Smoothed Particle Hydrodynamics Implementation of the

Standard Viscous-Plastic Sea-Ice Model and Validation in

Simple Idealized Experiments

ORESTE MARQUIS, Atmospheric and Oceanic Sciences

McGill University, Montreal
AUGUST, 2022

A thesis submitted to McGill University in partial fulfillment of the requirements of the

degree of

Master of Science

©ORESTE MARQUIS, 2022-08-10

Table of Contents

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
List of Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
Abrégé . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

1 Foreword 1
1.1 Manuscript Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Author and Co-authors Contributions . . . . . . . . . . . . . . . . . . . . . . 2

2 Introduction 3

3 Model 8
3.1 Smoothed Particle Hydrodynamics (SPH) . . . . . . . . . . . . . . . . . . . . 8
3.2 Momentum and continuity equations . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Constitutive laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4 Governing differential equations: SPH framework . . . . . . . . . . . . . . . 13
3.5 Numerical approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.6 Particle interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.7 Smoothing length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.8 Boundary treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

i
4 Results and Discussion 20
4.1 Plastic wave propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 Ridging experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3 Arch experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5 Conclusions and Summary 36

6 Appendix 41
6.1 Derivations of vector operator in SPH . . . . . . . . . . . . . . . . . . . . . . 41
6.1.1 Divergence of a vector . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.1.2 Divergence of a 2D tensor field . . . . . . . . . . . . . . . . . . . . . . 43
6.1.3 Gradient of a vector field . . . . . . . . . . . . . . . . . . . . . . . . . 44

ii
List of Figures

3.1 Graphical representation of the SPH kernel W (|rp − rq |, lp ) (solid red line),
the smoothing length lp (red arrow), the particle p, the neighbouring par-
ticles q, the support domain A (dashed red line) and the distance between
any neighbour particle q and the particle p within the support domain
rp − rq (black arrow). Note that particles are points in space and that their
size in this schematic is arbitrary. . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Graphical representation of the initial position of the particles and the rele-
vant parameter for the smoothing length evolution : the ice area carried by
the particle Ap (solid red square), the parameter α (= 2 in this schematic for
visibility), the support domain A (dashed red line), the smoothing length
lp (red arrow) and the initial distance between particle Lp . Black circles are
neighbouring particle q and the red circle is the current particle p. Note
that, as for the figure 3.1, the particle sizes in this schematic are also arbi-
trary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

iii
4.1 SPH plastic wave speed as a function of the normalized wavelength (λ/lp )
for the Wendland C 6 kernel. Panel a) show the classical VP rheology with
fixed concentration (Eq. 4.13) normalized by the FDM plastic wave speed
with fixed concentration (Eq. 4.14), panel b) show the classical VP rheology
with a variable concentration and the density definition ρCp = ρi hp Ap (Eq.
4.15) normalized by the FDM plastic wave speed with a variable concentra-
tion (Eq. 4.16) and panel c) show the classical VP rheology with a variable
concentration and the density definition ρp = ρi hp (Eq. 4.17) normalized
by the FDM plastic wave speed with a variable concentration (Eq. 4.16).
Different homogeneous base state of concentration A0 are shown varying
from 0 to 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Idealized domain of the ridging experiment. The blue circles represent the
ice particles and the black ones are the boundary particles. The grey arrow
shows the wind forcing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Temporal evolution of simulated sea-ice thickness along the central hori-
zontal line of the domain for a) the ridge experiment initialized with a con-
centration A = 1 and average thickness h = 1 and b) the ridge experiment
initialized with a concentration A = 0.5 and average thickness h = 0.5. The
wall is located at x = 0 and the wind speed is −5x̂ [m · s−1 ]. The theory
follows Eq. (4.18). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4 Evolution in time of a) the thickness normalized by concentration rate of
d(h/A)
change in time dt
, b) the average thickness h and c) the concentration
df
A at x = 300 [km]. The rate of change in time is computed from dt
(x, t) =
f (x,t+∆t)−f (x,t−∆t)
2∆t
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.5 Idealized domain of the ice arch experiments. The blue circles represent
the ice particles and the black ones are the boundary particles. The grey
arrow shows the wind forcing. . . . . . . . . . . . . . . . . . . . . . . . . . . 32

iv
4.6 Strain rate and stress invariants ( ϵ̇I , ϵ̇II , σI , σII ) at time t = 24 [h] for an
initial particle spacing of a) 7.5, b) 5 and c) 3.75 [km] (8, 12 and 16 particles
can fit in the strait respectively) and the initial total integrated surface stress
at the entry of the channel is 26.325 [kN · m−1 ]. . . . . . . . . . . . . . . . . . 33
4.7 Ice concentration, thickness and total velocity (h, A, |ui |) at time t = 24
[h] for an initial particle spacing of a) 7.5, b) 5 and c) 3.75 [km] (8, 12 and
16 particles can fit in the strait respectively) and the initial total integrated
surface stress at the entry of the channel is 26.325 [kN · m−1 ]. . . . . . . . . . 34
4.8 Thickness field at time t = 0, 48, 168 [h] for an initial particle spacing of
7.5 [km] and a total integrated stress at the entry of the channel of 13.146
[kN · m−1 ]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

v
List of Tables

3.1 Physical parameters used across all simulations . . . . . . . . . . . . . . . . 19

vi
List of Variables

Physical property

t Simulation time

∆ Total deformation

η Shear viscosity

Γ 1D stress proportionality factor

λ Wavelength

ϵ̇ Strain rate tensor

σ Cauchy’s stress tensor

τ Drag force

r General position vector

u Velocity vector

x Ice position vector

ω Angular velocity

ζ Bulk viscosity

A Ice concentration

vii
c Phase speed

h Mean ice thickness

k Wavenumber

m Mass

P Ice strength

S Ice shear strength

Smoothed Particle Hydrodynamics specifics

α A free parameter

κ Boundary force parameter

l Smoothing length

N Number of particles

R Normalize distance in the kernel referential

W SPH kernel function

Other symbols

δf A small element

δij Dirac delta function

x̂ X direction unit vector

ŷ Y direction unit vector

T A given tensor

V A given vector

viii
A An Area

F Fourier transform

S A Surface

V A Volume

f A given function

L A Length

Physics constants

ρ Material density

C Drag coefficient

C∗ Ice strength decay

e Ellipse yield curve aspect ratio

kt Tensile strength factor

P∗ Ice compressive strength

Superscript

C Relative to a classical definition

n Relative to an iteration number

Subscript

0 Relative to unperturbed/initial value

A Relative to the concentration

a Relative to the air

ix
b Relative to a boundary particle

d Relative to dimension

i Relative to the ice

max Relative to a maximum

min Relative to a minimum

p Relative to the current particle

q Relative to a neighbour particle

r Relative to a replacement

SP H Relative to the SPH Viscous-Plastic theory

T OT Relative to a total

VP Relative to the Viscous-Plastic theory

w Relative to the water

x
List of Acronyms

DEM Discrete-Element Method. 4, 5, 31, 33, 35, 37

EVP Elastic-Viscous-Plastic. 4, 34

FDM Finite-Difference Method. iv, xii, xiii, 4, 5, 22, 23, 25, 31, 35–38, 40

FEM Finite-Element Method. 4, 5, 35, 38

FVM Finite-Volume Method. 4, 5

GCM Global Climate Model. 39

LKFs Linear Kinematic Features. 4, 5, 39

MIZ Marginal Ice Zone. 27–29, 37

SPH Smoothed Particle Hydrodynamics. iii, iv, xii–xiv, 8–10, 13–16, 20, 22, 23, 25, 26, 28,
31, 32, 34–40

VP Viscous-Plastic. iv, xii, xiii, 4–7, 13, 20, 23, 25, 36

xi
Abstract

The Viscous-Plastic (VP) rheology with an elliptical yield curve and normal flow rule
is implemented in a Lagrangian modelling framework using the Smoothed Particle Hy-
drodynamics (SPH) meshfree method. Results show, from perturbation analysis of SPH
sea-ice dynamic equations, that the classical SPH particle density formulation expressed
as a function of sea-ice concentration and mean ice thickness, leads to incorrect plastic
wave speed. We propose a new formulation for particle density that gives a plastic wave
speed in line with theory. In all cases, the plastic wave in the SPH framework is dispersive
and depends on the smoothing length (i.e., the spatial resolution) and on the SPH kernel
employed in contrast with its finite difference method (FDM) implementation counter-
part. The steady-state solution for the simple 1D ridging experiment is in agreement with
the analytical solution within an error of 1%. SPH is also able to simulate a stable up-
stream ice arch in an idealized domain representing the Nares Strait in low wind regime
(5.3 [m · s−1 ]) with an ellipse aspect ratio of 2, an average thickness of 1 [m] and free-
slip boundary conditions in opposition to the FDM implementation that requires higher
shear strength to simulate it. In higher wind regime (7.5 [m · s−1 ]) no stable ice arches are
simulated — unless the thickness is increased — and the ice arch formation showed no
dependence on the size of particles contrary to what is observed in the discrete element
framework. Finally, the SPH framework is explicit, can take full advantage of parallel
processing capabilities and show potential for pan-arctic climate simulations.

xii
Abrégé

La rhéologie visqueuse-plastique (VP) avec une courbe de rendement elliptique et une

règle d’écoulement normale est mise en œuvre dans un cadre de modélisation Lagrangien
en utilisant la méthode sans maillage de l’hydrodynamique des particules lissées (SPH).
Les résultats montrent, à partir d’une analyse des perturbations des équations SPH de
la dynamique de la glace de mer, que la formulation classique de la densité des partic-
ules SPH exprimée en fonction de la concentration de la glace de mer et de l’épaisseur
moyenne de la glace conduit à une vitesse incorrecte des ondes plastiques. Nous pro-
posons une nouvelle formulation pour la densité de particules qui donne une vitesse
d’onde plastique conforme à la théorie. Dans tous les cas, l’onde plastique dans le cadre
SPH est dispersive et dépend de la longueur de lissage (c’est-à-dire de la résolution spa-
tiale) et du noyau SPH employé, contrairement à des modèles homologues basés sur la
méthode des différences finies (FDM). La solution à l’état stationnaire pour l’expérience
simple de crête 1D est en accord avec la solution analytique avec une erreur de 1%. Le
SPH est également capable de simuler une arche de glace stable en amont dans un do-
maine idéalisé représentant le détroit de Nares pour un régime de vent faible (5,3 [m · s−1 ])
avec un rapport d’aspect d’ellipse de 2, une épaisseur moyenne de 1 [m] et des condi-
tions limites de glissement libre, contrairement à l’implémentation FDM qui nécessite
une force de cisaillement plus élevée pour la simuler. Dans un régime de vent plus élevé
(7,5 [m · s−1 ]), aucune arche de glace stable n’est simulée — à moins que l’épaisseur ne soit
augmentée — et la formation d’arches de glace ne montre aucune dépendance à la taille
des particules, contrairement à ce qui est observé dans le cadre des éléments discrets.

xiii
Enfin, le cadre SPH est explicite, peut tirer pleinement parti des capacités de traitement
parallèle et présente un potentiel pour les simulations climatiques panarctiques.

xiv
Acknowledgements

I’d like to express my deepest thanks to Bruno Tremblay, who guided me during this
whole process while letting me enough space and liberty to explore what I felt interesting
and produce my own research.
I must also thank Jean-François Lemieux and Mohammed (Shameem) Islam for the con-
structive feedback they gave to my work. I would also like to extend my gratitude to the
Atmospheric and Oceanic Science Department and the sea ice research group for their
enthusiasm in science which promotes an awesome social environment.
I would further like to thank McGill University, Quebec-Océan, National Research Coun-
cil and Arctrain Canada for the opportunities and funding.
Finally, I must also thank my friends in this adventure Sandrine Trotechaud and Alice Le
Guern-Lepage, my sister Enza Marquis and my partner Anne-Frédérique Meilleur, whom
all supported me when I was complaining through this project.

xv
Chapter 1

Foreword

This Master’s Thesis will be adapted to a paper that will be submitted to The Cryosphere
journal in the early fall of 2022. All of the required elements for a thesis are included: an
introduction with a comprehensive review of the relevant literature, a model description
section, research findings with a comprehensive scholarly discussion and a conclusion.
The FORTRAN SPH sea-ice model code was developed from scratch by Oreste Marquis
and is now public. It can be found at https://github.com/McGill-sea-ice/SIMP. Output
data from the SPH sea-ice model along with a version of the model used, simulations pre-
sented in the thesis and the analyzing programs are available at https://10.5281/zenodo.6950156.

