Changeset 11031 for NEMO/trunk/doc/latex/SI3/subfiles/chap_model_basics.tex
- Timestamp:
- 2019-05-21T21:41:02+02:00 (5 years ago)
- File:
-
- 1 edited
Legend:
- Unmodified
- Added
- Removed
-
NEMO/trunk/doc/latex/SI3/subfiles/chap_model_basics.tex
r11015 r11031 28 28 29 29 \subsection{Scales, thermodynamics and dynamics} 30 Because sea ice is much wider -- $\mathcal{O}$(100-1000 km) -- than thick -- $\mathcal{O}$(1 m) -- ice drift can be considered as purely horizontal: vertical motions around the hydrostatic equilibrium position are negligible. The same scaling argument justifies the assumption that heat exchanges are purely vertical\footnote{The latter assumption is probably less valid, because the horizontal scales of temperature variations are $\mathcal{O}$(10-100 m)}. It is on this basis that thermodynamics and dynamics are separated and rely upon different frameworks and sets of hypotheses: thermodynamics use the ice thickness distribution \citep{ Thorndikeetal75} and the mushy-layer \citep{Worster92} frameworks, whereas dynamics assume continuum mechanics \citep[e.g.,][]{Lepparanta05}. Thermodynamics and dynamics interact by two means: first, advection impacts state variables; second, the horizontal momentum equation depends, among other things, on the ice state.30 Because sea ice is much wider -- $\mathcal{O}$(100-1000 km) -- than thick -- $\mathcal{O}$(1 m) -- ice drift can be considered as purely horizontal: vertical motions around the hydrostatic equilibrium position are negligible. The same scaling argument justifies the assumption that heat exchanges are purely vertical\footnote{The latter assumption is probably less valid, because the horizontal scales of temperature variations are $\mathcal{O}$(10-100 m)}. It is on this basis that thermodynamics and dynamics are separated and rely upon different frameworks and sets of hypotheses: thermodynamics use the ice thickness distribution \citep{thorndike_1975} and the mushy-layer \citep{worster_1992} frameworks, whereas dynamics assume continuum mechanics \citep[e.g.,][]{lepp_ranta_2011}. Thermodynamics and dynamics interact by two means: first, advection impacts state variables; second, the horizontal momentum equation depends, among other things, on the ice state. 31 31 32 32 \subsection{Subgrid scale variations} 33 33 34 Sea ice properties -- in particular ice thickness -- feature important changes at horizontal scales $\mathcal{O}$(1m) \citep{ Thorndikeetal75}. An explicit representation of these variations is not and will not be -- at least in the next twenty years or so -- accessible to large-scale sea ice models. Yet important features, such as energy exchanges through the ice, quite non-linearly depend on ice thickness \citep{Maykut86}; whereas ice motion depends on the presence of open water, thin and thick ice at the very least, suggesting that subgrid-scale variations in ice properties must be accounted for, at least in a statistical fashion \citep{MaykutThorndike73}.34 Sea ice properties -- in particular ice thickness -- feature important changes at horizontal scales $\mathcal{O}$(1m) \citep{thorndike_1975}. An explicit representation of these variations is not and will not be -- at least in the next twenty years or so -- accessible to large-scale sea ice models. Yet important features, such as energy exchanges through the ice, quite non-linearly depend on ice thickness \citep{maykut_1986}; whereas ice motion depends on the presence of open water, thin and thick ice at the very least, suggesting that subgrid-scale variations in ice properties must be accounted for, at least in a statistical fashion \citep{maykut_1973}. 35 35 36 36 %-------------------------------------------------------------------------------------------------------------------- … … 49 49 %-------------------------------------------------------------------------------------------------------------------- 50 50 51 The \textit{multi-category} framework \citep{ MaykutThorndike73} addresses this issue by treating the ice thickness as an independent variable next to spatial coordinates and time, and introducing a thickness distribution\footnote{$g(h)$, termed the \textit{ice thickness distribution} is the density of probability of ice thickness \citep{Thorndikeetal75}.