active layer thickness is a threshold depth of material to generate sufficient shear force along the low-cohesion permafrost boundary.
It is important to note that the permafrost parameters used in this comparison characterize the previous decade (2001–2010). Based on partial re-vegetation of some of the surveyed landslides, the time frame of the DNP permafrost model likely describes active layer depth and temperature of the study area when many of the inventoried landslides initiated; landslides in the high arctic may take >50 years to recover original plant communities (Cannone et al., 2010). Some landslides, however, have occurred several years after the time frame of this model. For example, eyewitness accounts constrain initiation of multiple inventoried landslides within the past five years, including the Eagle’s Nest landslide (Fig. 4C), which occurred in August 2016, and the Ptarmigan landslide (Fig. 4E), which occurred in July 2016. Although mean ground temperature at landslide initiation sites was slightly lower than the regional average (−0.23 ± 0.04 °C vs. −0.09 ± 0.001 °C, respectively), both mean temperatures are very near the 0 °C freezing threshold, indicating that in the previous decade, permafrost was already unstable. Given this context, it is likely that ongoing permafrost thaw has increased landslide initiation within the study area, allowing landslides to occur along partially melted ice boundaries.
4.3 Clay Formation and Slope Stability
Montmorillonite, beidellite, and vermiculite are the most prevalent clay minerals present in the 13 samples. Sample I8, derived from basalt, also contains measurable Fe/Mg chlorite. These specific mineral assemblages indicate a saturated, alkaline weathering environment, consistent with the modern subsurface characteristics of discontinuous permafrost zones (Walvoord and Kurylyk, 2016). Although the effect of climate on clay composition generally overwhelms other factors (Velde and Meunier, 2008), the minerals present also indicate generalized characteristics of the parent material. The primary minerals identified in our samples are typical of weathering feldspars and felsic volcanic rocks; contacts between felsic and mafic volcanic rocks; and tuff or glassy felsic flow deposits (Anthony et al., 2001). These environments are consistent with the sequences of volcanic rocks in both the Mount Galen and Teklanika formations, where felsic lithologies and complex contact relationships provide ample opportunity for montmorillonite, vermiculite, and other clay minerals to develop.
Clay weathering products are an important control on landslide susceptibility by impeding groundwater flow and increasing local pore pressure (Badger and Ignazio, 2018) and reducing rock strength (Bittelli et al., 2012; Borrelli and Gullà, 2017). Expansive clays (smectites and vermiculites, as identified in our samples) may even trigger landslides in response to changes in soil saturation (Velde and Meunier, 2008; Bittelli et al., 2012; Isobe and Torii, 2016). In most crustal rocks, coefficients of friction range from 0.6 to 0.85 (Byerlee, 1978). Empirical data suggest that the coefficients of friction of most clay minerals are much lower, ranging from 0.12 (montmorillonite) to 0.38 (illite) when samples are wet (Behnsen and Faulkner, 2012). This reduction in friction can dramatically reduce slope stability where alteration increases clay content in bedrock. At a mechanistic level, weathering of felsic volcanic rocks and the production of clay minerals have the potential to reduce shear strength of bedrock slopes below thresholds of stability. The high clay content in the sampled landslide deposits, including abundant low-strength, expanding clay minerals (e.g., montmorillonite), indicates that the disproportionate number of landslides that initiated in felsic volcanic units in DNP at least partially results from increased clay content due to weathering of feldspars and volcanic glass. Permafrost thaw in clay-rich substrate will exacerbate the existing susceptibility of clay-rich slopes to slope failure.
