Earth and Planetary Science Letters Catastrophic emplacement of giant landslides aided by thermal decomposition: Heart Mountain, Wyoming

that closely resemble microstructures in the basal slip zone of the natural landslide. Our experimental results are consistent with an emplacement mechanism whereby catastrophic slip was aided by carbonate decomposition and release of CO 2 , allowing the huge upper plate rock mass to slide over a ‘cushion’ of pressurized material.

Although long-runout landslides are known to occur on highangle surfaces (e.g. Brian Dade and Huppert, 1998), one of the long-standing paradoxes in geomechanics is how large landslides are able to move great distances on low-angle surfaces when frictional resistance demands that they not. There have been a number of hypotheses to explain this phenomenon (e.g. Pudasaini and Miller, 2013    Limestone. The gray horizontal layer is the ∼3 m-thick basal layer overlying a thin sliver of the lower plate Bighorn Dolomite and Cambrian Snowy Range Formation. The dark trachyandesite peak in the upper right is an Eocene stock of the Absaroka Group Volcanics. Marbleization is due to stock igneous intrusion prior to slide emplacement. Modified from Anders et al. (2010). (c) A clastic dike/injectite at Silvergate, Montana (light colored injectite is intruded into dark Eocene Absaroka Volcanic rocks). This dike is located within a meter of the basal contact. At this location carbonized Eocene wood was found in the dike (see Anders et al., 2010, their Fig. 3f). (d) Cut slab of clastic dike/injectite from Silvergate, Montana location. of the Heart Mountain landslide include hydroplaning on cognate waters (Voight, 1973), injection of pressurized volcanic gases at the slide's base (Beutner and Gerbi, 2005;Hughes, 1970), earthquake acoustic fluidization (Melosh, 1983), granular dynamic fluidization (Anders et al., 2000), hydrothermal overpressuring (Aharonov and Anders, 2006;Goren et al., 2010aGoren et al., , 2010b and, more recently, overpressuring from generation of CO 2 gas caused by thermal decomposition of the basal carbonates (Anders et al., 2010;Beutner and Gerbi, 2005).
Here we present experimental results that support the unique role of carbonate decomposition in the long-runout of large terres-trial landslides. We show that hypotheses of landslide fluidization by CO 2 gas generation, hitherto speculated at, are well supported by a combination of our experimental results and field observations.

Natural observations
One of the best exposures of the basal section of the HM landslide is at White Mountain (Figs. 1,2b). At this locality a Paleozoic section of marbleized Ordovician Bighorn Dolomite to Mississippian Madison Limestone in the upper plate was emplaced (Fig. 1)  onto Bighorn Dolomite and mudstones of the Cambrian Snowy Range Formation in the lower plate. The basal slip layer at this locality is up to 3.5 m thick and lies above an irregular contact that delimits the top of the lower plate. Clastic dykes that have previously been interpreted as injectites (Anders et al., 2010;Craddock et al., 2012) are commonly found originating within the basal slip layer and extending sub-vertically into the upper plate  for distances of up to 120 m (Anders et al., 2010;Craddock et al., 2012). The composition of the basal slip layer varies considerably along the length of the landslide depending on local protolith, but a significant proportion of carbonate material is generally found at every location previously examined (Anders et al., 2010). At White Mountain it comprises millimeter to centimeter sized well-rounded clasts of limestone and dolomite both ranging from 20-100 μm in diameter, less abundant angular fragments of volcanic glass up to 1 cm in diameter, and other types of volcanic clast suspended in a microcrystalline carbonate matrix ( Fig. 3a-d).
Field-emission scanning electron microscope (SEM) and Energy-Dispersive X-Ray spectroscopy (EDS) observations indicate that the carbonate matrix is predominantly composed of calcite grains 50-100 μm in size displaying well-defined "polygonal" grain shapes, with grain boundaries regularly forming triple junctions ( Fig. 3d-f). Dolomite clasts dispersed throughout the matrix contain an outer concentric layer up to 100 μm wide of fine-grained (<5 μm) mottled calcite (up to 70% of the entire clast) (Fig. 3d, e). The margins of the concentric layer of calcite are highly irregular and in places the calcite are observed to infill cracks within the large dolomite grain core, indicating that it may be a sec-ondary generation of calcite (Fig. 3e). On the outer boundary of the dolomite grains, aggregates of much smaller euhedral tabular grains (<10 μm in size) are commonly found (Fig. 3d, e). EDS analysis shows that many of these grains are composed of pure forsterite ( Fig. 3d-g). Although previously identified (Anders et al., 2010), there was no direct evidence that forsterite was generated during the sliding and that it might be associated with the decarbonation of dolomite. However, here we found near-perfect basal sections (Fig. 3e) suggesting the forsterite grains may have grown in-situ on the rims of the dolomite grains. Euhedral grains of forsterite <10 μm in size are also found distributed throughout the calcite matrix, commonly located at grain triple junctions (Fig. 3e). Where forsterite grains have higher density within the matrix, calcite grains tend to be significantly reduced in size, suggesting pinning may locally restrict the calcite grain growth (e.g. Herwegh et al., 2008) and that both calcite and forsterite growth are synchronous.

