Considerations on scaling behavior in avalanche flow: Implementation in a simple mass block model

Observations of runout distances combined with velocity measurements suggest that “major” dry-mixed avalanches show a scale invariance to the total drop height HSC. This is in accordance to the proposed upper-limit envelope of the maximum velocity by McClung and Schaerer (2006). The observations are also supported by a simple scaling analysis using a simple mass block model on cycloidal and parabolic tracks (Gauer, 2018b), concluding U gH ~ /2 max SC . In this supplementary paper, a simple mass block model is presented that includes basic observations of major dry-mixed avalanches, such as mass entrainment and deposition, and that reflects this scale invariance. Almost all model parameters can principally be observed in the field. Model results are compared with a series of avalanche observations of runout and velocity and match well, considering that the model is a first order approximation.


Introduction
Snow avalanches are severe threats to the population and their infrastructure in many snow covered areas, typically in mountainous regions, but also in regions that have only small relief (< 100 m) (e.g. Issler et al., 2016;Hetu et al., 2011). Delineation of avalanche endangered areas or the design of sufficient mitigation measures require in-depth understanding on the avalanche phenomenon. Practitioners are mostly interested in expected runout distances, velocity distributions along the track, and for some purposes flow heights.
In (Gauer, 2018b), the author tried to derive the principle scaling behavior of major (dry-mixed) avalanches by simply considering the behavior of a mass block along cycloidal and parabolic tracks. Comparing the results with runout observations and velocity measurements suggested that the maximum velocity scales as U gH /2 max SC and that the mean retarding acceleration dependence on the mean slope angle. However, the author did not specify the form of the retarding acceleration in more detail. In this supplementary paper, a simple mass block model is presented that combines a series of recent avalanche observations and measurements and reflects the observed scaling behavior. Mass block models have long been proposed, e.g. (Voellmy, 1964;Perla et al., 1980;Norem et al., 1987), but their proposed parameter choices fail to reproduce the observed scaling behavior. For a discussion on the scale dependency of Voellmy-type models and the NIS model see Gauer (2014).
The following section gives a brief summary of observations relevant for the model framework. It is followed by a section introducing the governing equations of the mass block model that incorporates these observations. In Section 4 various model results are presented. To demonstrate the model performance, model results are compared qualitatively with avalanche measurements of runout and front-velocity. Thereby, a more descriptive or exploratory statistics approach is used; focus of this paper is not a detailed calibration of the model. In Section 5 some limitations and outlooks are given. In this paper, we focus mainly on dense or fluidized dry-mixed avalanches but disregard the accompanying powder (suspension) part. This distinction is not always done in field observations, which can cause some uncertainty in their interpretation.

Observations
In this section, avalanche observations are presented that are relevant for avalanche models and can provide constraints for relevant parameter values. These observations constitute the framework for the present model. Lied and Bakkehøi (1980) were the first to propose a relation between the path geometry and the observed runout distance of "major" avalanches (i.e. avalanches of relative size ≥R4, cf. Greene et al., 2016). Many of their avalanches have probably returns periods of the order of 100 years and can be considered as rare events, but not necessarily as the most extreme ones. Lied and Bakkehøi expressed the runout by the ratio between vertical drop and horizontal runout distance, tanα = H/L. Based on several hundred avalanche observations, they developed the so-called α − β model for Norway, whereby the most extreme α-angle observed in a given path is related to the mean steepness of this path. The mean steepness is approximated by the slope angle β of the line connecting the upper boundary of release area and the point in the track where the terrain gradient falls below 10 ∘ . The proposed final relation does not include any length scale, suggesting a scale invariance or more specific an invariance to a change in drop height. The proposed relation implies also that any other length scales involved can only occur as constant ratios or within a dimensionless parameter. The model was later adapted in many other countries. The background data of the model provide a valuable data set and despite the variety of the data sets, the tendency in the various countries is very similar. Fig. 1 a shows a collection of observed runout angles α versus the corresponding β-angle. For more information on those data see Gauer et al. (2010) and references therein. The figure shows also the estimated exceedance probability of α versus β according to the α − β model. McClung and Schaerer (2006) suggested, mainly based on data from Rogers Pass (BC), that the relation = U H 1.8 max SC provides an upperlimit envelope for the maximum velocity in an avalanche path, where H SC is the total vertical drop height. Fig. 1  at Ryggfonn and for avalanches from various other locations (for more information on those data see McClung and Gauer, 2018). g is the gravitational acceleration. Gauer (2013Gauer ( , 2014) investigated a series avalanche observations from various test-sites for which velocity data along the track or at least for a major part of it are available. He suggested that the velocity of major (dry-mixed) avalanches scales as U gH(0.5 0.7) .