1.1 Manuscript Information

Title: Smoothed Particle Hydrodynamics Implementation of the Standard Viscous-Plastic

Sea-Ice Model and Validation in Simple Idealized Experiments
Authors: Oreste Marquis, Bruno Tremblay, Jean-François Lemieux and Mohammed (Shameem)
Islam
To be submitted to: The Cryosphere

1
1.2 Author and Co-authors Contributions

Oreste Marquis designed, developed and coded the model. Oreste Marquis carried out
the literature review, ran all the simulations, analyzed the results with the supervision
and suggestion of Bruno Tremblay. The writing was done in collaboration between Bruno
Tremblay and Oreste Marquis, with comments from Jean-Francois Lemieux and Mo-
hammed (Shameem) Islam.

2
Chapter 2

Introduction

Sea-ice plays an important role in climate (Budikova, 2009). It modulates the radiative
fluxes at the surface and influence the atmospheric energy balance (Gardner, Sharp, 2010;
Södergren et al., 2017). It insulates the relatively warm ocean from the atmosphere be-
cause of its low thermal conductivity and reduces the vertical heat and moisture fluxes
affecting mesoscale atmospheric processes (Maykut, 1982; Kottmeier, Engelbart, 1992;
Haid, Timmermann, 2013; Dethleff, 2013). Sea-ice modifies the seasonal cycle of tem-
perature because of the latent heat released when freezing and absorbed when melting
(Walsh, 1983; Fichefet, Maqueda, 1997). It influence the ocean circulation (Aagaard et al.,
1981; Ohshima et al., 2016) and deep-water formation (Smith et al., 1990; Maqueda, 2004)
when frazil crystals form and brine is rejected in one location and freshwater released in
another. The vertical mixing and convection induced by this ice formation and melting
also brings nutrient-rich benthic water that supports biological productivity and wildlife
(Stirling, 1980, 1997; Arrigo, 2004; Kalenitchenko et al., 2019).

For accurate climate projection of the Earth’s system a sea-ice component must be in-
cluded. Numerical models for geophysical sea-ice have historically employed a con-
tinuum approach where the material is discretized on an Eulerian mesh using various
constitutive relations. For example, the standard Viscous-Plastic model (Hibler, 1979) or

3
modifications (e.g., Elastic-Viscous-Plastic or EVP and Elastic-Plastic-Anisotropic or EPA
Hunke, Dukowicz, 1997; Tsamados et al., 2013), solves a set of partial differential equa-
tions using the finite-difference method (FDM). FDM is the simplest method to discretize
and solve partial differential equations numerically. However, it is based on a local Taylor
series expansion to approximate the continuum equations and construct a topologically
rectangular network of relations between nodes (e.g., Arakawa grids).

Even though the VP (and EVP) rheologies are commonly used to describe sea-ice dy-
namics and are able to capture important large-scale deformation features (Bouchat et al.,
2022; Hutter et al., 2022), they still have difficulties to represent smaller scale properties
(Schulson, 2004; Weiss et al., 2007; Coon et al., 2007) such as Linear Kinematic Features
(LKFs) unless run at very high resolution (≈2 km, Ringeisen et al., 2019; Hutter et al.,
2022). To improve the simulation of small-scale ice features and to alleviate the problem
of FDM with complex geometries (Peiró, Sherwin, 2005), the community also consid-
ered new sea-ice rheologies (Schreyer et al., 2006; Girard et al., 2011; Dansereau et al.,
2016; Ringeisen et al., 2019) and explored different space discretization frameworks like
the finite-element method (FEM) (Rampal et al., 2016; Mehlmann et al., 2021), the finite-
volume method (FVM) (Losch et al., 2010; Adcroft et al., 2019) or the discrete-element
method (DEM) (Hopkins, Thorndike, 2006; Herman, 2016; Damsgaard et al., 2018).

In recent decades, spatial resolution of sea-ice models became comparable to the char-
acteristic length of the ice floes. This makes the continuum assumption of current FDM,
FVM and FEM models questionable. Also, Eulerian models are known to have difficulties
determining the precise locations of inhomogeneity, free surfaces, deformable boundaries
and moving interfaces (Liu, Liu, 2010). These shortcomings have led to an increase inter-
est in the DEM approach. Another advantage of using DEM is that granularity of the ma-
terial (Overland et al., 1998) is directly represented using discrete rigid bodies from which
the physical interactions are calculated explicitly in the hope that larger scale properties

4
naturally emerge. In practice, the emergent properties still depend on the assumed floe
size and the nature of collisions. Nevertheless, DEM easily captures formation of cracks,
leads and large deformation making it a consistent framework for the numerical simula-
tion of granular material like sea-ice (Fleissner et al., 2007).

Despite the shortcomings of the continuum approaches, FDM, FVM and FEM are still
the most commonly used framework in the community because they have been devel-
oped and tested for a longer period, they are well understood, more computationally effi-
cient and easily coupled for synoptic scale simulations. In an attempt to take advantages
of both continuum and discrete formulation, blends between the two approaches have
been proposed — e.g., the finite-discrete element (Lilja et al., 2021) or the material-point
method (Sulsky et al., 2007). Those framework, however, still use a mesh to solve the dy-
namic equations in addition to considering sea ice as discrete elements making them even
more computationally expensive. Finally, a fairly new approach for sea-ice modelling —
also taking from both continuum and discrete framework — uses a Lagrangian meshfree
continuous method called smoothed particle hydrodynamics (SPH) (Lucy, 1977; Gingold,
Monaghan, 1977). This meshfree method is known to facilitates the numerical treatment
and description of free surfaces (Liu, Liu, 2010) which are common in sea-ice dynamics
with polynyas, LKFs, free drifting ice floes and unbounded ice extent. As in DEM, the
physical quantities are carried out by particles in space (an analogy for ice floes in the real
world), but evolve according to the same dynamic equations used in the continuum ap-
proach. Furthermore, the method has the advantage of treating the system of equations
in a Lagrangian framework discretized explicitly making it well suited for parallelization.

SPH has been used successfully for the modelling of other granular materials such as
sand, gravels and soils (Salehizadeh, Shafiei, 2019; Yang et al., 2020; Sheikh et al., 2020).
In the context of mesoscale and larger sea-ice modelling, Gutfraind, Savage (1997) initi-
ated the SPH study of sea-ice dynamics using a VP rheology based on a Mohr-Coulomb

5
failure criterion. The ice concentration and thickness were fixed at 100% and 1 [m] with
a continuity equation expressed in terms of a particle density. The internal ice strength
between particle was derived diagnostically from ice density assuming ice was a nearly
incompressible material. Later, Wang (2000) developed a sea ice model of the Bohai Sea
(east coast of China) using an SPH viscous-plastic rheology (Hibler, 1979) with continuity
equations for ice concentration and mean thickness, and ice strength calculated from static
ice jam theory (Shen et al., 1990). Following Wang (2000), Ji et al. (2005) implemented a
new viscoelastic-plastic rheology that was in better agreement with observations from
the Bohai Sea. Recently, Staroszczyk (2017) proposed a sea ice model considering ice
to behave as a compressible non-linear viscous material with a (dimensionless) contact
length dependent parameterization for floe collisions and rafting (Gray, Morland, 1994;
Morland, Staroszczyk, 1998). In all of the above, except for Gutfraind, Savage (1997), the
same ice particle density definition is used.

In this work, we use the standard VP sea-ice model with an elliptical yield curve and
normal flow rule (Hibler, 1979), and propose a reformulation of the ice particle density
that is internally consistent with the model physics. One goal of the study is to investigate
differences coming from the numerical framework. To this end, we theoretically investi-
gate the plastic wave propagation, a fundamental property of a sea-ice physical model,
throughout a 1D perturbations analysis and we test the model in a ridging and ice arch
experiment following earlier works by Lipscomb et al. (2007); Dumont et al. (2009); Ra-
batel et al. (2015); Dansereau et al. (2017); Williams et al. (2017); Damsgaard et al. (2018);
Ranta et al. (2018); Plante et al. (2020); West et al. (2022). We chose to investigate the
SPH method performance on a ridging experiment since it has an analytical steady-state
solution that can be used to establish the model accuracy and it is possible to evaluate
the coherent evolution of the continuity equations. We also test SPH performance on ice
arches simulation since this classic problem is an example of large-scale features resulting
from small-scale interaction involving fractures of the material. The two experiments al-

6
low a direct comparison with previous work and identify advantages and disadvantages
with the continuum and discrete sea-ice dynamic. The long-term goal is to lay the foun-
dation for an SPH sea-ice formulation that can be used in synoptic scale models.

The thesis is organized as follows. In chapter 3, the SPH framework and how the sea-
ice VP rheology, momentum and continuity equations can be implemented in this frame-
work are described. Results of a plastic wave propagation analysis, ridging experiments,
and ice-arching simulations are presented in the chapter 4. Finally, chapter 5 discuss the
SPH advantages and limitations of the framework and model developed and concludes.

7
Chapter 3

Model

3.1 Smoothed Particle Hydrodynamics (SPH)

The SPH method is at the interface between finite element method and discrete element
methods. In this framework any function f (r) at a point r is approximated from neigh-
bouring values in the parameter space f (r′ ) using an integral interpolant (see figure 3.1):

Z
f (r) = f (r′ )W (|r − r′ |, l)dr′ , (3.1)
V

where W (|r − r′ |, l) is the interpolating kernel and V is the entire space volume. In two
dimensions, the space volume is an area A and the kernel has units of [m−2 ]. This integral
interpolant approximation is based on the singular integral mathematical framework of
Natanson (1961) and imposes the following restrictions on the kernel:

Z
W (|r − r′ |, l)dr′ = 1, (3.2)
A

and
lim W (|r − r′ |, l) = δ(r − r′ ), (3.3)
l→0

8
Figure 3.1: Graphical representation of the SPH kernel W (|rp − rq |, lp ) (solid red line), the
smoothing length lp (red arrow), the particle p, the neighbouring particles q, the support
domain A (dashed red line) and the distance between any neighbour particle q and the
particle p within the support domain rp − rq (black arrow). Note that particles are points
in space and that their size in this schematic is arbitrary.

where l is the smoothing length of the kernel and δ is the Dirac delta function. Using
the particle approximation, Eq. (3.1) can be written as a weighted summation over all
neighbouring points within the area A:

N N
X X mq
f (rp ) ≈ f (rq )W (|rp − rq |, lp )∆Aq ≈ f (rq )W (|rp − rq |, lp ) , (3.4)
q=1 q=1
ρq

where N is the number of points in space referred as neighbour particles, ∆Aq (= m/ρ) is
the area associated with the particle p, m represent the mass [kg] and ρ is the 2D density
[kg · m−2 ]. From the above approximations, we reformulate differential operators relevant
to our study in their discrete SPH forms. We write the divergence of a vector field (V),
the divergence of a tensor (T) and the gradient of a vector field (V) (Monaghan, 2005) as

9
(see Appendix 6.1 for complete derivation) :

N
1 X
(∇ · V)p = mq (V(rq ) − V(rp )) · ∇p Wpq , (3.5)
ρp q=1
N  
X T(rq ) T(rp )
(∇ · T)p = ρp mq 2
+ 2
· ∇p Wpq , (3.6)
q=1
ρ q ρ p