Experimental methods
In an attempt to reproduce some of the distinct microstructures seen in the basal slip zone of the landslide at White Mountain, we performed ten low-to-high velocity friction experiments on powdered carbonate rocks collected from the landslide. The carbonates were sheared using a sample holder for incohesive rocks within SHIVA (slow-to high-velocity rotary-shear friction apparatus) at the Istituto Nazionale di Geofisica e Vulcanologia (INGV), Rome, Italy (Di Toro et al., 2010;Smith et al., 2013). Powders were prepared by crushing and sieving (to a particle size of <250 μm) initially intact host rocks of Madison Limestone and Bighorn Dolomite collected from White Mountain (Figs. 1a, 2b). As the basal layer of the landslide at this locality is composed of comminuted pieces of both the upper and lower plates, three types of gouge were used in the experiments; two end members of 100 wt% Bighorn Dolomite and 100 wt% Madison Limestone, and an intermediate mix of 50 wt% dolomite and 50 wt% limestone. Powdered samples were sheared for up to 3 m displacement at velocities between 0.01 m s −1 and 1 m s −1 at constant normal stresses up to 10 MPa, corresponding to a depth of around 450 m. Normal stresses were limited in this case due to minor gouge extrusion at displacements approaching 3 m. Previous estimates put the upper plate thickness at 2-4 km prior to landslide motion (Aharonov and Anders, 2006), so our results should be considered as a lower bound for representing conditions during the movement phase. Experiments were performed in a closed environmental chamber containing a mass spectrometer to detect CO 2 output during the experiments Violay et al., 2013). CO 2 gas emission was monitored using an Om-niStar™ GSD 301 O mass spectrometer, for general gas analysis at atmosphere pressure. The experimentally deformed gouges were recovered and impregnated with epoxy, and thin sections were cut perpendicular to the gouge layers and approximately parallel to the slip direction to conduct geochemical and microstructural investigations. A JEOL 6500 field-emission scanning electron microscope at the Istituto Nazionale di Geofisica e Vulcanologia (Rome) was used to image microstructures. Chemical analysis was carried out using a JEOL JXA8100 Superprobe with an Oxford Instruments INCA system (EDS) and an accelerating voltage of 15 kV, current of 1 μA and a beam diameter of 1 μm.

Results
Four representative experiments are shown in Fig. 4a. At a slip rate of 0.01 m s −1 , the friction coefficient (μ, shear stress divided by normal stress) remained high during the experiments (s638, μ = 0.7-0.8, Fig. 4a) and no CO 2 output was detected (Fig. 4b).
At slip rates of 1 m s −1 , all three gouge types (pure calcite, pure dolomite, 50/50 dol./cal.) show peak friction coefficients of ∼0.6 followed by dynamic weakening to reach minimum friction coefficients as low as 0.25 (Fig. 4a). This was broadly concomitant with a pronounced spike in CO 2 emissions (increasing from background levels in the lab. of 360-385 ppm to maximum values of 3580 ppm) that were registered by the mass spectrometer within 6-10 seconds of slip initiation (Fig. 4b).
Two very low-displacement tests (8.5 mm and 0.1 mm) were performed to constrain whether CO 2 is generated during the earliest stages of slip (Fig. 5). The target slip velocity in these experiments was 6.5 m s −1 but because the displacements were so small, the peak velocities reached (with acceleration and deceleration rates of 90 m s −2 ) were ca. 0.7 m s −1 (for the 8.5 mm test; Fig. 5a) and ca. 0.06 m s −1 (for the 0.1 mm test; Fig. 5b). For 8.5 mm of total slip, a significant increase of around 80 ppm CO 2 was detected above background variations (360-385 ppm), and for just 0.1 mm of total slip a smaller increase of up to 20 ppm was detected (Fig. 5c).