Weathered volcanic sequences in other areas of the world are also susceptible to landsliding, particularly where clay weathering products are abundant. Hydrothermal alteration of volcanic rocks in an active volcanic complex in Ecuador contributes to the initiation of clay-rich debris flows (Detienne et al., 2017). In the Southern Rocky Mountains of Colorado, weathered volcanic rhyolite tuff increases debris-flow susceptibility in the Upper Colorado River Valley (Grimsley et al., 2016). In 2005, a complex landslide (Sutherland landslide) in British Columbia occurred in a lithologic setting that is similar to the Mount Galen and Teklanika formations, where relatively resistant Eocene basalt overlies a weaker felsic volcaniclastic sequence (Blais-Stevens et al., 2015a). Clay samples collected from the Sutherland landslide included abundant smectite-group minerals, including montmorillonite, located stratigraphically above saturated volcaniclastics. Blais-Stevens et al. (2015a) suggest that weathered ash and/or feldspar provided the primary source material for the clay minerals and that expandable clays formed the primary slip plane of the Sutherland landslide. The global occurrence of landslides in weathered felsic volcanic units suggests that clay alteration products contribute to slope susceptibility in diverse physiographic and climatic settings.
With expected rapid climate warming and subsequent permafrost thaw, landslide hazards pose an ongoing management challenge within DNP. Our geologic map provides a framework for monitoring and investigation of geomorphic process in the Alaska Range. It describes surficial geology at high resolution (1:24,000 scale) and documents detailed relationships between Cretaceous sedimentary rocks, volcanic units, and Quaternary sediments; structural boundaries; and geologic history. The comprehensive inventory of modern landslides in the road corridor provides a baseline data set for comparison of future landslides and time-series analysis of landslide frequency in the Alaska Range.
Within the study area, multiple landslide styles occur primarily on high-elevation hillslopes on all slope aspects. The bimodal distribution of slope angles (Fig. 5B) where landslides initiate in the study area indicates two primary failure mechanisms in discontinuous permafrost regions, including atmospheric events and ice-rich permafrost thaw. Landslides in the study area preferentially initiate in areas underlain by permafrost, and landslides in permafrost terrain occurred on slope angles ∼7° shallower than landslides on seasonally thawed hillslopes. Modeled mean decadal ground surface temperatures were very near 0 °C for the previous decade. Thaw of sensitive permafrost provides a mechanism to reduce cohesion and allow landslides to develop on relatively low slope angles, particularly where a threshold thickness of ∼1 m of unconsolidated material is present in the active layer. Shallow-angle landslides (<20° slopes) in permafrost and the abundance of active layer detachments (11 of 85 landslides) demonstrate that ice-rich permafrost thaw is an important triggering mechanism in high-latitude regions. Melting permafrost reduces shear strength by lowering cohesion and friction values along ice boundaries and temporarily increases pore pressure. Increased permafrost degradation associated with climate change will make this and other mountainous areas more susceptible to shallow-angle landslides.
Unconsolidated units, including colluvium and glacial deposits, generate the largest numbers of landslides, and generate more landslides than would be predicted by area alone. Felsic volcanic rocks also generate a disproportionate number of landslides due to the significant weathering of feldspars and glassy matrix to clay minerals (particularly montmorillonite, vermiculite, and beidellite). The presence of clay minerals promotes landslide initiation by impeding groundwater flow and increasing local pore pressure, reducing rock strength and friction. Montmorillonite, in particular, has a very low friction coefficient when wet (0.12) (Behnsen and Faulkner, 2012). The presence of swelling clays may even trigger landslides in response to changing saturation. The disproportionate number of landslides that initiated in felsic volcanic units in DNP at least partially results from high concentration of clay content (>80% in some areas) due to weathering of feldspars and volcanic glass. Permafrost thaw in clay-rich substrate will likely exacerbate the existing susceptibility of clay-rich slopes to landslide occurrence as climate continues to warm in high-latitude regions.
Funding to support this research was provided by Colorado State University, Denali National Park and Preserve, EDMAP (a component of the U.S. Geological Survey National Cooperative Geologic Mapping Program) (award no. G18AC00093), the Geological Society of America, the Quaternary Geology and Geomorphology Division of GSA, and the Sigma Xi Research Society. Additionally, we would like to thank individuals who have provided technical support or contributed to the major ideas expressed in this paper, including Michael Frothingham, Cal Ruleman, Russell Rosenberg, Nick Virgil, Sawyer Finley, Jerry Magloughlin, John Ridley, and Britta Schroeder. We also thank Joshua Roering, Jim Finley, Marten Geertsema, and an anonymous reviewer for their thoughtful feedback, which helped to clarify the conclusions of this study.