} $g(h)$ as the main prognostic model field. In the discrete world, the thickness distribution is converted into $L$ thickness categories. Ice thickness categories occupy a fraction of each grid cell, termed ice concentration ($a_l, l=1, 2, ..., L$), with specific thickness and properties.52 53 The \textit{single-category} framework \citep{ Hibler79} tackles the subgrid-scale issue by drastically simplifying the ice thickness distribution. The grid cell is divided into open water and sea ice characterized by a single ice concentration $A$ and mean thickness $H$. Single-category models (in particular LIM2) typically add parameterizations to represent the effects of unresolved ice thickness distribution on ice growth and melt \citep[see, e.g.][]{MellorKantha89,FichefetMaqueda97}.51 The \textit{multi-category} framework \citep{maykut_1973} addresses this issue by treating the ice thickness as an independent variable next to spatial coordinates and time, and introducing a thickness distribution\footnote{$g(h)$, termed the \textit{ice thickness distribution} is the density of probability of ice thickness \citep{thorndike_1975}.} $g(h)$ as the main prognostic model field. In the discrete world, the thickness distribution is converted into $L$ thickness categories. Ice thickness categories occupy a fraction of each grid cell, termed ice concentration ($a_l, l=1, 2, ..., L$), with specific thickness and properties. 52 53 The \textit{single-category} framework \citep{hibler_1979} tackles the subgrid-scale issue by drastically simplifying the ice thickness distribution. The grid cell is divided into open water and sea ice characterized by a single ice concentration $A$ and mean thickness $H$. Single-category models (in particular LIM2) typically add parameterizations to represent the effects of unresolved ice thickness distribution on ice growth and melt \citep[see, e.g.][]{mellor_1989,fichefet_1997}. 54 54 55 55 SI$^3$ provides the choice between either a multi- or a single-category framework. The default mode is multi-category. The single-category mode can be activated by setting the number of categories ($jpl=1$) and by activating the virtual thickness distribution parameterizations (nn\_monocat=1). … … 70 70 & Description & Value & Units & Ref \\ \hline 71 71 $c_i$ (cpic) & Pure ice specific heat & 2067 & J/kg/K & ? \\ 72 $c_w$ (rcp) & Seawater specific heat & 3991 & J/kg/K & \cite{TEOS -10} \\73 $L$ (lfus) & Latent heat of fusion (0$^\circ$C) & 334000 & J/kg/K & \cite{ BitzLipscomb99} \\74 $\rho_i$ (rhoic) & Sea ice density & 917 & kg/m$^3$ & \cite{ BitzLipscomb99} \\75 $\rho_s$ (rhosn) & Snow density & 330 & kg/m$^3$ & \cite{ MaykutUntersteiner71} \\76 $\mu$ (tmut) & Linear liquidus coefficient & 0.054 & $^\circ$C/(g/kg) & \cite{ Assur58} \\72 $c_w$ (rcp) & Seawater specific heat & 3991 & J/kg/K & \cite{TEOS_2010} \\ 73 $L$ (lfus) & Latent heat of fusion (0$^\circ$C) & 334000 & J/kg/K & \cite{bitz_1999} \\ 74 $\rho_i$ (rhoic) & Sea ice density & 917 & kg/m$^3$ & \cite{bitz_1999} \\ 75 $\rho_s$ (rhosn) & Snow density & 330 & kg/m$^3$ & \cite{maykut_1971} \\ 76 $\mu$ (tmut) & Linear liquidus coefficient & 0.054 & $^\circ$C/(g/kg) & \cite{assur_1958} \\ 77 77 \hline 78 78 \end{tabular} … … 85 85 % 86 86 87 Ice thermodynamics are formulated assuming that sea ice is covered by snow. Within each thickness category, both snow and sea ice are horizontally uniform, hence each thickness category has a specific ice thickness ($h_i^l$) and snow depth ($h_s^l$). Snow is assumed to be fresh, with constant density and thermal conductivity. Sea ice is assumed to be a \textit{mushy layer}\footnote{Mushy layers are two-phase, two-component porous media.