Microstructural observations
Microstructural observations indicate that displacement at slip rates of 0.01 m s −1 was accommodated by distributed cataclastic deformation (Fig. 6), formation of Riedel shears (primarily of R -type following Logan et al., 1979) and grain size reduction across the entire 1 mm thick gouge layer. Dolomite and calcite grains deformed at this slip rate are typically rounded and range in size from several hundred nm to 20 μm (Fig. 6b, c). Longer  Fig. 7a, b) defined by a marked reduction in grain size compared to the bulk gouge layer (Fig. 7b, e, f). Secondary SEM observations of the discrete principal slip surface cutting the PSZ show small (<500 nm in size), tightly-packed and polygonal calcite grains that form a surface "pavement" with negligible porosity (Fig. 7h). Adjacent to the PSZ is a layer ∼100 μm thick (Fig. 7a, b) in which the overall size and distribution of dolomite clasts is similar to that in the starting materials, indicating that the layer was not highly strained. However, the layer contains two distinct microstructures not found elsewhere in the gouge layers; (1) a crystalline calcite matrix showing roughly polygonal, interlocking calcite grains <10 μm in size (Fig. 7c, d, g) and (2) dolomite grains containing vesicular rims and planar trails of vesicles (Fig. 7c, d, g), typical of decarbonation. We refer to this zone as the decarbonated dolomite zone (DDZ). All of the dolomite clasts within the DDZ have outer rims with a vesicular texture. The Mg content in dolomite clasts increases towards the principal slip zone (Fig. 7b, Fig. 8) indicating a gradient in both temperature and the amount of decarbonated material. In the DDZ adjacent to the principal slip zone the dolomite clasts are often completely engulfed by the vesicular texture, forming highly porous grains surrounded by the crystalline calcite matrix (Fig. 7c, d).

Estimates of temperature increase within experimental slip zones
To estimate the temperature distribution across the gouge layer produced by frictional heat diffusing away from the principal slip surface in experiment s458, we calculated the temperature increase T with time within the 10 μm thick experimental PSZ and adjacent gouge layer using a simplified solution for heat wave propagation in one dimension which accounts for heat loss during  the endothermic decarbonation reaction (Brantut et al., 2011;Sulem and Famin, 2009). Temperature is estimated by solving the one dimensional heat diffusion equation in the presence of a heat source rate τ (t)V (t) (frictional power, where V = slip rate, and τ = shear stress) and a heat sink rate φ(t) due to the latent heat of decarbonation of dolomite and calcite. Experiments described in our study demonstrate that decomposition occurs within a zone c. 100 μm wide. The net heat rate available at any given time t is hence: The temperature increase T (t, x) within the experimental slip zone is estimated using the classical solution for 1D diffusion with imposed heat flow at the boundary (Carslaw and Jaeger, 1959). Assuming frictional power is generated within a very thin slip zone (we neglect the initial strain localization phase): where x is the distance away from the slip zone, ρ = 2800 kg m −3 is mass density, κ = 1.74 × 10 −6 m 2 s −1 is thermal diffusivity and  Fondriest et al., 2013), is reached at around 75 cm of slip (Fig. 9a). Release of CO 2 detected for slips as small as 0.1 and 0.85 mm in the experiments (Fig. 5c) is probably due to grain crushing inducing localized high temperatures by flash heating, while the bulk temperature of the sliding interface is still much lower. According to the relatively high temperatures obtained without heat sinks, decarbonation should have occurred up to about 500 μm from the slip zone, during the last half-second of slip. However, microstructural observations show that decomposition occurred only within about 100 μm of the slip zone (Fig. 8). We try to match this experimental observation by including a heat sink due to the latent heat of decarbonation.