max SC
A summary of the measured velocity data from Ryggfonn (NO) is included in the later Fig. 7. Similarly, Fig. 8 includes a comparison for several "major" avalanches from various other locations covering a range of H SC ≈ [120 m, 1200 m ] (for information on those avalanches see Gauer, 2013Gauer, , 2014, and the references therein). Based on these measurements, Gauer (2014) concluded that the proposed scaling behavior has implications for the choice of the empirical parameters for avalanche models, which use a velocity dependent friction law with fixed parameters. He argued that the measurements suggest a lower explicit velocity dependency of the bed friction than generally proposed. His suggestion is in line with back-calculations by Ancey and Meunier (2004).
Little hard data on typical release masses or fracture depths of major avalanches is published; a little more is reported on estimated deposition volumes. Fig. 2 presents a collection of observed deposition volumes related to the drop height of the avalanches. The relation suggests larger deposition volumes are observed in larger tracks. The power law dependency is in accordance with observations by McClung (2009), but can also partly be attributed to entrainment along the track. Gauer and Issler (2004) and Sovilla (2004) emphasized the importance of mass entrainment long the avalanche track. Gauer and Issler noted that erosion and entrainment of mass can contribute considerably to the retardation of an avalanche. They estimated a ratio between the contribution due to entrainment and due to frictional losses of the order of 0.2. Even if an avalanche does not entrain all the available snow, ploughing through an intact snowpack will cause a resistance on the avalanche. Gauer (2016) presented estimations on the averaged erosion depth for Ryggfonn and compared those with estimates from other locations. At Ryggfonn, avalanches entrained on average approximately 0.25 m snow along the track, whereas the mean entrainment depth for the other observations is about 0.4 m. This higher value might be expected as these avalanches are regarded more extreme events on average. The estimates are in accordance with measurements presented by Sovilla et al. (2006). Hence, entrainment Fig. 1. a) Runout observations and estimated exceedance probabilities of α versus β according to the α − β model (α m = 0.96β − 1.4°; gray shaded area ± σ; for explanation see Lied and Bakkehøi, 1980 at Ryggfonn and avalanches at various locations. The gray rectangle indicates a region that covers typical "major" events (c.f. Fig. 8).

Fig. 2.
Observed avalanche deposit volumes of "major avalanche events" versus total drop height H SC (for references to the data see Gauer et al., 2010). The lines show the estimated exceedance probabilities derived from these observations.
plays an important role in the mass balance, which is indicated by Fig. 2.
On the other hand, observations by Sovilla et al. (2010) indicate that avalanches may begin to deposit-mainly starting from the tail-at a slope angle of approximately 30 ∘ , which is in the range of typical friction angles for snow on snow (Lang and Dent, 1982;van Herwijnen and Heierli, 2009).
Both processes, entrainment and deposition, are important for the mass balances of an avalanche and therefore for its dynamics.
As far as the fracture height is concerned, Fig. 3 shows a comparison of the complementary cumulative distribution function (survival function) of observed fracture depths D rel and proposed relations in the literature (Rosenthal and Elder, 2003;McClung, 2003;Perla, 1977;Faillettaz et al., 2006;Gaume et al., 2012;Bazant et al., 2003). The comparison suggests that values between 1 and 2 m could be considered rare (i.e. have probabilities in the range of 0.01 and 0.1). There is however a considerable spread in the data and the observational data might be biased to minor events.
Dry-mixed avalanches can be considered as granular flows consisting of small snow clouds with sizes typically less than 0.1 m, where the dissipation is considered to be dominated by a Coulomb-type friction with an effective friction factor tanδ e (e.g. Coaz, 1910;Savage and Hutter, 1989). Savage and Hutter proposed an ansatz where p xz is the shear stress, p zz the normal stress, ν the solids fraction, and = du dz ( / ) 2 2 is the shear rate squared. For a free surface flow, Savage and Hutter argued that in the limit of very small shear rates Eq.
(1) reduces to (2) δ s is termed the quasi-static friction angle. For high shear rates they argued that Eq. (1) tends to where δ D (ν) is the dynamic friction angle which, based on experiments, was suggested to be fairly close to δ s . They further mentioned that experiments showed very little rate-dependence in the stress ratio and that experiments suggested that the standard deviations of the stress ratios was approximately 10% of the mean for a given material. Savage and Hutter advocated that both the interior and the basal friction angles are rate independent.
A shear-rate independence is in accordance with the observations on runout and maximum velocity for major dry-mixed avalanches-both are governed by the total drop height. Effects due to varying flow heights, where the flow height is a second length scale, should only be of second-order at most.