N
X mq
(∇V)p = (V(rq ) − V(rp )) ⊗ ∇p Wpq . (3.7)
q=1
ρq

In Eq. (3.7), ⊗ denotes the outer product. ∇p Wpq is the gradient of the kernel at the
coordinate rp − rq in the reference frame of particle p and is written as :

rp − rq ∂W (|rp − rq |, lp )
∇p Wpq = . (3.8)
|rp − rq | ∂|rp − rq |

Note that Wpq is a scalar function and consequently ∇p Wpq is a vector, the inner product
in Eq. (3.5) is a scalar, the inner product in Eq. (3.6) is a 2D vector and the outer product
in Eq. (3.7) is a 2D tensor of rank 2. In addition to Eq. (3.2 - 3.3), the smoothing kernel
must have the following set of properties to avoid non-physical behaviour and costly
computation (Liu, Liu, 2003):

Compact support : W (|rp − rq |, lp ) = 0, for |rp − rq | > lp , (3.9)

Positive definite : W (|rp − rq |, lp ) ≥ 0, (3.10)

∂W (|rp − rq |, lp )
Monotonically decreasing : ≤ 0, (3.11)
∂(|rp − rq |)
Symmetric : W (|rp − rq |, lp ) = W (−|rp − rq |, lp ), (3.12)
∂ n W (|rp − rq |, lp )
Differentiable : ∃, (3.13)
∂(|rp − rq |)n

where ∃ stands for exist. In the above, differentiable means that the kernel derivatives
exist up to the highest order present in the equations. Finally, to ensure the consistency
of the SPH method approximations to the nth order, all kernel moments of order 1 to n

10
need to vanish. In practice, the consistency conditions are satisfied when the number
of neighbouring particles is sufficiently large to be evenly distributed in the domain of
influence (Fraga Filho, 2019). Note that, at the boundaries, the domain of influence of the
particle is truncated making it impossible to satisfy the kernel moments equations. This
phenomenon is referred as the particle inconsistency and leads to poorer approximations
of physical properties. No clear solutions to this problem are proposed in the literature
yet.

3.2 Momentum and continuity equations

Following Plante et al. (2020), we consider sea-ice to behave as a two-dimensional gran-

ular material described by the 2D momentum equation (neglecting the Coriolis and sea
surface tilt terms):
Du
ρi h = ∇ · σ + τ, (3.14)
Dt

where ρi is the ice density, h is the mean ice thickness (ice volume over an area), u =
ux̂ + vŷ is the ice velocity vector, σ is the vertically integrated internal stress tensor acting
in the ŷ direction on a face with a unit outward normal pointing in the x̂ direction, τ is the
D ∂
sum of water drag and surface air stress and Dt
= ∂t
+ u · ∇ is the Lagrangian derivative
operator. We neglected the Coriolis and sea surface tilt force in the momentum equation
to make it easier to validate the model and study the ice arch formation. Note that using
the Lagrangian derivative operator naturally incorporates the advection of momentum in
the ice dynamics — a term that is typically neglected for most continuum based Eulerian
sea-ice models. The surface air stress and the water stress can be written using bulk
formulation as (McPhee, 1979):

τ = ρa Ca |ua − u|(ua − u) + ρw Cw |uw − u|(uw − u), (3.15)

≈ ρa Ca |ua |(ua ) + ρw Cw |uw − u|(uw − u), (3.16)

11
where ρa and ρw are air and water densities, ua and uw are air and water velocity vectors,
Ca and Cw are the air and water drag coefficients and where u is neglected in the formula-
tion of the wind stress since u ≪ ua . The continuity equations for the mean ice thickness
h and the ice concentration A can be written as:

Dh
+ h∇ · u = 0, (3.17)
Dt
DA
+ A∇ · u = 0, (3.18)
Dt

where the thermodynamic source terms are omitted.

3.3 Constitutive laws

The constitutive relations for the viscous-plastic ice model with an elliptical yield curve,
a normal flow rule and tensile strength can be written as (Beatty, Holland, 2010):

 
Pr (1 − kt )
σij = 2η ϵ̇ij + (ζ − η)ϵ̇kk − δij , (3.19)
2
   
1 ∂uj ∂ui 1 ⊺
ϵ̇ij = + = ∇u + ∇u , (3.20)
2 ∂xi ∂xj 2

where ϵ̇ij is the symmetric part of the strain-rate tensor, ζ and η are the non-linear bulk
and shear viscosities, Pr is the replacement pressure, kt is the tensile strength factor and
δij is the Kronecker delta. Following Bouchat, Tremblay (2017) we write :

P (1 + kt )
ζ= , (3.21)
2∆∗
 2
ζ 2S
η= 2 =ζ , (3.22)
e P (1 + kt )
∆∗ = max(∆, ∆min ), (3.23)
 1/2
2 2 −2 −2 2 −2
∆ = (ϵ̇11 + ϵ̇22 )(1 + e ) + 4e ϵ̇12 + 2ϵ̇11 ϵ̇22 (1 − e ) , (3.24)

12
where P = P ∗ h exp(−C(1−A)) is the ice strength (Hibler, 1979), P ∗ and C are respectively
the ice compressive strength and ice concentration parameters, S is the ice shear strength
and e is the ellipse aspect ratio. In the limit where the strain rates ϵ̇ go to zero, ζ and η
would tend to infinity. To avoid this situation, the deformation ∆ is capped to ∆min =
2 × 10−9 s−1 . Using the ∆∗ formulation, the replacement pressure Pr can be written as

∆
Pr = P , (3.25)
∆∗

which ensures that the stresses are zero when the strain rates are zero.

3.4 Governing differential equations: SPH framework

To solve ice dynamic system of equations in the SPH framework, equations involving
spatial derivatives (Eqs. 3.14 - 3.17 - 3.18 - 3.20) are reformulated using Eqs. (3.5 - 3.6 -
3.7) with the particle subscripts p and q (see Fig. 3.1) and a temporal evolution for the ice
particle position is defined:

Dxp
= up , (3.26)
Dt
N  
Dup X σq σp
ρi hp = ρp mq 2 + 2 · ∇p Wpq + τp , (3.27)
Dt q=1
ρq ρp
N
Dhp hp X
+ mq (uq − up ) · ∇p Wpq = 0, (3.28)
Dt ρp q=1
N
DAp Ap X
+ mq (uq − up ) · ∇p Wpq = 0, (3.29)
Dt ρp q=1
 XN  XN ⊺ 
1 mq mq
(ϵ̇ij )p = (uq − up ) ⊗ ∇p Wpq + (uq − up ) ⊗ ∇p Wpq . (3.30)
2 q=1
ρ q q=1
ρq

It is important to make the distinction between the intrinsic ice density ρi and the particle
densities ρp . For consistency reasons with the standard VP rheology, we consider the
following definition of density independent of ice concentration in contrast with previous

13
work (Wang, 2000; Ji et al., 2005; Staroszczyk, 2017) (see results section for discussion):

ρp = ρi hp . (3.31)

By formulating density as Eq. (3.31), the continuity Eq. (3.28) has the same form as the
more commonly used continuity density equation (Monaghan, 2012) :

N
Dρp X
= −ρp ∇ · up = mq (up − uq ) · ∇p Wpq , (3.32)
Dt q=1

except for the fact that the divergence of the velocity field is scaled by the ice material
density ρi ( Dρ
Dt
p
= ρi Dh
Dt
p
). Note that since the particle density ρp is independent of the
concentration, the particle concentration Ap is a quantity that measures the compactness
of the floes at the particle location, but does not relate to the amount of ice carried by a
particle. With this formulation, the concentration can be interpreted as the probability
of ice floes carried by a particle to come in contact with ice floes of another particle (and
repel each other) within the unresolved area ∆Ap .

3.5 Numerical approach

Following Hosseini et al. (2019), we use a second order predictor-corrector scheme to

evolve in time the SPH ice system of equation (see algorithm 1 below). This integration
scheme takes a given function f (here f can be x, u, A and h) and used a predictor step to
calculate its value f n+1/2 at time t = (n + 21 )∆t (where ∆t is the time step) followed by a
correction step to calculate the solution f n+1 at time t = (n + 1)∆t from f n+1/2 :

∆t Dfpn
fpn+1/2 = fpn + + O(∆t2 ), (3.33)
2 Dt
n+1/2
n+1/2 ∆t Dfp
fp corrected = fpn + , (3.34)
2 Dt
n+1/2
fpn+1 = 2fp corrected − fpn + O(∆t3 ). (3.35)

14
Following Lemieux, Tremblay (2009), a simple 1D model taking into account only the
viscous term — the most restrictive condition — leads to the following stability criterion:

2
ρi hlmin e2 ρi lmin
2
∆min
∆t ≤ = ∗
, (3.36)
ηmax P (1 + kt )

where lmin is the minimum smoothing length across all the particles.

Algorithm 1 Sea-ice SPH

Require: Domain shape and boundaries, Spatial resolution, Total integration time
initialize particle and boundary according to input
for i = 0 to IntegrationT ime do
nInteraction ← nearestN eighbourP articleSearch
for j = 0 to nInteraction do
kernel ← smoothingF unctionCalculation
internalF orce ← kernel
end for
for all particles do
externalF orce
physicalQuantities ← (externalF orce,internalF orce)
density ← iceT hickness
smoothingLength ← density
end for
timeStep ← smoothingLength
monitor particle interaction statistics
output
end for

3.6 Particle interactions

Following Rhoades (1992), we use the bucket search algorithm parallelized using shared
memory multiprocessing (OpenMP) to find all the neighbours of each particle in favour
of the explored tree algorithm (Cavelan et al., 2019) which involve pointers and complex
memory structure that are not easy to manipulate in OpenMP.

After the neighbour search, the interactions between pairs of particles are computed us-

15
ing the Wendland C 6 kernel — Wendland kernels have the best stability properties for
wavelengths smaller than the smoothing kernel (Dehnen, Aly, 2012) — which is written
as:


(1 − R)8 (32R3 + 25R2 + 8R + 1), 0 ≤ R < 1,


W (|rp − rq |, lp ) = WC 6 (R) = αd

0,
 R ≥ 1,

(3.37)

−22R(16R2 + 7R + 1)(1 − R)7 lκp ,

0 ≤ R < 1,

∂W (|rp − rq |, lp ) ∂WC 6 (R) 
= = αd
∂|rq − rp | ∂|rq − rp | 
0,
 R ≥ 1,

(3.38)

where αd is a normalization factor depending on the dimension of the problem. Note

that R (= κ|rp − rq |/lp ) is the normalized distance between particles in the referential
rp − rq . Consequently, we always integrate from 0 to lp (the smoothing length) indepen-
dently of the kernel instead of 0 to κlp as shown in (Liu, Liu, 2010). The constant αd
78κ2
becomes 7πl2
in 2D, with a factor of κ2 different from the usual definition. Note that the
scaling factor κ has a value of 1 for the Wendland C 6 kernel. The choice of kernel was
validated using stability tests with six different kernels including the original Gaussian
kernel (Gingold, Monaghan, 1977), a quartic spline Gaussian approximation (Liu, Liu,
2010), a quintic spline Gaussian approximation (Morris et al., 1997), a quadratic kernel
(Johnson, Beissel, 1996) and the Wendland C 2 , C 4 and C 6 kernels (Wendland, 1995).

3.7 Smoothing length

The smoothing or correlation length is a key element of SPH and has a direct influence
on the accuracy of the solution and the efficiency of the computation. For instance, if lp is
too small, there may not be enough particles in the support domain violating the kernel
moments requirements. If the smoothing length lp is too large, all the local properties

16
of particles would be smoothed out over large number of neighbours and the computa-
tion time would increase with the number interactions. In two dimensions the optimal
number of neighbours interacting with any particle p should be about 20 to balance the
precision and the computational cost (Liu, Liu, 2003). We therefore implement a variable
smoothing length that evolves in time and space to maintain this approximate number
of neighbours. To this end, we keep the mass of particles constant in time and evaluate
the smoothing length from the particle density. Note that keeping the mass of a particle
constant has the advantage of ensuring mass conservation. This assumption is justified
in our case since we are only interested in sea-ice dynamics and ridging change the area
cover by ice floes but not their mass. However, fixing the ice mass is only valid when
neglecting the thermodynamics and need to be modified for synoptic scale simulation.

