ArticlePDF Available

Abstract

Dry snow avalanches consist of two distinct layers. A dense-flow layer is superposed by a powder-snow layer, a cloud of relatively small ice particles suspended in air. The density of this suspension is one order of magnitude smaller than that of the dense flow. A simulation model for dry avalanches has been developed, based on separate sub-models for the two layers. The sub-models are coupled by an additional transition model, describing the exchange of mass and momentum between the layers. The fundamentals of the two-dimensional granular flow model for the dense flow and of the three-dimensional turbulent mixture model for the powder flow are presented. Results of the complete coupled model, SAMOS (Snow Avalanche MOdelling and Simulation), applied to observed catastrophic avalanche events, are discussed, and the prediction of powder-snow pressures acting on a tunnel bridge is briefly described. SAMOS is used routinely for hazard zoning at the Austrian Federal Service for Torrent and Avalanche Control.
International Snow Science Workshop, Davos 2009, Proceedings
519
International Snow Science Workshop Davos 2009
Avalanche Simulation with SAMOS-AT
Peter Sampl1 and Matthias Granig2
1 AVL List GmbH, Graz, Austria
2 Stabstelle Lawinensimulation, Schwaz, Austria
ABSTRACT: The simulation software SAMOS for dry snow avalanches has been used by the Aus-
trian Service for Torrent and Avalanche Control since 1999. The software was based on a two-
dimensional, depth-averaged model with Coulombian and turbulent bottom friction for the dense flow
part and on a three-dimensional mixture-model for air and ice-particles, which form the powder snow
part. The mixture model involved separate mass-balances for the air and ice-particles and one single
balance for the momentum. Additional balances for the turbulent energy and dissipation rate were
considered. The amount of powder snow generated from the dense flowing snow was predicted as-
suming an analogy of momentum- and mass-transfer above the dense-flow-surface. A triangular grid,
moving with the flowing mass over the terrain surface, was used for the numerical solution of the
dense-flow equations in Lagrangean formulation, while a three-dimensional, non-moving (Eulerian)
grid, adapted to the terrain surface, was employed to solve the equations for the air-and-ice mixture.
An improved version, SAMOS-AT, has been completed in 2007. The dense flow friction model was
changed, so that the turbulent friction part depends also on the flow-depth. The mixture-model for the
powder snow part was replaced by a full two-phase-model for air and ice-particles, with separate mass
and momentum balances. A smoothed-particle hydrodynamics method is used now to solve the
dense-flow-equations. The improved model is described in this paper.
KEYWORDS: Avalanche dynamics, simulation, dense flow avalanches, powder avalanches.
1 INTRODUCTION
SAMOS is a simulation-software for dry
snow avalanches. It describes both the dense-
flow-layer (DFL) and the powder-snow-layer
(PSL) of an avalanche, as well as the interaction
between them. The software has been devel-
oped for the Austrian Service for Torrent and
Avalanche Control (WLV), which has used a first
version since 1999 (Zwinger et al., 2003; Sampl
and Zwinger, 2004). The use of the software
revealed several problems: too large DFL-width
in channelled terrain and often too large run-out
distances of the DFL and also the PSL in terms
of the 5 kPa zone. An improved version,
SAMOS-AT, has been developed and com-
pleted in 2007. The main changes are different
bottom friction models for the DFL, the treatment
of the PSL as two-phase-flow and a different
numerical solution scheme for the DFL. Fur-
thermore, the usage of the software has been
substantially simplified.
2 SAMOS-AT AVALANCHE MODEL
In SAMOS, dry snow avalanches are
thought to consist of air and ice particles in vari-
able concentrations. With the volume-fraction c
and intrinsic density
ρ
ice of the ice particles and
the air density
ρ
air, the mixture density
ρ
is given
by
airice
cc
ρρρ
)1( += . (1)
A vertical three-layer-structure (Figure 1) is
assumed in the avalanche, the layers being
identified by different ranges of the mixture den-
sity: a dense flow layer (DFL) with 200 to 300
kgm-3 at the bottom, a powder snow layer (PSL)
with 1 to 10 kgm-3 at the top and a transition
layer in between.
Powder snow layer
(PSL)
Transition layer
Dense flow layer
(DFL)
Sliding surface
(DHM)
Powder snow layer
(PSL)
Transition layer
Dense flow layer
(DFL)
Sliding surface
(DHM)
Figure 1. Vertical layer-structure of a dry snow
avalanche as assumed in the SAMOS-AT
model.
The transition layer and the PSL are as-
sumed to develop above the DFL with time, de-
pending on the flow conditions. The transition
layer is not spatially resolved in the model, but
reduced to an interface which connects DFL and
PSL. The DFL is modelled as shallow flow in
two dimensions on the terrain surface (pre-
scribed by a digital height model (DHM)), while
______________________
Corresponding author address: Peter Sampl,
AVL List GmbH, Hans-List-Platz 1, 8020 Graz,
Austria;
tel: +43 316 787 439; fax: +43 316 787 1923;
email: peter.sampl@avl.com
International Snow Science Workshop, Davos 2009, Proceedings
520
International Snow Science Workshop Davos 2009
the PSL is modelled as three-dimensional flow
above the DFL, including the air masses a few
hundred meters above ground.