Governing equations
In this section, we formulate the governing equations for a simple sliding block model (center of mass model) with varying mass. The model includes basic observations of major dry-mixed avalanches, such as mass entrainment and deposition, and reflects the observed scale invariance to changes in drop height. Almost all model parameters can be traced back to field observations. A similar model was proposed, e.g., by Cherepanov and Esparragoza (2008);Cherepanov (2019). However, in our case, we also allow for mass loss and we disregard the explicit velocity squared dependent resistance term (often referred to as 'turbulent' velocity-dependent resistance), which is included in most present-days avalanche models. The 'turbulent' velocity-dependent resistance introduces a scale dependency that is in this way not observed in the measurements. On the other hand, we retain implicit terms that dependent on the velocity such as the effect of curvature or terms arising from a varying mass.
The governing equation for the mass balance is where the avalanche mass per unit width M = ρ a V with the density of the flowing avalanche ρ a and V its volume per unit width. The density of the intact snowpack along the track is ρ s and d e (s) is the depth of snow measured perpendicular to the track that is entrained by the avalanche and ρ a d d (s) is the mass loss due to deposition. The traveled distance (arc-length) along the avalanche track at time t is s(t).
The momentum of the sliding block will change due to entrainment and deposition of mass. Let us recall the principle of linear momentum, which states that the momentum L (t+Δt) at time t + Δt of a system (in our case the system of the avalanche, entrained snowpack and deposited mass) is equal to the sum of the momentum of the system L t at time t and the impulse ∫ t