The initial mass of a particle is defined from the ice area it represents within its support
domain (∆Ap in Fig. 3.2). To avoid creating porosity in the medium, we divide the space
in equal square area (= L2p ) that covers the whole domain. Since we want approximately
20 neighbours for every particle, we introduce α (= 3 in all simulations) a parameter that
stands for the approximate number of particles desired in any direction within the sup-
port domain. The parameter α can also be interpreted as the proportionality constant
between the particle spacing Lp and the smoothing length lp . Note that to increase accu-
racy of the particle approximation, α can be increased by any desired factor (see Fig. 3.2).
The mass carried by a particle is therefore written as :

mp = Ap ρi h0p = L2p ρi h0p , (3.39)

where h0p is the initial mean thickness of the particle. The smoothing length is then up-
dated at each time step diagnostically from:

mp
r
lp = αLp = α . (3.40)
ρp

17
The smoothing length lp is capped to 10 times its initial value when the particle density

Figure 3.2: Graphical representation of the initial position of the particles and the rele-
vant parameter for the smoothing length evolution : the ice area carried by the particle
Ap (solid red square), the parameter α (= 2 in this schematic for visibility), the support
domain A (dashed red line), the smoothing length lp (red arrow) and the initial distance
between particle Lp . Black circles are neighbouring particle q and the red circle is the cur-
rent particle p. Note that, as for the figure 3.1, the particle sizes in this schematic are also
arbitrary.

tends to zero. This capping prevents conservation of mass for density lower than 1% of
its initial value (see Eq. (3.39)). We justify this capping because such small densities do
not affect the ice dynamics.

3.8 Boundary treatment

We implemented the boundary treatment of Monaghan, Kajtar (2009) because of its sim-
plicity, versatility and low computational cost. The boundaries are set up by placing
stationary particles with fixed smoothing length lb and a mass mb equal to the average
ice particle mass mp . The boundary smoothing length lb is chosen in a way that only one

18
layer of ice particles initially interact with the boundary (this makes lb resolution depen-
dent). The boundary particles are (equally) spaced apart by a factor one quarter of their
smoothing length (lb /4). In this manner, all ice particles p within a support domain lb will
interact with approximately four boundary particles (denoted by the subscript b) at a time
resulting in a net normal repulsive force FNp :

Nb
X (rp − rb ) 2mb
FNp = κn 2
Wpb , (3.41)
b=1
|rp − rb | mp + mb

that is added to their momentum equation. In Eq. (3.41), κn is a constant with units
of [kg · m4 ·s−2 ] used to adjust the repulsion strength and is also simulation dependent
because it needs to counterbalance the particle acceleration, and prevent them from es-
caping the domain. This free parameter is not suited for complex pan-arctic simulations,
but is sufficient in our idealize experiment study. A free-slip boundary condition in all
simulations, i.e., no tangential friction force between boundary particle and ice particle is
applied.

Table 3.1: Physical parameters used across all simulations

Parameter Symbol Value Unit

Ice concentration parameter C 20 -
Ice compressive strength parameter P∗ 27.5 kN · m−2
Air density ρa 1.3 kg · m−3
Water density ρw 1026 kg · m−3
Ice density ρi 900 kg · m−3
Wind drag coefficient Ca 1.2 × 10−3 -
Water drag coefficient Cw 5.5 × 10−3 -
Minimal total deformation ∆min 2 × 10−9 s−1
1
Values of the parameter used for the simulations are the same as the one presented
in Williams et al. (2017) to facilitate comparison in the results section.

19
Chapter 4

Results and Discussion

4.1 Plastic wave propagation

We first compare the plastic wave speed for the VP dynamic equations with and without
the SPH approximations. To this end, we do a perturbation analysis for a one-dimensional
case with a fixed sea-ice concentration (A = 1). In this case, the 1D SPH sea-ice dynamic
equations (Eqs. 3.26 - 3.29) form a system of three equations and three unknowns (x, u
and h):

Dxp
= up (4.1)
Dt
N  
Dup X mq 1 1 xpq ∂W
=Γ + + τp , (4.2)
Dt q=1
ρ2i hq hp |xpq | ∂xpq
N
Dhp 1 X xpq ∂W
=− mq (uq − up ) , (4.3)
Dt ρi q=1 |xpq | ∂xpq

 
P∗
where xpq is a short form for xp − xq and Γ = 2
± (e−2 + 1)1/2 − 1 . In the above, we
made use of the following 1D normal stress for convergent plastic motion (see Gray, 1999;
Williams et al., 2017, for 1D normal stress derivation):

P∗
 
−2 1/2
σ = σxx = ± (e + 1) − 1 h = Γh. (4.4)
2

20
Linearizing around a mean state (ū = 0 and h̄ = h0 ), considering small perturbations (δx,
δu and δh) and ignoring 2nd order term, we obtain:

Dδxp
= δup (4.5)
Dt !
N
Dδup ΓX x̄pq −1 ∂W ∂ 2W
= ∆Aq (δhq + δhp ) + 2(δxp − δxq ) 2 , (4.6)
Dt ρi q=1 |x̄pq | h0 ∂ x̄pq ∂ x̄pq
N
Dδhp X x̄pq ∂W
= −h0 ∆Aq (δuq − δup ) , (4.7)
Dt q=1
|x̄pq | ∂ x̄pq

mq mq
where ∆Aq = ρi h0
= ρq
(Eq. 3.4) and where we have used the binomial expansion
1 1 1 δh
h
= h0 +δh
≈ h0
(1 − h0
). Assuming perturbations have a wavelike solution of the form
δf = fˆ exp(i(kx̄ − ωt)) — where i is the imaginary number, k is the wavenumber and
ω is the angular velocity — the set of equations in the reference frame following the ice
motion reduces to:

i
x̂ = û, (4.8)
ω !
N
∂ 2 Wpq
 
iΓ X x̄pq ĥ ∂W
û = Aq − (1 + exp(−ikx̄pq )) + 2x̂(1 − exp(−ikx̄pq )) ,
ωρi q=1 |x̄pq | h0 ∂ x̄pq ∂ x̄2pq

(4.9)
N
ih0 û X x̄pq
ĥ = − Aq ( exp(−ikx̄pq ) − 1). (4.10)
ω q=1 |x̄pq |

Note that since the ice is initially at rest, the Lagrangian and the Eulerian frameworks are
equivalent. For large enough wavelength (so that the perturbation can be resolved across
multiple particles with high accuracy i.e., λ ≥ lp and N → ∞), the summations can be
x̄pq R∞
written as integrals, i.e., N
P
q=1 Aq |x̄pq | becomes −∞ dx̄pq . Taking advantage of the kernel

21
properties — i.e., all moments higher than 0 vanish — we can write Eqs. 4.9 - 4.10 as:

−iΓ ∞ ĥ ∂W ∂ 2 Wpq
Z    
Γ ĥ 2
û = + 2x̂ exp(−ikx̄pq )dx̄pq = k + i2k x̂ W̃ , (4.11)
ωρi −∞ h0 ∂ x̄pq ∂ x̄2pq ωρi h0
ih0 û ∞
Z
∂W h0 ûk
ĥ = − exp(−ikx̄pq ) dx̄pq = W̃ , (4.12)
ω −∞ ∂ x̄pq ω

where the integrals have been converted to Fourier transform using F( ∂∂W
x̄pq
) = ikF(W ) =
ik W̃ . Finally, combining Eqs. (4.8 - 4.11 - 4.12), the phase speed for the plastic wave ( ωk )
can be written as: s  
ω Γ 2
cSPH = = ±W̃ − −1 . (4.13)
k ρi W̃
1
For wavelengths much larger than the smoothing length (λ ∝ k
≫ lp ), the Fourier trans-
form of the kernel tends to 1 (W̃ ≈ 1) and the SPH formulation reduces to the Viscous-
Plastic theory without SPH approximations (see for instance Williams et al., 2017), i.e.:

s
Γ
cVP = ± − , (4.14)
ρi

with a plastic wave propagation speed cVP ≈ 5.7 [m · s−1 ] for typical sea-ice parameters
(see Table 3.1). Consequently, a major difference of SPH with the FDM framework is
that the plastic wave speed is dispersive with a phase velocity cSPH that is dependent
on the wavelength and the smoothing length. In general, only the plastic waves with a
wavelength between approximately 1 and 11 times the smoothing length will have their
travelling speed modified by more than 1%. More specifically, in the limit where the
wavelength λ approaches the smoothing length lp , the plastic wave speed increases in the
SPH framework for a maximum value of ≈ 6.7 [m · s−1 ] (see Fig. 4.1 panel a). Note that
for wavelength smaller than the smoothing length, the summations in Eqs. (4.11- 4.12)
cannot be written as integrals but the particles still respond partially to the perturbations.
This sometimes leads to the tensile and the zero-energy modes instabilities (Swegle et al.,
1995). As mentioned above, Dehnen, Aly (2012) showed that Wendland kernels, can di-
minish the tensile instability and the pairing of particles. A deeper analysis of unresolved

22
waves (λ < lp ) in the context of sea-ice SPH dynamic equations is beyond the scope of the
current study.

For the more general case when the base state allows for a variable concentration (lin-
earized around a mean state Ā = A0 ) and considering the classical — denoted by a super-
script C — particle density definition (ρCp = ρi hp Ap ) used by Wang (2000); Ji et al. (2005);
Staroszczyk (2017), the plastic wave speed becomes:

s
Γ∗
 
2
cCA,SPH = ±W̃ − CA0 − 3 + , (4.15)
ρi W̃

where Γ∗ = Γ exp(−C(1−A0 )). We argue that the plastic wave speed cCA,SPH obtained with
the classical density definition does not converge (see Fig. 4.1 panel b) to the Viscous-
Plastic theory, cA,VP , derived from FDM (see Williams et al., 2017, for derivation):

s
Γ∗
 
cA,VP =± − CA0 + 1 , (4.16)
ρi

because the ice concentration is taken into account in both the definition of ρCp and implic-
itly in the definition of the average thickness hp . When we consider the new formulation
of particle density independent of concentration as proposed above (Eq. 3.31) the wave
speed equation becomes:

s
Γ∗
 
2
cA,SPH = ±W̃ − CA0 − 1 + , (4.17)
ρi W̃

which reduces to the FDM VP theory (Eq. 4.16) when the wavelength is large compared
to the smoothing length (see Fig. 4.1 panel c). However, the classical density definition
is not wrong, Wang (2000); Ji et al. (2005); Staroszczyk (2017) used different formulation
of the continuity equation in their model which makes our perturbation analysis only
valid in the current study. In a similar manner as for the plastic wave speed with a fixed

23
concentration (Eq. 4.13), the wave speed cA,SPH (Eq. 4.17) is dispersive and the wavelength
between 1 and 11 times the smoothing length are those that are mostly affected (more than
1%). However, in this case, the plastic wave speed is damped for wavelengths close to
the smoothing length for mean concentration state higher than 0.1. Note that while the
plastic wave speed is defined for all A, it does not have a physical meaning for A < 0.85
since there are negligible ice-ice interactions.

24
Figure 4.1: SPH plastic wave speed as a function of the normalized wavelength (λ/lp ) for
the Wendland C 6 kernel. Panel a) show the classical VP rheology with fixed concentra-
tion (Eq. 4.13) normalized by the FDM plastic wave speed with fixed concentration (Eq.
4.14), panel b) show the classical VP rheology with a variable concentration and the den-
sity definition ρCp = ρi hp Ap (Eq. 4.15) normalized by the FDM plastic wave speed with
a variable concentration (Eq. 4.16) and panel c) show the classical VP rheology with a
variable concentration and the density definition ρp = ρi hp (Eq. 4.17) normalized by the
FDM plastic wave speed with a variable concentration (Eq. 4.16). Different homogeneous
base state of concentration A0 are shown varying from 0 to 1.