2.1 Dense Flow Layer (DFL) Model
The DFL model is obtained by dividing the
DFL into a large number of small mass elements
and writing the integral mass- and momentum-
balances for each of it. Each element contains a
mass me, touches the bottom surface at an area
Ae and extends up to the DFL surface. The bal-
ances result in a set of ordinary differential
equations, which are coupled by the lateral
forces between the elements. The DFL is as-
sumed to be shallow: the ratio of the character-
istic DFL depth H (normal to the terrain) to the
DFL length (and width) L is taken to be small. All
terms of order (H/L)1.5 or smaller are neglected
(following the work of Savage and Hutter, 1989;
derivation see Zwinger et al., 2003). A constant
bulk density
ρ
is assumed for the entire DFL. A
smoothed-particle hydrodynamics (SPH)
method (Monaghan, 1988) is used, so that the
flow depth h at each point
x
on the terrain sur-
face is expressed as weighted sum over all ele-
ment with centers at position
e
x
:
=
e
ee
xxWmxh ),(
1
)(
ρ
(2)
The SPH-kernel function
),(
e
xxW
(see 3.1)
has a dimension of m-2. The bottom area Ae of
an element follows as
)(
e
e
e
xh
m
A
ρ
=
(3)
The mass and momentum balances for each
element are formulated in a local coordinate
system with direction 1 along the elements cur-
rent flow direction
u
(defined as average veloc-
ity in the element), direction 3 normal to the ter-
rain and direction 2 normal to both. For brevity,
the element indices e are dropped below.
1
,eu
3
e
2
e
A
h
1
,
eu
3
e
2
e
A
h
Figure 2. A DFL mass element with bottom sur-
face A, average depth h and average velocity
u
and the local coordinate system
321
,, eee
.
Snow entrainment is considered at the DFL
front and specified by qent, the entrainable mass
per unit surface area. The front width wf within
an element, measured normal to the flow direc-
tion, is zero for elements not at the front. If the
mass transferred to the PSL per unit surface
area is termed js and the time t, the mass bal-
ance for an element can be written as
Ajuqw
dt
dm
sentf = . (4)
The momentum balance in direction 3 yields
an equation for the bottom pressure in the ele-
ment
A
m
ugp
b
)(
2
13
)(
κ
= (5)
with gi the vector of gravitational accel-
eration and
κ
1 the terrain curvature in flow direc-
tion. The momentum balances for directions i
=1,2 can be written
)(
)(
)(
)( 2
)(
res
i
i
b
b
i
i
i
i
F
u
u
A
h
p
K
x
Amg
dt
mud
+
=
τ
(6)
with
τ
(b) the bottom shear stress and Fi
(res)
the resistance force resulting from obstacles (an
additional resistance force to break entrained
snow from the ground can be considered op-
tionally; see Sailer et al, 2008). The second term
on the right hand side describes the force result-
ing from the lateral normal stresses, which are
expressed as product of the bottom pressure
with the normal stress coefficients K(i). A linear
drop of the stresses from the bottom to zero at
the DFL surface is assumed, hence the average
over the depth is half the bottom value. The
shallowness assumption allows to neglect lateral
shear stresses. The resistance force due to ob-
stacles like trees in a forest or fields of boulders,
defined by a characteristic obstacle height hres,
diameter dres, average lateral distance sres and
shape resistance coefficient cw is computed as
u
u
s
A
hhd
u
cF
i
res
resre sw
res
i
2
2
)(
),min(
2
ρ
=
(7)
For a granular material with internal Mohr-
Coulombian friction, as assumed in SAMOS, the
coefficients K(i) depend on the flow state and the
friction parameters (Zwinger et al., 2003). The
normal stress is larger than p(b), if the material is
compressed, and smaller, if it expands. This
results in a net energy loss in each compres-
sion-expansion cycle. In practice, this led to a
considerable reduction of runout-distances in
SAMOS when finer DHMs were used, since
these resolve more surface-ripples, each caus-
ing a compression-expansion-cycle. Hence K(i)
1 is assumed in SAMOS-AT, as for a Newtonian
fluid, and the retarding effect of ripples is con-
sidered in the bottom friction parameters.
International Snow Science Workshop, Davos 2009, Proceedings
521
International Snow Science Workshop Davos 2009
The bottom friction in SAMOS-AT is com-
puted as
2
2
0
0
)(
0
)(
ln
1
+
1
+
+
++=
D
DD
ss
s
bb
B
R
h
u
RR
R
p
κ
ρ
µττ
(8)
In the above relation,
τ
0 is a yield stress
(which is applied in small-avalanche-simulations
only; see Sailer et al., 2008),
µ
is the Coulom-
bian bed friction coefficient, Rs a “fluidization fac-
tor”, RD the surface roughness, and Rs0,
κ
D and
BD empirical constants. The values used for
µ
,
Rs0,
κ
D, RD and BD are 0.155, 0.222, 0.43, 0.1 m,
and 4.13, respectively.
The fluidization factor Rs is defined as the ra-
tio of dispersive stresses to the effective bottom
pressure:
)(
2
b
s
p
u
R
ρ
= (9)
The Rs-term serves to increase the Coulom-
bian bed friction at low fluidization, so as to stop
slow avalanche parts already in terrain steeper
than atan(
µ
) and to prevent spreading when
mass is close to stopping.
The last term on the right hand side in (8) is
inferred from classical boundary layer theory for
turbulent fluid flow and the logarithmic law-of-
the-wall for rough plates (see e.g. in Gersten
and Herwig, 1992). These laws may be obtained
by assuming turbulent eddies (i.e. larger clusters
of non-coherent ice particles, or coherent snow
clods), whose size is proportional to the distance
from the terrain surface. The difference to the
turbulent friction term as in the classical Voellmy
model is the (moderate) reduction of friction with
increasing flow depth h.