Case with heat sink due to decarbonation
Here we account for the heat loss by decarbonation. For simplicity, we assume a priori that such a heat sink is consistent in the region immediately adjacent to the slip zone, and consider  that both the main frictional heat sources and heat sinks are located within a layer of negligible thickness so that we can locate the net heat source on the plane x = 0 at the center of the PSZ for the solution of the problem. This approximation is justified by the experimental observation that only negligible reactions took place outside a region 100 μm thick adjacent to the principal slip surface.
We assume an Arrhenius thermal dependence of the reaction kinetics. We compute the rate of mass decomposed as a function of time and the corresponding rate of heat loss. After Sulem and Famin (2009), we express the reaction rate (in mass per unit volume m per unit time t) as: (2) where R = 8.31 J K −1 mol −1 is the gas constant, E a = 319 × 10 3 J mol −1 is the activation energy and A = 2.95 × 10 15 s −1 is the pre-exponential factor. For simplicity, we assume that all forms of decarbonation of calcite and dolomite will not differ significantly in their energy absorption and describe a single reaction with the above values of E a and A, and L = 3190 × 10 3 J kg −1 for latent heat (Sulem and Famin, 2009). n(t) is the proportion of carbonated material which has been decarbonated at time t and α is the initial fraction of carbonated material in the bulk rock which we approximate by α = 1.
We may write the heat sink rate per unit fault area and the proportion of mass loss per unit time per unit fault area, respectively, as: The larger the thickness h of rock affected by decarbonation around the slip zone, the smaller the net heat and its penetration distance. Different values of h were tested for consistency: the penetration distance of the 600 • C isotherm resulting from the computation should be close to the h value used in the computation. For a layer of about h = 130 μm around the slip zone, we do obtain a heat penetration of the same order (see Fig. 9b). Such a value is also consistent with the microstructural observations. Fig. 9b shows a clear "thermal buffering effect" of the reaction, as noted in Sulem andFamin (2009) andBrantut et al. (2011), which starts when temperatures approach 600 • C (decarbonation triggering) and is fully effective at about 800 • C where decarbonation is fast.
Inserting the value of φ (3) into (1) and solving iteratively in time, we obtain the temperature and the percentage of decarbonated volume at different distances from the PSZ as indicated in Fig. 9c from the computation including heat sinks. At the end of the experiment, 80% of the mass is decarbonated at distances of x = 0.1 mm while the percentage drops rapidly thereafter to around ∼25% at x = 0.5 mm. A distance of 0.1 mm corresponds to a thickness of 0.2 mm, which is roughly compatible with the slightly smaller h value adopted in the computation (0.13 mm).