25
4.2 Ridging experiments

We validate our implementation of the SPH model (with the new definition of particle
density ρp ) in a 1D ridging experiment for which we have an analytical solution (see
Williams, Tremblay, 2018, for derivation):

dσ dh 2ρa Ca |ua |ua

− = ρa Ca |ua |ua =⇒ = √ , (4.18)
dx dx P ( e−2 + 1 + 1)
∗

i.e., a linear profile in thickness with a slope proportional to the square of the wind ve-
locity and inversely proportional to the ice strength. We consider a rectangular domain
of 1000 by 2000 [km] including the boundary (the ice field is 1900 [km] to ensure that no
particle escape on the open side) with 37240 particles, an initial homogeneous smooth-
ing length lp of 21.429 [km] (spacing lp /α = 7.14 [km]) and a smaller — to limit boundary
effect — boundary particle smoothing length lb of 4 [km] (spacing lb /4 = 1.0 [km]) to rep-
resent the wall (see Fig. 4.2). Particles are initialized with an average thickness h = 1
[m] and a concentration A = 1. They are forced against the wall by a constant unidirec-
tional wind of 5 [m · s−1 ]. Note that the water drag force is removed in the simulation
for a faster convergence to the steady state which enables higher resolution — a water
current of 0 [m · s−1 ] would slow down the ice and the ridge formation since it is driven
by the advection speed. The Coriolis force should normally also have to be considered
with this domain size and classical polar latitude — the Rossby number is O(10−2 ) —, but
is neglected in this idealized experiment to conserve the symmetry of the solution and
compare it to the theoretical 1D equation ( Eq. 4.18). In results presented below (Fig. 4.3 -
4.4), the particles properties are averaged over a grid of approximately 10 by 5 [km] cells
for plotting purposes. Results show that the simulated thickness field converges to the
analytical solution (within an error of ≈ 1%) after ≈ 5 days with a slope of 1.33 × 10−3
[m · km−1 ], compared with 1.34 × 10−3 [m · km−1 ] for the theory. Artifacts are observed
close to the boundary where the repulsive force prevent the particle from reaching the
”wall”. Additionally, when a particle comes into contact with the boundary with a cer-

26
Figure 4.2: Idealized domain of the ridging experiment. The blue circles represent the ice
particles and the black ones are the boundary particles. The grey arrow shows the wind
forcing.

tain inertia (due to the 1/r dependence of the boundary force), we observe oscillations in
the motion of particles which can propagate far in the domain ( e.g., Fig. 4.3 panel a, at
x ≈ [50, 300] [km] and t = [30, 45] [h]). The oscillations are damped and the energy is dis-
sipated by the rheology term with time until an equilibrium is reached. A more physical
boundary treatment is beyond the scope of this study.

We also tested the ridge formation in a concentration regime that transitions from 0.5
to 1 with the same forcing conditions and total sea-ice volume as before, i.e., an initial
average thickness h = 0.5 [m] and an initial concentration A = 0.5 (see Fig. 4.3 panel b).
To this end, the domain including the boundaries was extended in the x̂ direction to 4000
[km] (for a total ice field extent of 3800 [km]) and the initial particles spacing is changed
from 7.14 [km] to 10.0 [km] with a corresponding initial smoothing length lp of 30.0 [km]
and a total number of particles of 38000. First, results show that the model converges in
≈ 10 days to a slope 1.36 × 10−3 [m · km−1 ], which is also in agreement with theory within
an error of ≈ 1% (see Fig. 4.3 panel b). A noticeable difference with the simulation initial-
ized at A = 1 is an average thickness greater than 1 building up outside of the ridge in
the marginal ice zone (MIZ). This is because when a particle reaches a maximum concen-

27
tration before its neighbour, it will start ridging, but will not be able to move far from its
current location because of the pressure applied by the surrounding particles resulting in
a local increase of thickness. This local ridging stabilize at a thickness of approximately
1.1 [m] and is akin to the wave radiation drag in the MIZ (Sutherland, Dumont, 2018),
even though no wave parametrization is implemented in the model. Further experimen-
tation of convergent ice flow in the MIZ with concentration close to 1 will be considered
in future work to explore this phenomenon. In order to test the behaviour of the continu-

Figure 4.3: Temporal evolution of simulated sea-ice thickness along the central horizontal
line of the domain for a) the ridge experiment initialized with a concentration A = 1 and
average thickness h = 1 and b) the ridge experiment initialized with a concentration
A = 0.5 and average thickness h = 0.5. The wall is located at x = 0 and the wind speed is
−5x̂ [m · s−1 ]. The theory follows Eq. (4.18).

ity equations in the context of SPH (Eqs. 3.28 - 3.29), we ensure that both h and A covary
h
in time in such a way that A
remains constant in the MIZ until significant ice interactions
take place. Results at x = 300 [km], away from the boundary effects, show that, as ex-

28
 d(h/A)
pected, thickness and concentration evolve coherently — h/A is constant in time or dt

is zero — before ice concentration reaches ≈ 85% (see Fig. 4.4 panel a). At that point
(t ≈ 22 [h]), the ice pack has sufficiently compacted to have an ice strength able to re-
pulse the incoming particle and the ridging process starts ( d(h/A)
dt
> 0). The ridge build-up
speed increase until a maximum concentration is reached after ≈ 70 [h] (see Fig. 4.4 panel
c). Subsequently, the rate of advance of the ridge slows down with time as it takes more
ice to be advected to build it up. When the ice thickness gradient is in balance with the
d(h/A)
surface wind stress (after ≈ 200 [h]), dt
reaches steady state. Overall, we can observe
3 stages in the ridge formation (see Fig. 4.4). First, a rapid compaction stage, when ice
particles are drifting close to their free drift speed since the ice strength is weak. Second,
a transition stage between A ≈ 0.85 − 1 when ridging occurs in the MIZ analogous to the
wave radiation drag mentioned above. Third, a ridging stage with changes in ice thick-
ness that are about one order of magnitude higher than the transition stage.

Note that oscillations (i.e., the ones between particles or coming from the boundaries)
in both ridging experiments, diminish when incorporating the water drag. The force re-
duce efficiently the advection speed of the ice which results in less kinetic energy carried
by the particles and fewer wobbles.

29
Figure 4.4: Evolution in time of a) the thickness normalized by concentration rate of
d(h/A)
change in time dt
, b) the average thickness h and c) the concentration A at x = 300
df f (x,t+∆t)−f (x,t−∆t)
[km]. The rate of change in time is computed from dt
(x, t) = 2∆t
.

30
4.3 Arch experiments

We next compare the SPH approach with the FDM and DEM sea-ice model in a second
well-studied idealized experiment: the ice arches formation. To this end, we run the SPH
model in an idealized domain representing the Nares Strait (see Fig. 4.5) with an up-
stream reservoir 5 times the length of the channel (L) to minimize boundary confinement
effect without sacrificing the spatial resolution.

The set of simulations uses a domain with L = 60 [km]. The initial condition for ice
thickness, concentration and velocity are h = 1 [m], A = 1 and u = 0 [m · s−1 ]. The
ice is forced with a constant unidirectional wind of −7.5 [m · s−1 ] in the ŷ-direction and
ocean current is fixed to uw = 0 [m · s−1 ]. The corresponding surface stress is ≈ 0.04
[kN · m−2 ] and the total integrated stress at the entry of the channel is slightly smaller
R 5L
than P ∗ ( 0 τa dx =26.325 [kN · m−1 ]). We use a weaker wind than what is common in
Nares Strait ice arches simulations (≈ 10 [m · s−1 ]) to limit the ridging phase prior to the
formation of the ice arch. In this experiment, we limit ourselves to ice with no tensile
strength (kt = 0) and a shear strength of 6.875 [kN · m−2 ], i.e., an ellipse aspect ratio of
2. We first test whether the SPH approach has the same sensitivity to the relative size of
particle with respect to the channel width as in DEM (Damsgaard et al., 2018). Results
showed that no stable arch can be formed with the specified forcing for all particle diam-
eter size tested (7.5, 5, 3.75 [km]) (see ice velocity field Fig.4.7). Instead, a ”continuous”
slow flow of ice is present in the channel. The discontinuity at the entry of the channel
visible in the concentration, thickness and velocity fields (Fig. 4.7) can be interpreted as
intermittent (unstable) ice arch formation. Also, we noted that larger particles are not
more prone to ice jam than smaller ones. This is contrary to what is know from granular
material theory and to results from Damsgaard et al. (2018) that show a transition from
stable to no ice arch formation for floe sizes ranging from approximately one quarter to
one sixteenth of the strait width. We explain this difference between SPH and DEM from

31
Figure 4.5: Idealized domain of the ice arch experiments. The blue circles represent the
ice particles and the black ones are the boundary particles. The grey arrow shows the
wind forcing.

the continuum description of the ice dynamics equation which describes the ice strength
as a function of ice concentration and mean thickness, not on the particle size. Even
though the increase in resolution — or particle size — has no effect on the arch stability,
it enables smaller fractures resolution that are visible at the entrance of the channel (see
ϵI and ϵII Fig. 4.6). In our SPH model, the stress invariants σI and σII shows oscillation
patterns in regions where ice is in the viscous regime (see the tree-like structure in the
normal and shear stress fields in Fig. 4.6). We hypothesized that those are associated with
over-damped viscous waves occurring with small movement of the particle undergoing
viscous deformation. Those structures are not symmetric, despite symmetrical initial con-
ditions, because of the domino effect between interacting viscous waves. Note that they
are absent from the strain-rate fields since viscous deformation are extremely small. They
are also absent in sea-ice model based on a continuum approach (Dumont et al., 2009;
Dansereau et al., 2017; Plante et al., 2020), but these tree-like structures are qualitatively

32
similar to the stress structure between floes observed in DEM (e.g., Damsgaard et al., 2018,
Fig. 5c).

Figure 4.6: Strain rate and stress invariants ( ϵ̇I , ϵ̇II , σI , σII ) at time t = 24 [h] for an initial
particle spacing of a) 7.5, b) 5 and c) 3.75 [km] (8, 12 and 16 particles can fit in the strait
respectively) and the initial total integrated surface stress at the entry of the channel is
26.325 [kN · m−1 ].

33
Figure 4.7: Ice concentration, thickness and total velocity (h, A, |ui |) at time t = 24 [h] for
an initial particle spacing of a) 7.5, b) 5 and c) 3.75 [km] (8, 12 and 16 particles can fit in the
strait respectively) and the initial total integrated surface stress at the entry of the channel
is 26.325 [kN · m−1 ].

Second, we explored the ability of the model to produce stable ice arches. To do so, we
reduce the total integrated surface stress at the entry of the channel to 13.146 [kN · m−1 ]
(or 5.3 [m · s−1 ]), which has the same effect as increasing the ice thickness. In this case,
results show a clear stable arch (see Fig. 4.8) with a shape that is qualitatively similar
to the one presented by Dansereau et al. (2017); Plante et al. (2020); West et al. (2022). It
has been shown by Dumont et al. (2009) that the elliptical EVP rheology can only form
stable arch for ellipse aspect ratio equal or lower than 1.8 (e ≤ 1.8). Yet, in an SPH model,
we are able to form one with e = 2. This suggests that SPH has a different sensitivity of
ice arching to the ellipse aspect ratio e and ice thickness h. This behaviour may change

34
when using a no-slip boundary condition instead of free slip. Further investigation is left
for future work. Nonetheless, this shows that SPH is able to capture large-scale features
coming from small-scale interactions. The stable ice arch (Fig. 4.8) also shows how SPH
can easily fracture and create discontinuity as seen in DEM models. Note that, in the SPH
framework, a lead can be defined by particles far enough apart to not be in the support
domain of each other — akin to DEM — or it can be defined by a set of particles with
neighbours, but with a concentration and an average thickness of 0 — akin to FDM. To
make an analogy of the lead formation in SPH with the continuum framework, it is as
if a grid splits up at the location of the fracture creating new edges in the domain. This
is a similar behaviour as the elastic-decohesive sea-ice constitutive model explored by
Schreyer et al. (2006) or the FEM model of Rampal et al. (2016).