(t+Δt)
Rdt of the resultant of all forces acting on the system: Now, the linear momentum of the avalanche and the mass entrained by the avalanche, ΔM e , at time t, shortly before ΔM e is entrained, is just the sum of both, where U is the velocity of the avalanche and U e the velocity of the mass ΔM e . After ΔM e is entrained and ΔM d has separated (deposited) from the avalanche, the new avalanche mass M + ΔM e − ΔM d moves at the same velocity U + ΔU. The momentum of the system is now ΔM d U is the momentum of the deposited mass separated from the avalanche. Using Eqs. (7) and (6) in (5), we obtain After rearranging and dividing by Δt, one gains which becomes in the limit Δt → 0 Here, we use The mass fluxes are dM e /dt = Q e and dM d /dt = Q d . With U(s) = ds/dt and Λ(s) ≡ [M(s)U(s)] 2 and by using Eq. (4), Eq. (10) can be rewritten as For a known resulting force R s, M s ( ) , Eq. (11) can be integrated numerically and the avalanche speed U(s) found as function of s: where the effective resulting force is The avalanche mass M(s) follows from Eq. (4) The initial conditions are s = 0, U(0) = U 0 , M(0) = M 0 , and in the case of entrainment from rest U e = 0. If one includes particle impacts and rebouncing (i.e. saltation), U e will be larger than zero. For signs of saltation see, e.g., (Gauer et al., 2008, Fig. 13), but for now we disregard particle saltation. In the following, we use the ansatz for the resulting force on the avalanche, where ϕ is the local slope angle, g the gravitational acceleration, a 0 (= tan δ be ) is the effective coefficient of friction in which δ be is the basal friction angle. κ is the local curvature of the track. The last term on the right hand side expresses the fracture resistance of the snowpack, where K e is the socalled entrainment toughness (cf. Cherepanov, 2019;Cherepanov and Esparragoza, 2008 or by using Eq. (16) as For dU/ds = 0 and U e ≈ 0 and ansatz (15) where the subscript (•) m marks the point where the maximum velocity is reached. Now, on parabolic or cycloidal tracks, the curvature can be given as κ = c(ς, ϕ 0 )/H SC , where H SC is the total drop height of the path, ς the scaled travel distance, and ϕ 0 is the initial slope angle. According to (14), it is furthermore reasonable to assume that the avalanche mass can be approximated as M(s) ≈ m(ς)H SC . For example, assuming an initial mass M 0 per meter width where, ρ r is the averaged density of the snowpack in the release area, D rel is the fracture depth and c 0 a coefficient describing the length of the release area relative to the length of the track (i.e. c H sin / SC 0 0 is the slope parallel length of the release area l rel ). Typically, c 0 might be in the range of 0.1 to 0.3, that is the height difference of in release area is about 10 to 30% of the total drop height H SC . If we now consider a cycloid, the mass can be written as where ϕ 1 (ς) is the slope angle along the track. Hence, the dimensionless maximum velocity, U gH / , m is nearly independent of the total drop height of the track, which is in accordance to the observations on major avalanches (cf. Gauer, 2014Gauer, , 2013Gauer, , 2018bMcClung and Gauer, 2018). For a cycloid, c(ς, ϕ 0 ) increases with increasing ϕ 0 and decrease with increasing ς. For parabolic tracks, there is an increase with increasing ς at first, and an increase for ϕ 0 . The increase with slope angle compensates partly the effect of the slope angle dependency of the Coulomb-friction term and contributes to a more constant retarding acceleration along the track. In addition, the increase in total mass causes a reduction of the velocity depending term, which is caused by mass entrainment itself. This counterbalances the increasing Coulombfriction in the lower part of the track. Mass deposition counteracts the reduction due to entrainment. These feedback mechanisms lead to a rather constant apparent retarding accelerations (i.e. the sum of frictional contribution and due to entrainment) for major avalanches, which is in concert with energy considerations for observations by Gauer (2013).
Eq. 23 implies also that the ratio between snow entrainment and initial mass (ρ s d e /c 0 ρ r D f ) has a decisive role for fast moving avalanches and long runouts. A smaller ratio favors higher velocities.
To account for the drag due to the quiet ambient air, let us assume a resistance term which is proportional to the area of the interface and the square of the avalanche velocity. Eq. (15) where ρ air is the density of air and l a is the length of the avalanche. In this case, Eq. 23 rewrites to In a first estimate, one may assume l a ~ l rel . As long as ρ s d e ≫ ρ air c Da l a , the drag due to the ambient air can safely be neglected. Otherwise, the air drag might introduce a dependency on the total drop height through l a . As a conservative estimate, the drag coefficient c Da is assumed to be in the order of 0.01 (see for example Curcic and Haus, 2020).
Recalling that for major dry snow avalanches m will typically be larger than 0.75/ 2 but probably not much larger than 1.1/ 2 . In this case, Eq. (25) provides some constraints on the parameter values and for contributions due to possible further terms that might have been neglected here.

Model test
The following section presents several comparisons between observations and model calculations. We do not try to provide a detailed parameter calibration, rather we use first estimates on the parameters based on the observations presented in Sec. 2 to demonstrate the general model performance. Thereby, we primarily consider major drymixed avalanches (i.e. relative size ≥ R4). Hence, the parameter choice might be biased in regard to those observations.