Discussion
We suggest that the microstructures within the ∼100 μm-thick DDZ adjacent to the principal slip zone are the result of thermal decomposition (e.g. Han et al., 2007aHan et al., , 2007b in the case of the vesicular rims in dolomite grains, and recrystallization in the case of the crystalline calcite matrix , both caused by the frictional heat generated within the 10 μm thick PSZ during localized shear and diffusing into the adjacent material. Similar vesicular decarbonation rims (VDRs) have been observed in rapidly heated and decomposed carbonates trapped in volcanic magmas (Deegan et al., 2010) and in calcite grains found in the PSZs of large-displacement thrust faults (Collettini et al., 2013). Electron Microprobe chemical analysis shows increased concentrations of Mg on the edges of the large dolomite clast within the DDZ closest to the PSZ (Fig. 8b) which likely indicates a decarbonation reaction product. Temperature calculations show that at the end of experiment s458 (1 m s −1 , 50% dolomite/50% limestone) 80% of the mass within 100 μm of the principal slip surface has decarbonated, while the percentage drops rapidly at greater distances. This is broadly consistent with microstructural observations showing that the width of the DDZ where vesicular rims are present is ca. 100 μm (Figs. 7b, 8).
The lack of significant grain size reduction or shear offset in the dolomite and calcite grains outside the PSZ (Fig. 7a-b, f-g) but within the DDZ suggests that no significant displacements were hosted outside of the PSZ, and heat was likely generated within the PSZ due to extreme localization of strain, and diffused away from the slip surface decarbonating the dolomite grains. Evidence of degassing in the very small-displacement experiments indicates that CO 2 starts to be liberated almost immediately after the onset of shearing. Temperature calculations suggest that the bulk temperature needed to initiate decarbonation of dolomite (600 • C) is only reached after around 75 cm of slip in these experiments (Fig. 9). This indicates that the early release of CO 2 (detected for slips as small as 0.14 mm in the experiments) is likely due to grain crushing and flash heating at grain-grain contacts (Beeler et (Fig. 3d) containing a concentric layer of calcite up to 100 μm wide within a calcite matrix, and experimentally produced dolomite clasts (Fig. 7d) with vesicular decarbonation rims suspended in a calcite matrix. Experimental dolomite clasts are smaller due to initial gouge being prepared to a grain size of <250 μm. Rice, 2006;Tisato et al., 2012) inducing localized high temperatures and subsequent decarbonation (e.g. De Paola et al., 2011aPaola et al., , 2011b. In the calcite matrix, no textural indicators for calcite decomposition such as the vesicular texture described in the dolomite grains were observed. Lack of evidence for calcium oxide (due to calcite decomposition) or portlandite (due to reaction of calcium oxide with water from room humidity) may suggest that the temperature in the DDZ was effectively buffered at 800 • C by the dolomite decarbonation reaction (Fig. 9), below the ca. 900 • C required to decarbonate calcite.
The dolomite grains with vesicular rims and recrystallized calcite matrix of the experimental samples (Figs. 7d, 10) show a remarkable similarity to the dolomite grains and calcite matrix microstructures found in the basal layer of the Heart Mountain landslide (Figs. 3d, 10). Anders et al. (2010) previously interpreted the calcite rich matrix in the basal layer as being a secondary cement precipitated from meteoric and hydrothermal fluids migrating up to and along the base of the slide block after cessation of slip. Our experiments suggest an alternative formation mechanism involving recrystallization and/or annealing of fine-grained calcite gouge due the high-temperatures attained during sliding. One can imagine a situation post-slip in which the porous decarbonated experimental grains may develop over time into microstructures resembling the natural dolomite grains with concentric calcite rims (Fig. 10a), where hydrothermal and meteoric fluids precipitate calcite within the vesicles. These vesicular textures have previously been interpreted as forming by dissolution over time to form skeletal grains, not dynamically, and then infilled by post-slip calcite and silica precipitation events related to hydrothermal and meteoric fluids identified by cathode luminescence and isotopic compositions (Anders et al., 2010).
We propose that thermal decomposition as evidenced by pervasive vesiculation textures, and the subsequent release of CO 2 , indicate that temperatures in the slip zone locally reached the decomposition temperature of carbonates (at least 600 • C for dolomite; Rat'ko et al., 2011) within just a few hundred μm of slip. This near-instantaneous CO 2 production combined with porosity reduction during grain crushing drives local overpressure and reduces the effective stress on the sliding surface (Billi and Di Toro, 2008;Rowe et al., 2012;Sulem and Famin, 2009). While experimental constraints limited normal stresses to 10 MPa, at depths of 2-4 km the rate and amount of CO 2 produced would be significantly increased. Assuming a low permeability of the wallrock adjacent to the slip zone (Goren et al., 2010a(Goren et al., , 2010b, once the slide was triggered the system may self-pressurize and promote slip from a very early stage. Prior to this stage other mechanisms must account for initiation of movement on a 2 • slope (Aharonov and Anders, 2006;Hughes, 1970;Melosh, 1983). We suggest that the forsterite (Fig. 3) discovered in the natural samples was formed dynamically during the extreme frictional heating associated with slip by the reaction 2 dolomite + quartz → forsterite + 2 calcite + 2 CO 2 .
Forsterite has been shown to form by the silicification of dolomite during contact metamorphism (Ferry et al., 2011) in the presence of SiO 2 and CO 2 rich fluids at temperatures of 655 • C and pressures of 50 MPa, equivalent to depth of the landslide basal surface.
Our experimental results compare favorably with theoretical calculations (Goren et al., 2010a(Goren et al., , 2010b and field observations made along the base of the Heart Mountain landslide (Anders et al., 2010;Craddock et al., 2012). The textures and chemical reactions we observed in the experiments suggest that the release of CO 2 gases by frictional heating of carbonates is a viable mechanism to explain catastrophic emplacement of landslides that move for large distances on low-angle surfaces.