2.2 Transition Layer Model
The transition layer is dominated by salta-
tion, particle-collisions, sedimentation and turbu-
lent suspension due to the aerodynamic forces
acting onto the particles at the DFL surface. The
interaction of particle-suspension and the turbu-
lence in the air is intense and complex. A
strongly simplified model thus has been devel-
oped for the initial version of SAMOS and has
been kept with a few modifications.
The DFL surface with the transition layer is
considered to be a “rough wall” for the PSL (this
wall is of course moving with the DFL-speed).
With
u the magnitude of the velocity difference
between the DFL surface and the PSL at a ref-
erence height y above, equal to the assumed
transition layer depth, the wall shear stress in
turbulent flows at rough walls is, again from
classical boundary layer theory, given by
+
=
P
P
air
w
B
R
y
u
ln
1
2
κ
ρ
τ
(10)
Rp is the roughness length and
κ
(again
0.43) and Bp (8.5) constants. The suspension
mass flux per DFL surface unit js is assumed to
be proportional to the momentum flux, which is
equal to
τ
w. Since
τ
w is proportional to
u and js
is to the particle concentration difference
c,
between DFL and PSL at y, the mass flux is re-
scaled accordingly and computed as
=
p
dsusp
ws d
cc
u
c
j
τ
(11)
The correction term in brackets is introduced
to account for dissimilarities between momen-
tum and particle transport. It shall reflect that
particles with larger diameter dp and smaller
form-drag-coefficient cd are transported to a
lesser extent. csusp is an empirical correction con-
stant with the dimension of a length. No size
distribution is considered for the particle diame-
ter in the model. The roughness seen by the
PSL is computed as
33
2
0g
c
g
u
cRR
ice
w
ressaltP
ρ
τ
++=
(12)
where R0 is the geometric roughness of the
DFL surface without saltation. The second term
models the increase of roughness due to salta-
tion (with empirical constant csalt) and the third
the increase due to the suspended particles
themselves (with yet another empirical constant
cres). Furthermore, it is assumed that suspension
starts only at a critical saltation height of the par-
ticles at the DFL surface. The saltation height,
relative to the particle diameter, is proportional
to a Froude number Frs, computed with the DFL
velocity and the condition for suspension is for-
mulated as
crits
p
sFr
dg
u
Fr ,
3
2
>=
(13)
This condition blocks suspension specifically
when the DFL comes to rest. Values typically
used for csusp, cd, Frs,c rit, R0, csalt and cres are 0.01,
3, 400, 0.1 m, 10-4 and 10, respectively. The
particle diameter dp is assumed in the range 0.5
to 1 mm.
2.3 Powder Snow Layer (PSL) Model
The PSL model in SAMOS-AT treats air and
ice particles as separate phases, i.e. separate
velocities are computed for both (in contrast to
International Snow Science Workshop, Davos 2009, Proceedings
522
International Snow Science Workshop Davos 2009
the initial SAMOS version, which assumed zero
slip velocity). The two phase approach allows for
sedimentation of the particles and a separation
of particle trajectories and air-streamlines, which
may be significant in highly curved terrain.
The well known Navier-Stokes equations
with turbulence-averaging are solved to com-
pute the flow of the air phase. The compressibil-
ity of air is considered via the ideal-gas-law
TRp air
ρ
=
(14)
with the static air pressure p, the air tem-
perature T , which is assumed to be constant
constant (273 K), and Rair the specific gas con-
stant of air. In the following equations, all vari-
ables refer to the air, if not indicated otherwise,
and all dependent variables are turbulence-
averaged. The volume of the ice particles is ig-
nored due to their small volume fraction. A
global, static (Eulerian) coordinate system (co-
ordinates xi) with directions i =1,2,3 is used and
Einsteins summation convention applies to the
equations below. The air mass balance reads
0
)( =
+
i
i
x
u
t
ρ
ρ
(15)
The momentum balance considers the parti-
cle drag fi and the drag fi
(res) resulting from ob-
stacles not representable via the geometric
boundary conditions (e.g. forest):
)(
)(
)(
res
ii
i
j
j
i
eff
j
i
i
j
ji
i
ff
x
u
x
u
x
x
p
g
x
uu
t
u
++
+
+
=
+
η
ρ
ρ
ρ
(16)
The effective viscosity
η
eff consists of the
laminar viscosity
η
lam and the apparent turbulent
viscosity, resulting from turbulent mixing. The
widely used k-
ε
turbulence model (Launder et
al., 1972) is employed to capture the turbulent
flow effects. It requires the solution of additional
transport equations for the turbulent fluctuation
energy k and the turbulent dissipation rate
ε
. For
brevity, these equations are not reproduced
here. Turbulence generation and dissipation due
to the particles are assumed to cancel. With the
k-
ε
-model constant C
µ
(0.09), the effective vis-
cosity reads
ε
ρηη
µ
2
k
C
lame ff += (17)
The resistance force per volume due to for-
est is written analogously to (7) as
u
u
u
c
s
d
f
i
w
res
res
res
i
2
2
)(
ρ
= (18)
The particles are treated as separate rigid
bodies. The drag force exerted by the surround-
ing air onto a particle is formulated as (suffix p
indicates particle values)
p
ipp
p
dip u
ud
u
cF
=,
2
2
,42
π
ρ
(19)
with the drag coefficient c
d
and the particle
diameter d
p
, introduced in 2.2. The slip velocity
ipiiip
uuuu
,,
)(
+= (20)
must consider the turbulent fluctuation part
u’
i
of the air velocity at the particle position. W ith
the k-
ε
–model, a fluctuation velocity can be de-
termined stochastically by picking a random vec-
tor according to a Gaussian distribution with
mean value zero and a standard deviation of the
vector magnitude of
ku
i
3
2
2
=
(21)
The same random velocity is kept as long as
the particle travels with the same turbulent eddy,
which is of size L
turb
and has a life time t
turb
, de-
termined from k and
ε
: This time interval t
t
ends
when either the eddy decays or it is traversed.