Figure 4.8: Thickness field at time t = 0, 48, 168 [h] for an initial particle spacing of 7.5
[km] and a total integrated stress at the entry of the channel of 13.146 [kN · m−1 ].

35
Chapter 5

Conclusions and Summary

In this paper, we have presented a first implementation of the Viscous-Plastic rheology

with an elliptical yield curve and normal flow rule in the framework of SPH with the
long-term goal of simulating synoptic scale sea-ice dynamics. We have described the ba-
sics of the SPH approach and how the sea-ice dynamic equations can be formulated in this
framework along with the implementation of key components of the numerical method
such as the smoothing length, the kernel, the boundaries and the time integration tech-
nique. We proposed a different definition of the particle density and showed that the
more commonly used density definition involving the ice concentration when used to-
gether with the average ice thickness leads to erroneous plastic wave speed propagation.
A particle density definition independent of the ice concentration corrects this and leads
to results that are inline with the VP theory. The SPH model thus developed is in excellent
agreement (error of ≈ 1%) with an analytical solution of the VP ice dynamic for a simple
1D ridging experiment. The approximations used at the core of the SPH framework, re-
sult in a dispersive plastic wave speed in the medium — contrary to its FDM counterpart
— which is dependent on the smoothing length (or resolution) and the choice of the ker-
nel. The plastic wave speed is mostly affected for wavelengths 11 times the smoothing
length and lower.

36
From the simple ridging experiment with fixed sea-ice concentration (A = 1), we ob-
serve nonphysical damped oscillations that propagate in the domain associated with our
choice of boundary conditions. The conclusions drawn from our simulations are robust to
the choice of boundary conditions. Nevertheless, this behaviour needs to be removed for
a proper simulation of sea-ice near coastlines. The ridging experiment with an initial ice
concentration below 100% showed that continuity equations for concentration and thick-
ness evolve coherently until a concentration of 85%. At that point, SPH particles start
to ridge locally in the MIZ in addition to the wall where the maximum stress is located.
This effect is not observed in continuum approach and is presumably related to particle
collisions in converging motion.

When compared to other numerical framework, the SPH model is able to reproduce sta-
ble ice arches in an idealized domain of a strait with an ellipse aspect ratio of 2 and a wind
forcing of 5.3 [m · s−1 ], contrary to other continuum approaches that require higher mate-
rial shear strength. However, when using a stronger wind field of 7.5 [m · s−1 ], no stable
arches are formed when increasing the particles size in the strait (stable arches are only
achieved when increasing particle average thickness). We concluded that the number of
particles in the strait does not influence the formation of ice arches contrary to DEM and
is analogous to an increase in resolution in a continuum framework : a larger number of
particles influence the number of fractures that can form and the resolution of fine-scale
structures. The stress fields produced by the SPH model in the channel experiment show
tree-like pattern upstream of the channel where there are low total deformations. This is
not observed in FDM experiment but is qualitatively similar to tensile stress network ex-
hibited in DEM (Damsgaard et al., 2018) that comes from individual contact force between
the ice floes and is hypothesized to be associated with damped viscous sound waves.

Even though we successfully implemented the standard sea-ice Viscous-Plastic rheology

with an elliptical yield curve and a normal flow rule in an SPH framework, the current

37
model does not outperform classical FDM model. In fact, there are inherent difficulties
and instabilities in SPH that do not exist in FDM. It is known that the SPH framework
trade consistency — i.e., the ability to correctly represent a differential equation in the
limit of an infinite number of points with a null spacing between them — for stability ,
which gives the SPH a distinct feature of working well for many complicated problems
with good efficiency, but less accuracy. However, the classical formulation of SPH used
and described in the present work does not usually respect zeroth-order consistency be-
cause of the unstructured particle position in space(see Belytschko et al., 1998, section 3
for derivation). Nevertheless, consistency can be improved at the expense of computa-
tional cost (Chen, Beraun, 2000; Liu et al., 2003) by reformulating the SPH core approxi-
mation (Eq. 3.1). Also, boundary description has been identified as a weak point of the
SPH framework since prescribing a Dirichlet or Neumann boundary condition is not as
straightforward as in continuum approaches and preventing particle penetration through
a boundary is still a challenging task (Liu, Liu, 2010) and the SPH consistency is usually
at its worst at the boundary because the support domain is truncated. In the present
study, a proper physical representation of the boundary was not adopted and the bound-
ary treatment was chosen for its numerical simplicity and should be modified in future
work. Other major issues with SPH are the zero-energy modes and the tensile instability
previously mentioned. The zero-energy modes can be found in FDM and FEM and they
correspond to modes at which the strain energy calculated is erroneously zero (Swegle
et al., 1995). The tensile instability results in particle clumping or nonphysical fractures in
the material. In the present work, we adopted a different kernel from the usual Gaussian
spline to avoid those instabilities, but other methods such as the independent stress point
(Dyka, Ingel, 1995; Chalk et al., 2020), artificial short length repulsive force (Monaghan,
2000), particle repositioning (Sun et al., 2018), adaptive kernel (Lahiri et al., 2020), etc. can
be used if more stabilization is needed. For example, at smaller scale, SPH simulation
of ice in uniaxial compression was improved by a simplified finite difference interpo-
lation scheme (Zhang et al., 2017). More specifically for sea-ice model, Kreyscher et al.

38
(2000) pressure closure is not well suited for long simulation. Indeed, particle can still
move when they are in the viscous state but, have low internal ice pressure because of
the replacement pressure scheme. Consequently, particles could pass through each other
resulting in erroneous location of the parameters carried. Finally, using SPH for sea-ice
modules in grid-based continuum global climate model (GCM) complicates the coupling
with ocean and atmosphere components since particle quantities need to be converted on
a grid and vice versa.

Nevertheless, SPH also has interesting properties that could be exploited. For example,
SPH can be used with little change for problems involving several fluids whether liquid,
gas, or dust fluids (Monaghan, 2012). This feature could be exploited in the creation of
a general approach for all components of a GCM (atmosphere, ocean and sea-ice). The
method developed is also a proper option for nowcasting sea-ice prediction because only
the ice dynamics need to be considered in nowcasting applications and the model has a
good ability to carry the ice property in space. SPH can fracture and transitions from the
continuum to fragments seamlessly, which is the main reason for our investigation of the
method for sea-ice dynamics. The elastic behaviour assumed for sea-ice in certain rhe-
ology can be associated to the weak compressibility inherent in the classical formulation
of SPH. Finally, the SPH discretization of the continuum into particles enables the imple-
mentation of several new features. For example, angular momentum to individual floes
(or pack of floes) can be added to take into account rotation along LKFs. A direct measure
of the concentration from the number of particles within a support domain (this takes
advantages of already computed number of neighbours and help ensuring the desired
number of neighbours in converging flow) can be computed. A subscale parametrization
of floe-floe contact force (this short length repulsive force could also help for the tensile
instability) can be implemented. A varying floe size distribution can be incorporated by
varying the mass carried by a particle for a given particle density.

39
For future work, a more physical boundary treatment should be investigated — e.g., us-
ing the immerse boundary method (Tu et al., 2018) with a fixed grid for the boundary and
an interpolation scheme to apply force on the particle to simulate the grounding of sea-ice
near the coast. Additionally, to generalize the model for pan-arctic simulation, the Corio-
lis and sea surface tilt force along with the treatment of the thermodynamics sources and
sinks needs to be implemented in the framework (work already started by Staroszczyk,
2018). The computation speed should be made comparable to the FDM models and, to do
so, improving the algorithm parallelization is required. The current work began the study
of perturbations coming from the SPH framework in the context of sea-ice, nonetheless
a deeper inspection of the stability and accuracy of the method should be done. All in
all, there is a lot of work remaining for the development of an SPH sea-ice model able
to compete with current state-of-the-art continuum sea-ice models, but the method has
shown to be a promising tool to investigate sea-ice material properties.

40
Chapter 6

Appendix

6.1 Derivations of vector operator in SPH

Vector operator takes different forms in the SPH framework because they only operate
on the smoothing kernel W . Also, to make sure the interaction between particles is sym-
metric, the operator form is changed so both particles (the current one and its interacting
neighbours) interact in the same way. The following subsection shows the demonstra-
tions.

6.1.1 Divergence of a vector

First to symmetries this operator we rewrite the divergence of vector as follows using the
identity of the divergence of a scalar function times a vector and chose the scalar function
to be density:

 
1
∇ · V⃗ = ∇ · (ρV⃗ ) − V⃗ · ∇ρ . (6.1)
ρ

Now forgetting about equation (6.1) and applying the integral interpolant approxima-
tion (3.1) to the divergence of density times a vector and to the density gives us the two

41
following equation:

Z Z Z
⃗
∇ · (ρV ) = ′ ′ ⃗′ ⃗′
∇ · (ρ V )W dr = ∇ · (ρ V W )dr − ρ′ V⃗ ′ · ∇′ W dr⃗′ ,
′ ′ ⃗′ ⃗′ (6.2)
Z V V V

ρ= ρ′ W dr⃗′ . (6.3)
V

In the above equation, the prime quantities are used to show that they are functions of the
whole space, in contrast to the quantities with no primes which refer to a specific location
in space. The kernel is the only function that depends on both primed and non-primed
position as defined at (3.1). Using divergence theorem it is possible to cancel the first term
in (6.2) by the following :

Z Z
∇ · (ρ V⃗ ′ W )dr⃗′ =
′ ′
(ρ′ V⃗ ′ W ) · ds⃗′ = 0. (6.4)
V S

In the equation (6.4), s⃗′ denotes an infinitesimal area element and S is the surface encom-
passing the volume V. The term is null because of the compact support of the kernel
(W = 0 at the edge of the domain). Now using the particle approximation (3.4) the equa-
tions (6.2) and (6.3) are written as:

X X
(∇ · (ρV⃗ ))p = − mq V⃗q · ∇q Wpq = mq V⃗q · ∇p Wpq , (6.5)
q q
X
ρp = mq Wpq , (6.6)
q

where we used the identity ∇p = −∇q and p and q represent the current particle referential
and the neighbour one respectively. Finally, substituting the last two equations (6.5, 6.6)
in the first one that we let behind (6.1) gives the desired form :

X 
1 X
(∇ · V⃗ )p = mq V⃗q · ∇p Wpq − V⃗p · ∇p mq Wpq (6.7)
ρp q q
X 
1
= mq (V⃗q − V⃗p ) · ∇Wpq . (6.8)
ρp q

42
6.1.2 Divergence of a 2D tensor field

In the following demonstration, we used the Einstein summation convention to simplify

the calculation and the tensor representation. So, let’s start by writing the divergence of a
2D tensor divided by the density to have the following equation:

 
∂ Tij 1 ∂Tij Tij ∂ρ
= − 2 . (6.9)
∂xi ρ ρ ∂xi ρ ∂xi

Reorganizing the terms gives :

   
∂Tij ∂ Tij Tij ∂ρ
=ρ + 2 . (6.10)
∂xi ∂xi ρ ρ ∂xi

Now applying the interpolant approximation (3.1) to the first term in the bracket leads to
:

∂ Tij′
  Z  
∂ Tij
= ′
W dr⃗′ (6.11)
∂xi ρ V ∂xi ρ′
∂ Tij′
Z  ′
Tij ∂W ′
Z  
= ⃗
W dr −′ dr⃗ . (6.12)
′ ′ ′ ′
V ∂x i ρ V ρ ∂x i

As for the divergence of a vector demonstration (section 6.1.1), the first integral above
vanish by using the divergence theorem since the kernel has a compact support domain.
Therefore, applying the particle approximation to the above equation leads to :

   X  (Tij )q  ∂Wpq X  (Tij )q  ∂Wpq

∂ Tij
=− mq 2 = mq 2 . (6.13)
∂xi ρ p q
ρq ∂(xi )q q
ρq ∂(xi )p

43
Substituting this in the equation 6.9 and using the equality 6.6 for the density in the gra-
dient operator, in the last term, we get the following expression :

  X  X 
∂Tij (Tij )q ∂Wpq (Tij )p ∂
= ρp mq 2 + mq Wpq (6.14)
∂xi p q
ρq ∂(xi )p ρ2p ∂(xi )p q
X   
(Tij )q (Tij )p ∂Wpq
= ρp mq 2
+ (6.15)
q
ρ q ρ2p ∂(xi )p
X  T⃗q T⃗p 
= ρp mq 2 + 2 · ∇p Wpq , (6.16)
q
ρq ρp

which is the form presented at equation (3.6).