Parameters
All parameters, except of the effective friction parameter a 0 and the drag coefficient c Da , can in principle be observed in the field. For the effective basal friction parameter, a 0 , we choose as a first estimate This corresponds to an effective basal friction angle of δ be ≈ 17 ∘ . The value is, e.g., in agreement with measurements by Platzer et al. (2007) for dry snow, which were done on a chute. Also Heimgartner (1977) found similar values. As previously indicated, the exact value of a 0 may to a certain degree depend on the state of flow and properties of the avalanche, like flow density, clod size, humidity, and other. It actually may vary to a certain degree along the track. However, this is outside of the scope of the present paper.
Observations of erosion depth at Ryggfonn, mentioned above, show typical erosion depths in the range of 0.1 m to 0.4 m with a median of around 0.25 m (Gauer and Kristensen, 2016). Estimates based on a series major avalanches in various tracks show a similar range, but a somewhat higher mean (Gauer, 2016). These estimates are in accordance with measurements by Sovilla et al. (2006). In the simulations, we use where H e is the erosion depth measured vertically. For the entrainment toughness K e in Eq. (15), we follow the ansatz for the fracture toughness by Schweizer et al. (2004): The density of ice is ρ ice ≈ 917 kg m −3 . Lacking better data at present, we assume A c ≈ 8 kPa m and B c ≈ 2.3. For cases of our concern, K d g H m 2 / ( ) e e SC is typically less than 0.01 and the effect can be neglected for a first order approximation.
Observations by Sovilla et al. (2010) suggest that avalanches start to deposit at slope angles of approximately 30 ∘ , which is in the range of typical friction angles for snow (cf. Lang and Dent, 1982;van Herwijnen and Heierli, 2009). Here, due to lack of better knowledge, we use a simple approach to include mass loss. We assume where d d0 is a parameter and F ϕ(s) and F κ(s) are functions of the slope angle and curvature, respectively.
where μ s (≈0.6) is the friction coefficient for snow on snow, μ d is an additional parameter regulating the intensity. In a first approximation, we assume d d0 = d e (s)ρ s /ρ a and μ d = 0. This implies that the avalanche will gain mass as long as d e > 0 and ϕ > 0. This may not be the case for all avalanches, especially for minor ones where the mass balance might become negative. A dependency on the curvature is also included by where κ m is set to 0.02 m −1 . This implies less tendency to deposit on convex slopes but a drastic increase of deposition on very concave slopes. Fig. 4 shows a plot of both factors.
Calculations with a simple release model (Gauer, 2018a) suggest that the expected fracture depth decreases with increasing slope angle of the release zone and can be approximated by where HS 0 is an appropriated snow depth for an avalanche release. Observations on fracture depths are shown in Fig. 3, which suggest HS 0 ~ 2 m as a first estimate for major avalanches. Table 1 gives a summary of the first guess parameter values used in the following sections, if not otherwise stated.

Parabolic and cycloidal tracks
In the first test, we follow the proposed approach in Gauer (2018b) and study tracks with a simple geometry. The simulations are carried out for cycloidal and parabolic tracks with varying mean steepness, which is approximated by the β-angle. Fig. 5 shows the runout simulations using the parameter values given in Table 1

P. Gauer
Cold The figure shows also the effect of different erosion rates with otherwise unchanged parameters-higher mass intake along the track causes higher retardation resulting in lower maximum velocities and shorter runouts.
The simulations are carried out for a total drop height H SC = 1000 m, however, as the model is invariant to H SC , simulations with other drop heights show no differences. That the model behaves invariant to H SC is demonstrated in Fig. 6, which shows U H max SC . The figure shows example calculations for a cycloidal and parabolic track with an initial slope angle ϕ 0 = 40 ∘ and H e = 0.5 m and H e = 0.25 m. The lines depict the probability of exceedance according to Fig. 1 b. For comparison, the measured maximum front velocity for a series of major avalanche events (c.f. Fig. 8) are presented too. The simulations capture the observed trend well. Fig. 7 shows a comparison of a series of avalanche observations at the Norwegian avalanche test-site Ryggfonn (for more details on the data see, e.g. Gauer and Kristensen, 2016) and a simulation with a slightly adapted parameter set. In accordance with the observations M 0 ≈ 5 ⋅ 10 4 kg m −1 and H e = 0.25 m is used. The remaining parameters are given in Table 1. The simulation provides a good first estimate on the larger events observed at the test-site. Fig. 8 illustrates the comparison of measured front-velocities from major avalanche events in various tracks with the mass block calculations. For details on the measurements, see (Gauer, 2014(Gauer, , 2013(Gauer, , 2018b. To facilitate a comparison, the path geometries are scaled by the drop height H SC and the velocity as U gH / /2 SC . As in the case for the Ryggfonn simulation, the model provides a reasonable good first estimate for those major events, both in respect to runout and velocity. As proposed, the apparent retarding acceleration is rather constant and the ratio between losses due to entrainment and frictional losses varies between approximately 0-0.3 along the track with a mean of ≈0.17. . The lines show the estimated exceedance probabilities derived from observations shown in Fig. 3. The figure shows example calculations for a cycloidal and parabolic track and two erosion depths. The initial slope angle ϕ 0 of the tracks is 40 ∘ . The remaining parameters are given in Table 1. The gray triangles depict measured maximum front-velocities from major avalanche events in various tracks (c.f. Fig. 8). The lines depict the probability of exceedance in Fig. 1 b.