=
==
turb
p
turb
t
turbturb
t
u
L
t
k
t
k
CL
,min
5.1
75.0
εε
µ
(22)
The particle drag force in (16) hence can be
formulated as limit of a sum over all particles
within a volume
V around the considered posi-
tion
=
V
p
ipVi
F
V
f
,0
1
lim
(23)
and the momentum balance for a particle
with mass
6/
3
πρ
picep dm =
reads
ipip
ip
pgmF
dt
du
m+= ,
,
(24)
Particles that hit obstacles are removed. The
hit-probability within a time interval
t is
res
p
s
tu
res
res
hit
s
d
P
=
11 (25)
A particle is removed, if a random number in
the range [0,1), queried each time interval
t, is
smaller than P
hit
. W hen a particle hits the DFL
surface, it is always reflected. If it hits the ter-
rain, it is reflected only if the velocity is larger
than a threshold (of 3 ms-1) and the angle be-
International Snow Science Workshop, Davos 2009, Proceedings
523
International Snow Science Workshop Davos 2009
tween its velocity vector and the surface normal
is smaller than 30°. Otherwise it is deposited.
3 NUMERICAL METHODS
3.1 DFL Numerics
The DFL momentum equation (6) is solved
for a large number of mass elements (typically
about 2000 kg each) explicitly in time (usually
steps of 0.1 s). A regular DHM grid with a reso-
lution of typically 5 m is used to represent the
terrain and to define the SPH-kernel-function W
in (2).
Figure 3. DHM-grid (detail), seen from above,
mass element centers (dots), element velocities
(light vectors) and smoothed velocity field at grid
points (bold vectors).
To obtain W, a bilinear interpolation is ap-
plied first to distribute element values to the
nodes of the grid cells containing the element-
centers, and the same bilinear interpolation is
used afterwards to get a smoothed value at the
element center from the values at the surround-
ing grid nodes. SPH methods have the advan-
tage of small numerical diffusion, while they may
show stability problems due to the forces be-
tween the elements. In the DFL the latter are
small (of order H/L), however. Furthermore, an
artificial viscosity is applied to the difference be-
tween element velocity and the smoothed veloc-
ity at the element center.
3.2 PSL Numerics
A fully implicit Finite Volume method is used
to solve the balances of the air flow in a SIMPLE
scheme (Patankar, 1990; for implementation
details, see FIRE Manual, 2009). The momen-
tum balances for the ice-particles are integrated
explicitly in time. Of course, not all particles can
be followed separately. Instead, they are
grouped in parcels (of about 100 kg each) and
the equations are solved for one exemplaric par-
ticle in each parcel. The numerical PSL-mesh is
created by extruding a regular grid (resolution
typically 15 m) in direction of the average terrain
normal. About 20 cell layers with increasing
depth are added, with the depth of the first ap-
proximately 4 m.
Figure 4. 3D-grid for simulation of PSL adapted
to the terrain surface.
4 APPLICATION RESULTS
Model validation, application and compari-
sons to the initial SAMOS version are described
in a separate article (Granig et al., 2009).
5 REFERENCES
FIRE CFD Solver Manual v2009, 2009. AVL List
GmbH, Graz.
Gersten, K. and Herwig, H. 1992.
Strömungsmechanik, Vieweg, Braunschweig,
Wiesbaden.
Granig, M., Sampl, P., Tollinger, C., Jörg, Ph., 2009.
Experiences in avalanche assessment with the
powder snow avalanche model SamosAT.
Accepted for the International Snow Science
Workshop 2009, Davos, Switzerland.
Launder, B.E. and Spalding, D.B., 1972.
Mathematical models of turbulence. Academic
Press, London.
Monaghan, J.J., 1988. An introduction to SPH.
Computer Physics Communications, vol. 48, pp.
88-96.
Patankar, S.V., 1980. Numerical heat transfer and
fluid flow. McGraw-Hill, New York.
Sailer, R., Fellin, W., Fromm, R., et al. 2008. Snow
avalanche mass-balance calculation and
simulation-model. Annals of Glaciology, Volume
48, 183-192.
Sampl, P., Zwinger, T., 2004. Avalanche simulation
with SAMOS. Annals of Glaciology, Volume 38,
393-398.
Savage, S.B. and Hutter, K., 1989. The motion of a
finite mass of granular material down a rough
incline. J. Fluid Mech. 199, 177-215.
Zwinger, T., Kluwick, A., Sampl, P., 2003. Simulation
of Dry-Snow Avalanche Flow over Natural
Terrain. In: Hutter, K., Kirchner, N. (Editors.),
Dynamic Response of Granular and Porous
Materials under Large and Catastrophic
Deformations, Lecture Notes in Applied and
Computational Mechanics. Volume. 11, Springer,
Heidelberg, p. 161-194.