6.1.3 Gradient of a vector field

To demonstrate the equation (3.7) we first have to recall that the zeroth-order moment of
the kernel equals 1 and it can be written as :

Z
M0 = W (⃗r − r⃗′ , lp )dr⃗′ = 1 (6.17)
V
X mq
= Wpq , (6.18)
q
ρq

where the particle approximation (3.4) was used to rewrite the integral. Using this iden-
tity it is possible to write:

∇(V⃗ 1) = 1∇V⃗ + V⃗ · ∇1 (6.19)

= ∇V⃗ − V⃗ · ∇1. (6.20)

44
Now, using the particle approximation (3.4) on V⃗ and substituting the 1 with the zeroth
moment identity in the expression 6.20 we get:

∂ X mq ∂ X mq
(∇V⃗ )p = (Vj )q Wpq − (Vj )p Wpq (6.21)
∂(xi )p q ρq ∂(xi )p q ρq
X mq ∂
= ((Vj )q − (Vj )p ) Wpq (6.22)
q
ρq ∂(xi )p
X mq
= (V⃗q − V⃗p ) ⊗ ∇p Wpq , (6.23)
q
ρ q

where Einstein summation convention was once again used to simplify the derivation.

45
Bibliography

Aagaard Knut, Coachman L.K., Carmack Eddy. On the halocline of the Arctic Ocean // Deep
Sea Research Part A. Oceanographic Research Papers. VI 1981. 28, 6. 529–545.

Adcroft Alistair, Anderson Whit, Balaji V., Blanton Chris, Bushuk Mitchell, Dufour Carolina O.,
Dunne John P., Griffies Stephen M., Hallberg Robert, Harrison Matthew J., Held Isaac M.,
Jansen Malte F., John Jasmin G., Krasting John P., Langenhorst Amy R., Legg Sonya, Liang Zhi,
McHugh Colleen, Radhakrishnan Aparna, Reichl Brandon G., Rosati Tony, Samuels Bonita L.,
Shao Andrew, Stouffer Ronald, Winton Michael, Wittenberg Andrew T., Xiang Baoqiang,
Zadeh Niki, Zhang Rong. The GFDL Global Ocean and Sea Ice Model OM4.0: Model
Description and Simulation Features // Journal of Advances in Modeling Earth Sys-
tems. X 2019. 11, 10. 3167–3211.

Arrigo Kevin R. Annual cycles of sea ice and phytoplankton in Cape Bathurst polynya,
southeastern Beaufort Sea, Canadian Arctic // Geophysical Research Letters. 2004. 31,
8.

Beatty Christof König, Holland David M. Modeling landfast sea ice by adding tensile
strength // Journal of Physical Oceanography. I 2010. 40, 1. 185–198. Copyright: Copy-
right 2010 Elsevier B.V., All rights reserved.

Belytschko T., Krongauz Y., Dolbow J., Gerlach C. On the completeness of meshfree particle
methods // International Journal for Numerical Methods in Engineering. XI 1998. 43,
5. 785–819.

46
Bouchat Amélie, Hutter Nils, Chanut Jérôme, Dupont Frédéric, Dukhovskoy Dmitry, Garric
Gilles, Lee Younjoo J., Lemieux Jean-François, Lique Camille, Losch Martin, Maslowski Wies-
law, Myers Paul G., Ólason Einar, Rampal Pierre, Rasmussen Till, Talandier Claude, Tremblay
Bruno, Wang Qiang. Sea Ice Rheology Experiment (SIREx): 1. Scaling and Statistical
Properties of Sea-Ice Deformation Fields // Journal of Geophysical Research: Oceans.
IV 2022. 127, 4.

Bouchat Amélie, Tremblay Bruno. Using sea-ice deformation fields to constrain the mechan-
ical strength parameters of geophysical sea ice // Journal of Geophysical Research:
Oceans. VII 2017. 122, 7. 5802–5825.

Budikova Dagmar. Role of Arctic sea ice in global atmospheric circulation: A review //
Global and Planetary Change. VIII 2009. 68, 3. 149–163.

Cavelan Aurélien, Cabezón Rubén M., Korndorfer Jonas H. M., Ciorba Florina M. Finding
Neighbors in a Forest: A b-tree for Smoothed Particle Hydrodynamics Simulations.
2019.

Chalk C.M., Pastor M., Peakall J., Borman D.J., Sleigh P.A., Murphy W., Fuentes R. Stress-
Particle Smoothed Particle Hydrodynamics: An application to the failure and post-
failure behaviour of slopes // Computer Methods in Applied Mechanics and Engi-
neering. VII 2020. 366. 113034.

Chen J.K., Beraun J.E. A generalized smoothed particle hydrodynamics method for nonlin-
ear dynamic problems // Computer Methods in Applied Mechanics and Engineering.
X 2000. 190, 1-2. 225–239.

Coon Max, Kwok Ron, Levy Gad, Pruis Matthew, Schreyer Howard, Sulsky Deborah. Arctic Ice
Dynamics Joint Experiment (AIDJEX) assumptions revisited and found inadequate //
Journal of Geophysical Research. XI 2007. 112, C11.

47
Damsgaard A., Adcroft A., Sergienko O. Application of Discrete Element Methods to Ap-
proximate Sea Ice Dynamics // Journal of Advances in Modeling Earth Systems. IX
2018. 10, 9. 2228–2244.

Dansereau Véronique, Weiss Jérôme, Saramito Pierre, Lattes Philippe. A Maxwell elasto-brittle
rheology for sea ice modelling // The Cryosphere. VII 2016. 10, 3. 1339–1359.

Dansereau Véronique, Weiss Jérôme, Saramito Pierre, Lattes Philippe, Coche Edmond. Ice
bridges and ridges in the Maxwell-EB sea ice rheology // The Cryosphere. IX 2017.
11, 5. 2033–2058.

Dehnen Walter, Aly Hossam. Improving convergence in smoothed particle hydrodynamics

simulations without pairing instability // Monthly Notices of the Royal Astronomical
Society. 09 2012. 425, 2. 1068–1082.

Dethleff Dirk. Polynyas as a Possible Source for Enigmatic Bennett Island Atmospheric
Plumes // The Polar Oceans and Their Role in Shaping the Global Environment. III
2013. 475–483.

Dumont Dany, Gratton Yves, Arbetter Todd E. Modeling the Dynamics of the North Water
Polynya Ice Bridge // Journal of Physical Oceanography. VI 2009. 39, 6. 1448–1461.

Dyka C.T., Ingel R.P. An approach for tension instability in smoothed particle hydrody-
namics (SPH) // Computers &amp Structures. XI 1995. 57, 4. 573–580.

Fichefet T., Maqueda M. A. Morales. Sensitivity of a global sea ice model to the treatment
of ice thermodynamics and dynamics // Journal of Geophysical Research: Oceans. VI
1997. 102, C6. 12609–12646.

Fleissner Florian, Gaugele Timo, Eberhard Peter. Applications of the discrete element method
in mechanical engineering // Multibody System Dynamics. VI 2007. 18, 1.

Fraga Filho Carlos Alberto. Smoothed Particle Hydrodynamics: Fundamentals and Basic
Applications in Continuum Mechanics. 01 2019.

48
Gardner Alex S., Sharp Martin J. A review of snow and ice albedo and the development of
a new physically based broadband albedo parameterization // Journal of Geophysical
Research. III 2010. 115, F1.

Gingold R. A., Monaghan J. J. Smoothed particle hydrodynamics: theory and application

to non-spherical stars // Monthly Notices of the Royal Astronomical Society. XII 1977.
181, 3. 375–389.

Girard Lucas, Bouillon Sylvain, Weiss Jérôme, Amitrano David, Fichefet Thierry, Legat Vincent.
A new modeling framework for sea-ice mechanics based on elasto-brittle rheology //
Annals of Glaciology. 2011. 52, 57. 123–132.

Gray J., Morland L.W. A Two-Dimensional Model for the Dynamics of Sea Ice // Philo-
sophical Transactions of The Royal Society B: Biological Sciences. 04 1994. 347. 219–290.

Gray J. M. N. T. Loss of Hyperbolicity and Ill-posedness of the Viscous–Plastic Sea Ice

Rheology in Uniaxial Divergent Flow // Journal of Physical Oceanography. XI 1999.
29, 11. 2920–2929.

Gutfraind Ricardo, Savage Stuart B. Smoothed Particle Hydrodynamics for the Simulation
of Broken-Ice Fields: Mohr–Coulomb-Type Rheology and Frictional Boundary Condi-
tions // Journal of Computational Physics. VII 1997. 134, 2. 203–215.

Haid V., Timmermann R. Simulated heat flux and sea ice production at coastal polynyas
in the southwestern Weddell Sea // Journal of Geophysical Research: Oceans. V 2013.
118, 5. 2640–2652.

Herman Agnieszka. Discrete-Element bonded-particle Sea Ice model DESIgn, version 1.3a –
model description and implementation // Geoscientific Model Development. IV 2016.
9, 3. 1219–1241.

Hibler W. D. A Dynamic Thermodynamic Sea Ice Model // Journal of Physical Oceanog-

raphy. VII 1979. 9, 4. 815–846.

49
Hopkins Mark A., Thorndike Alan S. Floe formation in Arctic sea ice // Journal of Geo-
physical Research. 2006. 111, C11.

Hosseini Khosrow, Omidvar Pourya, Kheirkhahan Mehran, Farzin Saeed. Smoothed particle
hydrodynamics for the interaction of Newtonian and non-Newtonian fluids using the
(I) model // Powder Technology. VI 2019. 351. 325–337.

Hunke E. C., Dukowicz J. K. An Elastic–Viscous–Plastic Model for Sea Ice Dynamics //

Journal of Physical Oceanography. IX 1997. 27, 9. 1849–1867.

Hutter Nils, Bouchat Amélie, Dupont Frédéric, Dukhovskoy Dmitry, Koldunov Nikolay, Lee
Younjoo J., Lemieux Jean-François, Lique Camille, Losch Martin, Maslowski Wieslaw, My-
ers Paul G., Ólason Einar, Rampal Pierre, Rasmussen Till, Talandier Claude, Tremblay Bruno,
Wang Qiang. Sea Ice Rheology Experiment (SIREx): 2. Evaluating Linear Kinematic
Features in High-Resolution Sea Ice Simulations // Journal of Geophysical Research:
Oceans. IV 2022. 127, 4.

Ji Shunying, Shen Hung, Wang Z., Shen Hayley, Yue Q. A viscoelastic-plastic constitutive
model with Mohr-Coulomb yielding criterion for sea ice dynamics // Acta Oceanolog-
ica Sinica. 01 2005. 24. 54–65.

Johnson G. R., Beissel S. R. NORMALIZED SMOOTHING FUNCTIONS FOR SPH IM-

PACT COMPUTATIONS // International Journal for Numerical Methods in Engineer-
ing. 1996. 39, 16. 2725–2741.