Tamokdalen (NO), 2019-01-02
An avalanche accident, which took place on 2 January 2019 below Blåbaertinden, Tamokdalen in North Norway, provided a rare data set of GPS-data (for more information on the accident, see Sandersen, 2019). One victim was warring a wristwatch/heartbeat monitor with a tracking frequency of approximately 1/s as he was transported by the avalanche over a drop height of about 200 m. The GPS-data allowed an estimate on the avalanche velocity during the descent of victim. Fig. 9 shows the comparison of the velocity estimates and a simulation with the "base" parameter in Table 1. The model provides again a good estimate for the event, both in respect to velocity and runout, even if a tip of the actual avalanche went slightly further.

Austria
The comparison between observed runout distances along the track, S f , and simulations is shown in Fig. 10 for two combined data sets from Austria (Klenkhart and Weiler, 1994;Wagner, 2016). The events were all natural releases and thought to have rather high return periods in the order of O (100 years ). No detailed velocity information are available. Out of the data set of 134 avalanches, 120 had sufficient geometry information (i.e., provided long enough profiles) for the comparison. The data cover a range of H SC ≈ [110 m, 1810 m] and β ≈ [18 ∘ , 39 ∘ ].
The maximum velocity for the simulations is U gH / /2~0.9

Se
, where H Se is the effective simulated drop height. In general, the model catches the observations rather well. The correlation between the modeled and observed runout distances is approximately 0.94 and the interquartile range of the relative error is 0.13. Nonetheless, there are several marked out-layers. In those cases a slight change in the parameters, which would also reflects the expected return period of the event better, may provide even lesser spreading. In some of these cases only a slight change in the parameters can cause a considerable change in the runout distance.