... Tools for the simulation of snow avalanches include a wide range of flow models and numerical implementations (e.g. 40 Christen et al., 2010b; Sampl and Zwinger, 2004;Zugliani and Rosatti, 2021;Li et al., 2021;Hergarten and Robl, 2015;Mergili et al., 2017;Rauter et al., 2018;Oesterle et al., 2022). Their tasks range from simulations for regional avalanche terrain analysis (Toft et al., 2023), to identify endangered terrain for hazard zone mapping or protection forest classification to detailed simulations used for dimensioning mitigation measures. ...
... For large scale or large area simulations, conceptual data driven models such as Flow-Py (D'Amboise et al.,45 2022) are used, but also process based, physical models (Issler et al., 2023;Bühler et al., 2018). Classically detailed simulations are performed for operational engineering practice with tools such as RAMMS (Christen et al., 2010b), the former SamosAT (Sampl and Zwinger, 2004) and now AvaFrame (Oesterle et al., 2022); or research questions are investigated in a scientific setting using OpenFOAM (Rauter et al., 2018) or lately for example the MPM method (Li et al., 2021). As the model parameters of complex avalanche flow models, such as the friction parameters, cannot practically be determined directly by laboratory 50 or field experiments, flow model applications rely on parameter suggestions from guidelines (Gruber and Bartelt, 2007) or parameter optimization through back calculations (Ancey et al., 2003). ...
... For avalanche simulation optimization the runout or deposition area is mostly used to determine the best-fit model parameters 170 (Sampl and Zwinger, 2004;Christen et al., 2010a;Bühler et al., 2011). These areas rely more on observations made after the event, making them easier to obtain and more numerous compared to inflow measurements taken during an avalanche experiment. ...
Preprint
Full-text available
Understanding particle motion in snow avalanches is essential for unravelling the driving processes behind transport phenomena and mobility. Our approach to investigating avalanche dynamics at the particle level combines data from a novel inflow sensor system, the AvaNodes, with radar measurements and simulation results from the thickness integrated flow module of AvaFrame, the open avalanche framework. The radar measurements offer a comprehensive view of the avalanche, serving as a reference for the AvaNodes' trajectories within it. This synthesis provides a holistic overview of the motion of avalanche particles and the front. The utilized com1DFA module in AvaFrame, equipped with a numerical particle grid method, enables a direct implementation of numerical particle tracking functionalities, facilitating a comparison between measurements and simulations. This unique combination prompts questions about the comparability of simulations and measurements on a particle level, yielding new insights into the thickness integrated model's ability to replicate real-scale snow avalanche particle behaviour assuming a modified Voellmy friction relation. Our work also highlights current limitations of comparing radar measurements and synthetic particle sensor systems with numerical simulation particles. Minimizing the differences between measured and simulated particle velocities and front positions allows to identify optimal parameter settings for an observed avalanche event at the Nordkette test site. Using the best-fit parameter values yields deviations below 5–10 % for the maximum velocities and the resulting travel lengths. Beyond the best-fit simulations, the applied optimization method shows a wide range of suitable parameter sets causing equifinality within the investigated parameter space. Additionally, the results show that there is a trade-off between the accuracy of an optimization on single observables or the simultaneous optimization of particle and front behaviour.
... Information on avalanche occurrences is crucial for many safety-related applications: for hazard mitigation, the dimensions of past avalanches are crucial for planning new and evaluating existing protection measures (e.g., Rudolf-Miklau et al., 2015). For the derivation of risk scenarios and the estimation of avalanche frequency, past events are an important piece of information as well (Bründl and Margreth, 2015). ...
... Numerous instance segmentation models have been proposed in recent years that are based on the superior image-understanding capabilities of deep learning. Besides the quest for fully automatic methods, there is also an area of research dedicated to interactive object segmentation (IOS), where a human collaborates with the computer vision model to segment the desired object with high accuracy but low effort (Boykov and Jolly, 2001;Gulshan et al., 2010;Xu et al., 2016;Sofiiuk et al., 2020;Kontogianni et al., 2020;Lin et al., 2022;Kirillov et al., 2023). The human operator explicitly controls the segmentation, first by an initial input to mark the desired object (e.g., through a click or scribbles) and then by iteratively adding annotations to correct the segmentation where the automatic model makes mistakes, gradually refining the result. ...
Article
Full-text available
For many safety-related applications such as hazard mapping or road management, well-documented avalanche events are crucial. Nowadays, despite the variety of research directions, the available data are mostly restricted to isolated locations where they are collected by observers in the field. Webcams are becoming more frequent in the Alps and beyond, capturing numerous avalanche-prone slopes. To complement the knowledge about avalanche occurrences, we propose making use of this webcam imagery for avalanche mapping. For humans, avalanches are relatively easy to identify, but the manual mapping of their outlines is time intensive. Therefore, we propose supporting the mapping of avalanches in images with a learned segmentation model. In interactive avalanche segmentation (IAS), a user collaborates with a deep-learning model to segment the avalanche outlines, taking advantage of human expert knowledge while keeping the effort low thanks to the model's ability to delineate avalanches. The human corrections to the segmentation in the form of positive clicks on the avalanche or negative clicks on the background result in avalanche outlines of good quality with little effort. Relying on IAS, we extract avalanches from the images in a flexible and efficient manner, resulting in a 90 % time saving compared to conventional manual mapping. The images can be georeferenced with a mono-photogrammetry tool, allowing for exact geolocation of the avalanche outlines and subsequent use in geographical information systems (GISs). If a webcam is mounted in a stable position, the georeferencing can be re-used for all subsequent images. In this way, all avalanches mapped in images from a webcam can be imported into a designated database, making them available for the relevant safety-related applications. For imagery, we rely on current data and data archived from webcams that cover Dischma Valley near Davos, Switzerland, and that have captured an image every 30 min during the daytime since the winter of 2019. Our model and the associated mapping pipeline represent an important step forward towards continuous and precise avalanche documentation, complementing existing databases and thereby providing a better base for safety-critical decisions and planning in avalanche-prone mountain regions.