} \citep{ Worster92} of constant density, made of pure ice and brine in thermal equilibrium, related by a linear liquidus relationship \citep{BitzLipscomb99}. A vertically-averaged bulk salinity $S_l$ uniquely characterizes brine fraction for each thickness category, and changes through time from a simple parametrization of brine drainage. The linear vertical salinity profile ($S^*_{kl}$) is reconstructed from the vertical mean \citep{Vancoppenolleetal09b}. The diffusion of heat affects the vertical temperature profile, discretized on a unique layer of snow and multiple ice layers (typically 2-5) for each category, whereas thermal properties depend on local brine fraction. Growth and melt rates are computed, also for each ice category. The choice of the main thermodynamic constants is described in Tab. \ref{PhysicalConstants_table}.87 Ice thermodynamics are formulated assuming that sea ice is covered by snow. Within each thickness category, both snow and sea ice are horizontally uniform, hence each thickness category has a specific ice thickness ($h_i^l$) and snow depth ($h_s^l$). Snow is assumed to be fresh, with constant density and thermal conductivity. Sea ice is assumed to be a \textit{mushy layer}\footnote{Mushy layers are two-phase, two-component porous media.} \citep{worster_1992} of constant density, made of pure ice and brine in thermal equilibrium, related by a linear liquidus relationship \citep{bitz_1999}. A vertically-averaged bulk salinity $S_l$ uniquely characterizes brine fraction for each thickness category, and changes through time from a simple parametrization of brine drainage. The linear vertical salinity profile ($S^*_{kl}$) is reconstructed from the vertical mean \citep{vancoppenolle_2009}. The diffusion of heat affects the vertical temperature profile, discretized on a unique layer of snow and multiple ice layers (typically 2-5) for each category, whereas thermal properties depend on local brine fraction. Growth and melt rates are computed, also for each ice category. The choice of the main thermodynamic constants is described in Tab. \ref{PhysicalConstants_table}. 88 88 89 89 … … 154 154 \subsection{Dynamic formulation} 155 155 156 The formulation of ice dynamics is based on the continuum approach. The latter holds provided the drift ice particles are much larger than single ice floes, and much smaller than typical gradient scales. This compromise is rarely achieved in practice \citep{ Lepparanta05}. Yet the continuum approach generates a convenient momentum equation for the horizontal ice velocity vector $\mathbf{u}=(u,v)$, which can be solved with classical numerical methods (here, finite differences on the NEMO C-grid). The most important term in the momentum equation is internal stress. We follow the viscous-plastic (VP) rheological framework \citep{Hibler79}, assuming that sea ice has no tensile strength but responds to compressive and shear deformations in a plastic way. In practice, the elastic-viscous-plastic (EVP) technique of \citep{Bouillonetal13} is used, more convient numerically than VP. It is well accepted that the VP rheology and its relatives are the minimum complexity to get reasonable ice drift patterns \citep{Kreyscheretal00}, but fail at generating the observed deformation patterns \citep{Girardetal09}. This is a long-lasting problem: what is the ideal rheological model for sea ice and how it should be applied are still being debated \citep[see, e.g.][]{Weiss13}.156 The formulation of ice dynamics is based on the continuum approach. The latter holds provided the drift ice particles are much larger than single ice floes, and much smaller than typical gradient scales. This compromise is rarely achieved in practice \citep{lepp_ranta_2011}. Yet the continuum approach generates a convenient momentum equation for the horizontal ice velocity vector $\mathbf{u}=(u,v)$, which can be solved with classical numerical methods (here, finite differences on the NEMO C-grid). The most important term in the momentum equation is internal stress. We follow the viscous-plastic (VP) rheological framework \citep{hibler_1979}, assuming that sea ice has no tensile strength but responds to compressive and shear deformations in a plastic way. In practice, the elastic-viscous-plastic (EVP) technique of \citep{bouillon_2013} is used, more convient numerically than VP. It is well accepted that the VP rheology and its relatives are the minimum complexity to get reasonable ice drift patterns \citep{kreyscher_2000}, but fail at generating the observed deformation patterns \citep{girard_2009}. This is a long-lasting problem: what is the ideal rheological model for sea ice and how it should be applied are still being debated \citep[see, e.g.][]{weiss_2013}. 