Aulta avalanche, Lukmanier (CH), 1984-02-08
How the various parameters influence the simulation results is shown in Fig. 11. The corresponding parameters are given in Table 2. For this comparison, the Aulta avalanche from 8 February 1984 (for details on the event, see Gubler et al., 1986;Gauer, 2013) is chosen as a complete velocity profile (solid black line) along the whole track exists and as it is a known event used in several publications.
The first simulation with the "base" parameters according to Table 1 (run 1) provides again a good estimate on the observations. Reducing the initial mass, M 0 , causes a decrease in runout and velocity (run 2) because the retarding effect due to erosion increases (i.e. the ratio ρ s d e / c 0 ρ r D rel in (23) increases). The same is observed for increasing the erosion depth H e (run 3). Mass loss has likewise an important effect, both on the velocity and runout (run 4). As easily seen, decreasing the  Table 1, however, in accordance with the observations M 0 ≈ 5 ⋅ 10 4 kg m −1 and H e = 0.25 m is used. For comparison, the range of observed front velocities along the track at the Ryggfonn-site is given. The blue line shows the mean, the shaded area the ± σ-range and the red dashed line shows the observed maximum derived from observations along the track. In addition, bars show the interquartile range of the front velocity U LC measured between load plates LC54 and LC321 and the front velocity U b at the base of the catching dam. The red dots show the corresponding means and red crosses mark the measured maxima. Fig. 8. Simulation (red solid line) along the "mean" path with M 0 ≈ 9.8 ⋅ 10 4 kg m -1 and H SC = 1000 m. In addition, a summary of measured front-velocities from major avalanche events in various tracks are shown. The blue line shows the mean, the blue shaded area the ± σ-range and the red dashed line the observed maximum derived from observations along the track. Red dots mark the maximum of the different measurements. The black line represents a "mean path" geometry and the gray shaded area the envelope of all path geometries. The light gray polygon gives an impression of the expected ranges for major avalanches according to the observations as presented in Fig. 1. The model parameters are given in Table 1. Fig. 9. Avalanche velocity as estimated by the model (red line) and as inferred from GPS records extracted from the wristwatch/heartbeat monitor of one of the victims (blue dots). Here H e = 0.25 m, the remaining model parameters are given in Table 1. The inlet shows the corresponding s-t plot (distance and time are scaled).
effective friction factor, a 0 , will increase the velocity and runout (run 5). That some of the effects can compensate each other is shown by run 6, where decreasing the friction factor, a 0 , compensates partly for the increase in erosion depth, H e . Likewise run 7, where the increase in initial mass, M 0 , compensates for the increased erosion depth. However, the required increase of initial mass is considerable in this case. This can be seen from Eq. (23). By keeping all parameters the same, except of the erosion depth, d e , and the fracture depth, D rel , the ratio d e /D rel should stay approximately the same to reach similar maximum velocities. Fig. 13 shows a Monte-Carlo simulation for the horizontally extended "mean track" in Fig. 8. The parameter distributions used are shown in Fig. 12. They are based on observations or best guesses. Relying upon observations by Perla (1977) (see also Fig. 3), the critical snow depth, HS 0 , is assumed to follow a half-normal distribution with a mean of approximately 1.4 m and a standard deviation of 0.6 m. The entrainment depth is given by a lognormal distribution with a mean of approximately 0.5 m and a standard deviation of 0.6 m. Using data by van Herwijnen and Heierli (2009) and Simenhois et al. (2012), μ s is approximated by a general extreme value distribution with a shape parameter, k gev ≈ 0.29, scale parameter, σ gev ≈ 0.036, and location parameter μ gev ≈ 0.567. The remaining parameters are taken as normal distributed with ρ s = N(150,15), ρ a = N(150,15), and a 0 = N (0.3,0.015) (a 0 = N(0.33,0.015) for the case of Ryggfonn in Fig. 14).

Sensitivity test
Comparing the range of observations in Fig. 8 with the model results in Fig. 13 suggests that the model gives reasonable first estimates for those major avalanches, covering more or less the full range. Fig. 14 shows a similar Monte Carlo simulation for the Ryggfonn path. The same parameters are used as before, except d d0 in Eq. (29) is given by a uniform distribution with a mean of 1 m and for μ d a normal distribution is used where μ d = max (0, N(0.26,0.052)). This will allow for actual mass loss and accounts to a degree for the spread in relative  Table 1. a) mean retarding acceleration < a ret > /g = ΔH/S f , where ΔH is the drop height between release area and runout; b) runout distance along the track S f ; and c) the relative error (S fm − S fo )/S fo , where S fo marks the observations and S fm the simulation. The dashed line shows the mean, the gray shaded area the corresponding standard deviation and the dash-dotted lines mark the interquartile range. The marker size indicates the avalanche size, if data are available. The scaled maximum velocity is color coded. The dotted lines in the two upper panels indicate a oneto-one relationship. Fig. 11. Simulations of the Aulta avalanche 1984-02-08 (Lukmanier, Switzerland). Measurement by Gubler et al. (1986) with a CW Doppler-radar. The model parameters are given in Table 2. size and ambient conditions of the events shown in Fig. 7. In our case, deposition accounts to large degree for that the avalanche stops even in steeper parts of the slope, which can often be observed in nature. The effect was already seen in Fig. 11 run 4. Also in the present case, the model shows a decent overall match compared to the observations in Fig. 7. Fig. 15 compares the comparative cumulative distribution functions (CCDF; probability of exceedance) of the observed runouts at Ryggfonn (for more deatils on the data see Gauer and Kristensen, 2016;Gauer et al., 2009) and corresponding ones of Monte Carlo simulations with (see Fig. 14) and without catching dam. To be consistent with the observations, for the comparison, only those avalanches are considered that reached the runout area (i.e. surpassed at least the elevation of the former transmission line marked as SC123). Considering the simple approach of the mass block model and the uncertainties involved in the measurements and their reporting (e.g., avalanche type; the distinction between the deposits of the dense or fluidized part and those of the powder part; or the degree of pre-filling of the catching dam), the model provides reasonable estimates on the avalanche runouts.