... With a change in the needs of consultants from targeting good and reliable hazard maps to more complex tasks, such as, determining pressure criteria, estimating avalanche flow times, dimensioning mitigation measures, and assessing climate change impact, changes must also come to the modelling tools designed to assist consultants. The prevalent numerical methods, such as RAMMS, DAN3D, SAMOS AT and others, were designed with optimisations towards runout modelling and provide valuable information for consultants within that context (Savage and Hutter, 1989;Christen et al., 2010;Sampl and Zwinger, 2004;McDougall and Hungr, 2004). However, these models often rely on oversimplified assumptions that inadequately capture the velocity-dependent properties of avalanches, such as impact pressures. ...
Conference Paper
Full-text available
Understanding the limitations of dynamical modelling approaches is essential to providing accurate and useful information to clients. In avalanche consultancy, one may find themselves addressing many avalanche problems at short notice, for example, assessing the impact of many avalanche paths along critical infrastructure. Fast-running dynamical models, such as the depth-averaged Voellmy friction-law type, serve as indispensable tools in this context, swiftly providing insights into avalanche properties like runout distances. While these models perform well in predicting runout, their optimisation for speed and scale may lead to the neglect of other crucial properties. For instance, they are implicitly restricted to dry dense configurations, with constant density assumptions neglecting powder or fluidised effects. Additionally, their empirical parameterisation on limited runout data raises concerns about their applicability across diverse climate and topographic regions. Moreover, their reliance on initial avalanche volume may not accurately capture the physical properties of real avalanches, introducing uncertainties. For certain consultancy applications, some models may outperform others. Improved practices are essential for generating more accurate runout maps, determining pressure criteria, estimating avalanche flow times, dimensioning mitigation measures, and assessing climate change impacts. A possible solution lies in leveraging adaptable frameworks that offer consultants a range of modelling options tailored to specific applications. For instance, if avalanche speed is crucial, more sophisticated depth-averaged approaches may be preferable. For scenarios involving transitions from wet snow avalanches to slush flows, multi-layer or multi-phase models may offer better insights. Likewise, for dimensioning purposes, models like Material Point Method (MPM) or 3D continuum methods could prove more effective, accounting for depth variations. This paper will discuss the importance of understanding and addressing the limitations of dynamical avalanche models in consultancy. By adopting versatile modelling frameworks, consultants can tailor their approaches to diverse applications, thereby improving the accuracy and reliability of their assessments for clients.
... The simulation software SAMOS for snow avalanches has been used by the Austrian Service for Torrent and Avalanche Control since 1999. The software is based on a two-dimensional, depth-averaged model with Coulombian and turbulent bottom friction for the dense flow and on a three-dimensional mixture model for air and ice-particles, which forms the powder snow part (Sampl and Zwinger, 2004). An improved version, SAMOS-AT, has been completed in 2007. ...
Conference Paper
Full-text available
The Lillet Lake avalanche site is in the Piedmont region, northwest Italian Alps, in the community of Ceresole Reale, Province of Turin, close to the border to France. In the runout zone of the avalanche is an inhabited hamlet called "Mua", with a small part referred to as "Inverso di Mua" on the other side of the Orco River. Several potential avalanche-starting zones endanger the site. The highest ones are located upslope of a small alpine lake, the Lillet Lake, giving the avalanche its name. Historical information about this avalanche and the exposure of Mua hamlet to avalanches is rather controversial. Of particular significance is the historical data for an event that occurred in 1972 in which the snow masses released upslope from Lake Lillet, causing it to overflow and generating a slushflow-like event, which was able to reach the valley bottom. However, the information on the effects of this event is unclear as available sources report different data for the affected area. To better investigate the exposure of the site to slushflow hazard, a modeling analysis with the state-of-the-practice avalanche dynamic models RAMMS and SAMOS has been performed, testing their suitability for the mod-eling of slushflow processes. At the time their performance was also compared with respect to mixed snow avalanche modeling. The analysis carried out confirms the potential exposure of "Inverso di Mua" to slushflow hazard and highlights its possible exposure to powder avalanches too. It emerged from the study that there is a need for further research in the field of slush-flow modeling, including a dedicated calibration of the current avalanche dynamic model. RAMMS and SAMOS have proven to be very flexible models able to simulate a wide range of snow-related flows, ranging from slushflows to powder snow avalanches, although non-negligible differences in the model's outputs may emerge as a result of the different approaches and choice of modeling parameters that have to be set by the user.
... al., 2018), Balkan Peninsula (Panayotov, 2007;Panayotov, 2011;Voiculescu et al., 2011;Pop et al., 2017) and South America (Mundo et al., 2007;Casteller et al., 2008). Once a record of avalanche activity is established it can successfully be used for modeling and improvement of local hazard mapping using modern model-based approaches such as the computer software RAMMS:EXTENDED (Casteller et al., 2008;Jamieson et al., 2008;Christen et al., 2010), SamosAT (Sampl and Zwinger, 2014), or Flow-Py (Horton et al., 2013;D'Amboise et al., 2022). ...