157 157 158 158 %------------------------------------------------------------------------------------------------------------------------- … … 162 162 %------------------------------------------------------------------------------------------------------------------------- 163 163 164 We first present the essentials of the thickness distribution framework \citep{ Thorndikeetal75}. Consider a given region of area $R$ centered at spatial coordinates $(\mathbf{x})$ at a given time $t$. $R$ could be e.g. a model grid cell. The ice thickness distribution $g(\mathbf{x},t, h)$ is introduced as follows:164 We first present the essentials of the thickness distribution framework \citep{thorndike_1975}. Consider a given region of area $R$ centered at spatial coordinates $(\mathbf{x})$ at a given time $t$. $R$ could be e.g. a model grid cell. The ice thickness distribution $g(\mathbf{x},t, h)$ is introduced as follows: 165 165 \begin{linenomath} 166 166 \begin{align} … … 168 168 \end{align} 169 169 \end{linenomath} 170 where $dA(h,h+dh)$ is the surface fraction of all parts of $R$ with ice thickness between $h$ and $h+dh$. Using this definition, the spatial structure of ice thickness is lost (see Fig. \ref{fig_g_h}), and $h$ becomes an extra independent variable, next to spatial coordinates and time, that can be thought as random. $g$ is by definition normalized to 1. The conservation of area, expressed in terms of $g(h)$, is given by \citep{ Thorndikeetal75}:170 where $dA(h,h+dh)$ is the surface fraction of all parts of $R$ with ice thickness between $h$ and $h+dh$. Using this definition, the spatial structure of ice thickness is lost (see Fig. \ref{fig_g_h}), and $h$ becomes an extra independent variable, next to spatial coordinates and time, that can be thought as random. $g$ is by definition normalized to 1. The conservation of area, expressed in terms of $g(h)$, is given by \citep{thorndike_1975}: 171 171 \begin{linenomath} 172 172 \begin{align} … … 199 199 \end{align} 200 200 \end{linenomath} 201 Ice volume per area is the extensive counterpart for ice thickness, connected with volume through $h_l^i = v_l^i / a_l$. Evolution equations for extensive variables can be readily derived from equation \ref{eq:gt} by integration between thickness boundaries of the $l^{th}$ category \citep{ Bitzetal01}. This applies to all model extensive variables (see Table \ref{GVariables_table}). For ice area, this reads:201 Ice volume per area is the extensive counterpart for ice thickness, connected with volume through $h_l^i = v_l^i / a_l$. Evolution equations for extensive variables can be readily derived from equation \ref{eq:gt} by integration between thickness boundaries of the $l^{th}$ category \citep{bitz_2001}. This applies to all model extensive variables (see Table \ref{GVariables_table}). For ice area, this reads: 202 202 \begin{linenomath} 203 203 \begin{align} … … 206 206 \end{align} 207 207 \end{linenomath} 208 wher $\Theta^a_l$ refers to the effect of thermodynamics. Enthalpy is a particular case because it also has a vertical depth dependence $z$, which corresponds to $K$ vertical layers of equal thickness. The solution adopted here, following from \cite{ ZhangRothrock01}, is that enthalpy from the individual layers are conserved separately. This is a practical solution, for lack of better.208 wher $\Theta^a_l$ refers to the effect of thermodynamics. Enthalpy is a particular case because it also has a vertical depth dependence $z$, which corresponds to $K$ vertical layers of equal thickness. The solution adopted here, following from \cite{zhang_2001}, is that enthalpy from the individual layers are conserved separately. This is a practical solution, for lack of better. 