Concluding remarks
This paper presents a simple mass block model that includes basic observations of major dry-mixed avalanches, such as mass entrainment and deposition and is basically scale invariant to the total drop height. All its parameters, except of the effective friction parameter, a 0 , are more or less observable in the field. For the effective friction parameter, a 0 , we take granular flow as guidance and use a value that is consistent with chute experiments using dry snow (e.g. Platzer et al., 2007). . The test is run for a horizontally extended "mean" path shown in Fig. 8. The parameter distributions used are shown in Fig. 12. The remaining parameters are given in Table 1. The light gray polygon gives an impression of the inter-quantile-range 0.1-0.9 and the blue dashed lines of the interquartile range. The errorbar depicts the mean of the simulated maximum velocity and its position. For comparison, the red line shows the simulation with the "base" parameter given in Table 1 and shown in Fig. 8.   Fig. 14. Monte Carlo simulation (median; solid blue line). The test is run for the Ryggfonn path shown in Fig. 7. The parameter distributions used are shown in Fig. 12. The remaining parameters are given in Table 1. The light gray polygon gives an impression of the inter-quantile-range 0.1-0.9 and the blue dashed lines of the interquartile range. The errorbars depict the mean and interquartile range of the simulations of the velocity taken between positions of load plates LC54 and LC321 and the velocity at the base of the catching dam. For comparison, the red line shows the simulation with the "base" parameter given in Table 1. The red dashed line redraws the run with adapted parameters shown in Fig. 7. Although it is only simple a mass block model, it is still an admissible first-order approximation. The model is dominated by a Coulomb like friction, which is in line with results, e.g., by Ancey and Meunier (2004); Ancey (2005). Mass erosion, however, causes an apparent velocity squared dependency. This dependency is governed by the ratio between entrained mass per unit area and the (initial) avalanche mass. A relatively large initial mass or little entrainment will favor slightly faster avalanches with longer runouts, which is quite in accordance with observations. In contrast to, e.g., Voellmy-type models or the NIS-model, the model parameters for the presented model are invariant to drop height.
Despite all the uncertainties associated with the model and the observations shown, the model provides reasonable results for major events as demonstrated above. The simulations demonstrate clearly that besides mass entrainment, mass loss can have a significant influence on the runout and velocity of avalanches. The obvious influence of mass erosion puts limits to the meaningfulness of granular experiments that neglect erosion. This holds also true for numerical models that do not include erosion and deposition. To improve the verification of the model approach, more combined observations on velocity, runout distances, mass balance, and snowpack properties from various observation sites that show different geometries (e.g., mean slope angles and drop heights) are desirable.
At present, we use a rather simple approach for the entrainment and loss of mass in the model. For the mass block model this is reasonable, especially the entrainment model. More sophisticated 2-D (or even 3-D) models will require more enhanced models for entrainment and mass deposition, models like the ones by Rauter and Köhler (2019); Naaim et al. (2004).
The effective friction parameter, a 0 , probably depends on the state of flow and may vary to a certain degree along the track, which needs to be accounted for. In our Monte Carlo simulations, we assumed that all parameter distributions are independent, however, some of the parameters might be correlated.
The presented model can be a useful simple tool for practitioners, especially in combination with Monte Carlo simulations, to obtain first estimates on runout distances of major avalanches as required in hazard mapping. The model circumvents the problem of finding a suitable βpoint (for a discussion on the issue see, e.g. Sinickas and Jamieson, 2014), which is necessary for the known statistical runout models, like the α − β model (Lied and Bakkehøi, 1980) or the run-out ratio model (McClung and Mears, 1991). At the same time, the model provides also speed estimates. As several input parameters are observable in the field, the model can be adapted to local conditions. Following the approach by Nohguchi (1989), it can easily be extended into a three dimensional space. This can reduce the uncertainty due to subjectively drawing profile lines. Furthermore, the model allows to include other resistance terms, e.g., the effect of forest to a certain degree. The latter, may require estimates on the flow height, which mass block models can principally not provide.

Author statement
The paper was solely written by the author.

Declaration of Competing Interest
The author declares that he has no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.