... For the derivation of risk scenarios and the estimation of avalanche frequency, past events are an important piece of information as well (Bründl and Margreth, 2015). Mapped avalanches are also used to validate and further develop numerical avalanche simulation software like SAMOS or RAMMS (Sampl and Zwinger, 2004;Christen et al., 2010). Today information on occurred 25 avalanches is still mainly reported and collected at isolated locations, unsystematically by observers and (local) avalanche warning services though more recent research has proposed using satellite imagery (e.g., Eckerstorfer et al., 2016;Wesselink et al., 2017;Bianchi et al., 2021;Hafner et al., 2022). ...
Preprint
Full-text available
For many safety-related applications such as hazard mapping or road management, well documented avalanche events are crucial. Nowadays, despite research into different directions, the available data is mostly restricted to isolated locations where it is collected by observers in the field. Webcams are getting more frequent in the Alps and beyond, capturing numerous avalanche prone slopes several times a day. To complement the knowledge about avalanche occurrences, we propose to make use of this webcam imagery for avalanche mapping. For humans, avalanches are relatively easy to identify, but the manual mapping of their outlines is time intensive. Therefore, we propose to support the mapping of avalanches in images with a learned segmentation model. In interactive avalanche segmentation (IAS), a user collaborates with a deep learning model to segment the avalanche outlines, taking advantage of human expert knowledge while keeping the effort low thanks to the model's ability to delineate avalanches. The human corrections to the prediction in the form of positive clicks on the avalanche or negative clicks on the background result in avalanche outlines of good quality with little effort. Relying on IAS, we extract avalanches from the images in a flexible and efficient manner, resulting in a 90 % time saving compared to conventional manual mapping. If mounted in a stable position, the camera can be georeferenced with a mono-photogrammetry tool, allowing for exact geolocation of the avalanche outlines and subsequent use in geographical information systems (GIS). In this way all avalanches mapped in an image can be imported into a designated database, making them available for the relevant safety-related applications. For imagery, we rely on current and archive data from webcams that cover the Dischma valley near Davos, Switzerland and capture an image every 30 minutes during daytime since the winter 2019. Our model and the associated mapping pipeline represent an important step forward towards continuous and precise avalanche documentation, complementing existing databases and thereby providing a better base for safety-critical decisions and planning in avalanche-prone mountain regions.
Preprint
Full-text available
Depth-averaged models of snow avalanches have hitherto lacked an objective arrest criterion. In this study, we investigate the stoppage mechanisms of simulated avalanches, considering the interplay between mesh resolution, simple/complex topographies, and cohesion. We use a second-order depth-averaged model including a modified Voellmy model with cohesion and a physical yielding criterion. Simulated results were found to be sensitive to the mesh resolution, until the cell width is less than 20 % of the characteristic flow depth. The yielding criterion is sufficient to unambiguously define flow arrest for highly cohesive avalanches, even on the complex topography with feature sizes comparable to the typical flow depth. In contrast, for weakly-cohesive avalanches on the complex topography, a fully static state is never reached, due to numerical diffusion enhanced by local non-zero slopes. We hence investigated different global and local arrest criteria applied during post-processing, complementing the yielding criterion. Most of these criteria require setting ad-hoc thresholds, the values of which depend on numerous factors. However, tracking the evolution to a static state of the highest point of the flow material in the runout zone appears to offer an objective and practical solution to indicate when the model enters a numerical-diffusion-dominated regime, whereupon simulations can reasonably be terminated.
Article
Full-text available
Avalanche risk assessment is complex and challenging, with terrain assessment as one of the most fundamental factors. To aid people's terrain assessment, Parks Canada developed the Avalanche Terrain Exposure Scale (ATES), a system that classifies the severity of avalanche terrain into five classes from non-avalanche terrain to extreme terrain. Manual classification is laborious and dependent on expert's assessments. To ease the process Larsen et al. (2020) developed an automated ATES model (AutoATES v1.0). Although the model allowed large-scale mapping, it had some significant limitations. This paper presents an improved AutoATES v2.0 model improving the potential release area (PRA) model, utilizing the new Flow-Py runout simulation package. Furthermore, it incorporates forest density data in the PRA, in Flow-Py, and in a newly developed post-forest-classification step. AutoATES v2.0 has also been rewritten in open-source software, making it more widely available. The paper includes a validation of the model measured against two consensus maps made by three experts at two different locations in western Canada. For Bow Summit, the F1 score (a measure of how well the model performs) improved from 64 % to 77 %. For Connaught Creek, the F1 score improved from 40 % to 71 %. The main challenge limiting large-scale ATES classification is the determination of optimal input parameters for different regions and climates. In areas where AutoATES v2.0 is applied, it can be a valuable tool for avalanche risk assessment and decision-making. Ultimately, our goal is for AutoATES v2.0 to enable efficient, regional-scale, and potentially global ATES mapping in a standardized manner rather than based solely on expert judgment.
Article
In the context of global warming and intensified human activities, glacier instability in plateau regions has increased, and glacier debris flows have become active, which poses a significant threat to the lives and property of people and socioeconomic development. The mass movement process of glacier debris flows is extremely complex, so this paper uses the 2018 Sedongpu glacier debris flow event on the Qinghai-Tibet Plateau as an example and applies a numerical simulation method to invert the whole process of mass movement. In view of the interaction between phases in the process of motion, we use the fluid-solid coupling method to describe the mass movement. The granular-flow model and drift-flux model are employed in FLOW3D software to study the mass movement process of glacier debris flows and explore their dynamic characteristics. The results indicate that the glacier debris flow lasted for 700 s, and the movement process was roughly divided into four stages, including initiation, scraping, surging and deposition; the depositional characteristics calculated by the fluid-solid coupling model are consistent with the actual survey results and have good reliability; strong erosion occurs during the mass movement, the clear volume amplification effect, and the first wave climbs 17.8 m across the slope. The fluid-solid coupling method can better simulate glacier debris flows in plateau regions, which is helpful for the study of the mechanism and dynamic characteristics of such disasters.