Kalenitchenko Dimitri, Joli Nathalie, Potvin Marianne, Tremblay Jean-Éric, Lovejoy Connie. Bio-
diversity and Species Change in the Arctic Ocean: A View Through the Lens of Nares
Strait // Frontiers in Marine Science. VIII 2019. 6.

Kottmeier Ch., Engelbart D. Generation and atmospheric heat exchange of coastal polynyas
in the Weddell Sea // Boundary-Layer Meteorology. VIII 1992. 60, 3. 207–234.

50
Kreyscher Martin, Harder Markus, Lemke Peter, Flato Gregory M. Results of the Sea Ice Model
Intercomparison Project: Evaluation of sea ice rheology schemes for use in climate sim-
ulations // Journal of Geophysical Research: Oceans. V 2000. 105, C5. 11299–11320.

Lahiri Saptarshi Kumar, Bhattacharya Kanishka, Shaw Amit, Ramachandra L S. A stable SPH
with adaptive B-spline kernel // ArXiv. 2020. abs/2001.03416.

Lemieux Jean-François, Tremblay Bruno. Numerical convergence of viscous-plastic sea ice

models // Journal of Geophysical Research. V 2009. 114, C5.

Lilja Ville-Pekka, Polojärvi Arttu, Tuhkuri Jukka, Paavilainen Jani. Finite-discrete element
modelling of sea ice sheet fracture // International Journal of Solids and Structures. V
2021. 217-218. 228–258.

Lipscomb William H., Hunke Elizabeth C., Maslowski Wieslaw, Jakacki Jaromir. Ridging,
strength, and stability in high-resolution sea ice models // Journal of Geophysical
Research. III 2007. 112, C3.

Liu G.R., Liu M.B. Smoothed Particle Hydrodynamics: A Meshfree Particle Method. 2003.

Liu M. B., Liu G. R. Smoothed Particle Hydrodynamics (SPH): an Overview and Recent
Developments // Archives of Computational Methods in Engineering. II 2010. 17, 1.
25–76.

Liu M. B., Liu G. R., Lam K. Y. A one-dimensional meshfree particle formulation for simu-
lating shock waves // Shock Waves. XI 2003. 13, 3. 201–211.

Losch Martin, Menemenlis Dimitris, Campin Jean-Michel, Heimbach Patick, Hill Chris. On the
formulation of sea-ice models. Part 1: Effects of different solver implementations and
parameterizations // Ocean Modelling. I 2010. 33, 1-2. 129–144.

Lucy L. B. A numerical approach to the testing of the fission hypothesis // The Astro-
nomical Journal. XII 1977. 82. 1013.

51
Maqueda M. A. Morales. Polynya Dynamics: a Review of Observations and Modeling //
Reviews of Geophysics. 2004. 42, 1.

Maykut Gary A. Large-scale heat exchange and ice production in the central Arctic //
Journal of Geophysical Research. 1982. 87, C10. 7971.

McPhee Miles G. The Effect of the Oceanic Boundary Layer on the Mean Drift of Pack
Ice: Application of a Simple Model // Journal of Physical Oceanography. III 1979. 9, 2.
388–400.

Mehlmann C., Danilov S., Losch M., Lemieux J. F., Hutter N., Richter T., Blain P., Hunke E. C.,
Korn P. Simulating Linear Kinematic Features in Viscous-Plastic Sea Ice Models on
Quadrilateral and Triangular Grids With Different Variable Staggering // Journal of
Advances in Modeling Earth Systems. X 2021. 13, 11.

Monaghan J J. Smoothed particle hydrodynamics // Reports on Progress in Physics. VII

2005. 68, 8. 1703–1759.

Monaghan J.J. SPH without a Tensile Instability // Journal of Computational Physics. IV

2000. 159, 2. 290–311.

Monaghan J.J. Smoothed Particle Hydrodynamics and Its Diverse Applications // Annual
Review of Fluid Mechanics. 2012. 44, 1. 323–346.

Monaghan J.J., Kajtar J.B. SPH particle boundary forces for arbitrary boundaries // Com-
puter Physics Communications. 2009. 180, 10. 1811–1820.

Morland L. W., Staroszczyk R. A material coordinate treatment of the sea–ice dynamics

equations // Proceedings of the Royal Society of London. Series A: Mathematical,
Physical and Engineering Sciences. XI 1998. 454, 1979. 2819–2857.

Morris Joseph P., Fox Patrick J., Zhu Yi. Modeling Low Reynolds Number Incompressible
Flows Using SPH // Journal of Computational Physics. 1997. 136, 1. 214–226.

52
Natanson I. P. Theory of Functions of a Real Variable. 2. New York: Frederick Ungar
Publishing Co., 1961. 11 – 21.

Ohshima Kay I., Nihashi Sohey, Iwamoto Katsushi. Global view of sea-ice production in
polynyas and its linkage to dense/bottom water formation // Geoscience Letters. V
2016. 3, 1.

Overland James E., McNutt S. Lyn, Salo Sigrid, Groves Joanne, Li Shusun. Arctic sea ice as a
granular plastic // Journal of Geophysical Research: Oceans. IX 1998. 103, C10. 21845–
21867.

Peiró Joaquim, Sherwin Spencer. Finite difference, finite element and finite volume methods
for partial differential equations // Handbook of materials modeling. 2005. 2415–2446.

Plante Mathieu, Tremblay Bruno, Losch Martin, Lemieux Jean-François. Landfast sea ice ma-
terial properties derived from ice bridge simulations using the Maxwell elasto-brittle
rheology // The Cryosphere. VII 2020. 14, 6. 2137–2157.

Rabatel Matthias, Labbé Stéphane, Weiss Jérôme. Dynamics of an assembly of rigid ice floes
// Journal of Geophysical Research: Oceans. IX 2015. 120, 9. 5887–5909.

Rampal Pierre, Bouillon Sylvain, Ólason Einar, Morlighem Mathieu. neXtSIM: a new La-
grangian sea ice model // The Cryosphere. V 2016. 10, 3. 1055–1073.

Ranta Janne, Polojärvi Arttu, Tuhkuri Jukka. Limit mechanisms for ice loads on inclined
structures: Buckling // Cold Regions Science and Technology. III 2018. 147. 34–44.

Rhoades Clifford E. A fast algorithm for calculating particle interactions in smooth particle
hydrodynamic simulations // Computer Physics Communications. VII 1992. 70, 3.
478–482.

Ringeisen Damien, Losch Martin, Tremblay L. Bruno, Hutter Nils. Simulating intersection
angles between conjugate faults in sea ice with different viscous–plastic rheologies //
The Cryosphere. IV 2019. 13, 4. 1167–1186.

53
Salehizadeh A. M., Shafiei A. R. Modeling of granular column collapses with (I) rheology
using smoothed particle hydrodynamic method // Granular Matter. III 2019. 21, 2.

Schreyer H. L., Sulsky D. L., Munday L. B., Coon M. D., Kwok R. Elastic-decohesive consti-
tutive model for sea ice // Journal of Geophysical Research. X 2006. 111, C11.

Schulson Erland M. Compressive shear faults within arctic sea ice: Fracture on scales large
and small // Journal of Geophysical Research. 2004. 109, C7.

Sheikh Bahman, Qiu Tong, Ahmadipur Amir. Comparison of SPH boundary approaches
in simulating frictional soil–structure interaction // Acta Geotechnica. X 2020. 16, 8.
2389–2408.

Shen Hung Tao, Shen Hayley, Tsai Shi-Ming. Dynamic transport of river ice // Journal of
Hydraulic Research. XI 1990. 28, 6. 659–671.

Smith Stuart D., Muench Robin D., Pease Carol H. Polynyas and leads: An overview of
physical processes and environment // Journal of Geophysical Research. 1990. 95, C6.
9461.

Södergren A. Helena, McDonald Adrian J., Bodeker Gregory E. An energy balance model ex-
ploration of the impacts of interactions between surface albedo, cloud cover and water
vapor on polar amplification // Climate Dynamics. XI 2017. 51, 5-6. 1639–1658.

Staroszczyk Ryszard. SPH Modelling of Sea-ice Pack Dynamics // Archives of Hydro-

Engineering and Environmental Mechanics. XII 2017. 64, 2. 115–137.

Staroszczyk Ryszard. Simulation of Sea-ice Thermodynamics by a Smoothed Particle Hy-

drodynamics Method // Archives of Hydro-Engineering and Environmental Mechan-
ics. XII 2018. 65, 4. 277–299.

Stirling Ian. The Biological Importance of Polynyas in the Canadian Arctic // ARCTIC. I
1980. 33, 2.

54
Stirling Ian. The importance of polynyas, ice edges, and leads to marine mammals and
birds // Journal of Marine Systems. I 1997. 10, 1-4. 9–21.

Sulsky Deborah, Schreyer Howard, Peterson Kara, Kwok Ron, Coon Max. Using the material-
point method to model sea ice dynamics // Journal of Geophysical Research. II 2007.
112, C2.

Sun P.N., Colagrossi A., Marrone S., Antuono M., Zhang A.M. Multi-resolution Delta-plus-
SPH with tensile instability control: Towards high Reynolds number flows // Com-
puter Physics Communications. III 2018. 224. 63–80.

Sutherland Peter, Dumont Dany. Marginal Ice Zone Thickness and Extent due to Wave
Radiation Stress // Journal of Physical Oceanography. VIII 2018. 48, 8. 1885–1901.

Swegle J.W., Hicks D.L., Attaway S.W. Smoothed Particle Hydrodynamics Stability Analysis
// Journal of Computational Physics. I 1995. 116, 1. 123–134.

Tsamados M., Feltham D. L., Wilchinsky A. V. Impact of a new anisotropic rheology on

simulations of Arctic sea ice // Journal of Geophysical Research: Oceans. I 2013. 118,
1. 91–107.

Tu Jiyuan, Yeoh Guan-Heng, Liu Chaoqun. Chapter 9 - Some Advanced Topics in CFD //
Computational Fluid Dynamics (Third Edition). 2018. Third Edition. 369–417.

Walsh John E. The role of sea ice in climatic variability: Theories and evidencesup1/sup
// Atmosphere-Ocean. IX 1983. 21, 3. 229–242.

Wang Zhilian. A coastal sea ice model with discrete parcel method. 2000. 146. Copyright -
Database copyright ProQuest LLC; ProQuest does not claim copyright in the individual
underlying works; Last updated - 2022-02-08.

Weiss Jérôme, Schulson Erland M., Stern Harry L. Sea ice rheology from in-situ, satellite and
laboratory observations: Fracture and friction // Earth and Planetary Science Letters.
III 2007. 255, 1-2. 1–8.

55
Wendland Holger. Piecewise polynomial, positive definite and compactly supported radial
functions of minimal degree // Advances in Computational Mathematics. XII 1995. 4,
1. 389–396.

West Brendan, O’Connor Devin, Parno Matthew, Krackow Max, Polashenski Christopher.
Bonded Discrete Element Simulations of Sea Ice With Non-Local Failure: Applications
to Nares Strait // Journal of Advances in Modeling Earth Systems. V 2022. 14, 6.

Williams James, Tremblay L. Bruno. The dependence of energy dissipation on spatial reso-
lution in a viscous-plastic sea-ice model // Ocean Modelling. X 2018. 130. 40–47.

Williams James, Tremblay L. Bruno, Lemieux Jean-François. The effects of plastic waves on the
numerical convergence of the viscous–plastic and elastic–viscous–plastic sea-ice mod-
els // Journal of Computational Physics. VII 2017. 340. 519–533.

Yang Edward, Bui Ha H., Sterck Hans De, Nguyen Giang D., Bouazza Abdelmalek. A scalable
parallel computing SPH framework for predictions of geophysical granular flows //
Computers and Geotechnics. V 2020. 121. 103474.

Zhang Ningbo, Zheng Xing, Ma Qingwei. Updated Smoothed Particle Hydrodynamics for
Simulating Bending and Compression Failure Progress of Ice // Water. XI 2017. 9, 11.
882.

56