Article
Full-text available
Following Norem's description of powder-snow avalanche formation and structure, we propose a mathematical model that consists of a suspension layer and a so-called saltation layer. The latter is only a few meters deep and is modelled by depth-averaged mass and momentum balances. In the suspension layer, the mass- and momentum-balance equations for the mixture are supplemented by the snow mass balance and the transport equations for turbulent kinetic energy and dissipation. Mass and momentum exchange between the two layers is determined by particle settling, turbulent diffusion against the concentration gradient and aerodynamic shear forces. The net erosion or deposition rate is a function of the kinetic energy of the impacting particle.s The saltation layer reacts on the suspension layer in that saltating particles extract momentum from the air flow. The preliminary estimates of the model parameters can be refined by means of saltation-trajectory simulations. Three-dimensional simulations with a simplified model have clearly shown the importance of snow erosion and deposition in practical applications. This approach is well suited for coupling to a dense-flow avalanche model.
Article
Die Überarbeitung für die 10. deutschsprachige Auflage von Hermann Schlichtings Standardwerk wurde wiederum von Klaus Gersten geleitet, der schon die umfassende Neuformulierung der 9. Auflage vorgenommen hatte. Es wurden durchgängig Aktualisierungen vorgenommen, aber auch das Kapitel 15 von Herbert Oertel jr. neu bearbeitet. Das Buch gibt einen umfassenden Überblick über den Einsatz der Grenzschicht-Theorie in allen Bereichen der Strömungsmechanik. Dabei liegt der Schwerpunkt bei den Umströmungen von Körpern (z.B. Flugzeugaerodynamik). Das Buch wird wieder den Studenten der Strömungsmechanik wie auch Industrie-Ingenieuren ein unverzichtbarer Partner unerschöpflicher Informationen sein.
Article
Article
Presents introductory skills needed for prediction of heat transfer and fluid flow, using the numerical method based on physical considerations. The author begins by discussing physical phenomena and moves to the concept and practice of the numerical solution. The book concludes with special topics and possible applications of the method.
Article
Dispersions of solid spherical grains of diameter D = 0\cdot 13 cm were sheared in Newtonian fluids of varying viscosity (water and a glycerine-water-alcohol mixture) in the annular space between two concentric drums. The density sigma of the grains was balanced against the density rho of the fluid, giving a condition of no differential forces due to radial acceleration. The volume concentration C of the grains was varied between 62 and 13%. A substantial radial dispersive pressure was found to be exerted between the grains. This was measured as an increase of static pressure in the inner stationary drum which had a deformable periphery. The torque on the inner drum was also measured. The dispersive pressure P was found to be proportional to a shear stress T attributable to the presence of the grains. The linear grain concentration lambda is defined as the ratio grain diameter/mean free dispersion distance and is related to C by lambda =1/(C0/C)1/3-1, where C0 is the maximum possible static volume concentration. Both the stresses T and P, as dimensionless groups Tsigma D2/lambda eta 2 and Psigma D2/lambda eta 2, were found to bear single-valued empirical relations to a dimensionless shear strain group lambda 1/2sigma D2(dU/dy)/eta for all the values of lambda < 12 (C = 57% approx.) where dU/dy is the rate of shearing of the grains over one another, and eta the fluid viscosity. This relation gives T propto \ sigma (lambda D)2 (dU/dy)2 and T propto \ lambda 3/2eta dU/dy, according as dU/dy is large or small, i.e. according to whether grain inertia or fluid viscosity dominate. An alternative semi-empirical relation T = (1 + lambda ) (1 + 1/2lambda ) eta dU/dy was found for the viscous case, when T is the whole shear stress. The ratio T/P was constant at 0\cdot 3 approx. in the inertia region, and at 0\cdot 75 approx. in the viscous region. The results are applied to a few hitherto unexplained natural phenomena.
Article
A physical model for dry snow avalanche flow is presented. The well established model for dense granular flow proposed by Savage and Hutter [21] is adapted to describe the dense, lower part of a snow avalanche. In order to account for the high velocities of snow avalanches, a velocity dependent bed shear stress in addition to the Coulomb law for dry friction is introduced. Since the model has to be applied to arbitrarily shaped topographies, certain simplifying assumptions of geometrical nature must be introduced. In contrast to former numerical implementations of the Savage-Hutter model based on Finite Difference schemes, a Lagrangian Finite Volume method is formulated using integral balance laws. For the powder snow avalanche forming on top of the denser part a mixture model with a constant slip velocity between ice particles and air is introduced. The k-j{cyrillic, ukrainian} turbulence model modified in order to account for buoyancy effects caused by the suspended particles is implemented. The resulting system of equations is solved by applying a Finite Volume scheme on a fixed grid. The transfer of snow mass from the dense flow into the powder snow avalanche is modelled by an analogy to turbulent momentum transfer.
Book
This book offers a modern approach to computational geo- metry, an area thatstudies the computational complexity of geometric problems. Combinatorial investigations play an important role in this study.