209 209 210 210 One of the major actions of LIM is to resolve conservation equations for all extensive variables that characterize the ice state. Let us now connect this detailed information with classical sea ice fields. The ice concentration $A$ and the ice volume per area\footnote{Ice volume per area is equivalent to the grid-cell averaged ice thickness.} $V_i$ (m) directly derive from $g$: … … 232 232 \end{align} 233 233 \end{linenomath} 234 where $m=\rho_i V_i + \rho_s V_s $ is the ice and snow mass per unit area, $\mathbf{u}$ is the ice velocity, $\mathbf{\sigma}$ is the internal stress tensor, $\mathbf{\tau}_a$ and $\mathbf{\tau}_w$ are the air and ocean stresses, respectively, $f$ is the Coriolis parameter, $\mathbf{k}$ is a unit vector pointing upwards, $g$ is the gravity acceleration and $\eta$ is the ocean surface elevation. The EVP approach used in LIM \citep{ Bouillonetal13} gives the stress tensor as a function of the strain rate tensor $\dot{\mathbf{\epsilon}}$ and some of the sea ice state variables:234 where $m=\rho_i V_i + \rho_s V_s $ is the ice and snow mass per unit area, $\mathbf{u}$ is the ice velocity, $\mathbf{\sigma}$ is the internal stress tensor, $\mathbf{\tau}_a$ and $\mathbf{\tau}_w$ are the air and ocean stresses, respectively, $f$ is the Coriolis parameter, $\mathbf{k}$ is a unit vector pointing upwards, $g$ is the gravity acceleration and $\eta$ is the ocean surface elevation. The EVP approach used in LIM \citep{bouillon_2013} gives the stress tensor as a function of the strain rate tensor $\dot{\mathbf{\epsilon}}$ and some of the sea ice state variables: 235 235 \begin{linenomath} 236 236 \begin{align} … … 250 250 251 251 Dynamical processes include the conservation of momentum, rheology, transport and mechanical redistribution. 252 To resolve the momentum equation, atmospheric stress is taken either as forcing or from an atmospheric model, oceanic stress and sea surface elevation from the ocean model, the Coriolis term is trivial. The last term, the divergence of the internal stress tensor $\sigma$, is the most critical term in the momentum equation and requires a rheological formulation. The EVP approach used in LIM gives the stress tensor components as \citep{ Bouillonetal13}:252 To resolve the momentum equation, atmospheric stress is taken either as forcing or from an atmospheric model, oceanic stress and sea surface elevation from the ocean model, the Coriolis term is trivial. The last term, the divergence of the internal stress tensor $\sigma$, is the most critical term in the momentum equation and requires a rheological formulation. The EVP approach used in LIM gives the stress tensor components as \citep{bouillon_2013}: 253 253 \begin{linenomath} 254 254 \begin{align} … … 257 257 \end{align} 258 258 \end{linenomath} 259 where $\Delta$ is a particular measure of the deformation rate, $\Delta_{min}$ a parameter determining a smooth transition from pure viscous vlow ($\Delta<<\Delta_{min}$) to pure plastic flow ($\Delta >> \Delta_{min}$), and $e$ is a parameter giving the ratio between the maximum compressive stress and twice the maximum shear stress. In the pure plastic regime, the stress principal components should lie on the edge of an elliptical yield curve (Fig. \ref{fig_yield}). In the viscous regime, they are within the ellipse. The ice strength $P$ determines the plastic failure criterion and connects the momentum equation with the state of the sea ice. $P$ is not well constrained and must be parameterized. The heuristic option of \cite{ Hibler79} was here adopted as a reference formulation:259 where $\Delta$ is a particular measure of the deformation rate, $\Delta_{min}$ a parameter determining a smooth transition from pure viscous vlow ($\Delta<<\Delta_{min}$) to pure plastic flow ($\Delta >> \Delta_{min}$), and $e$ is a parameter giving the ratio between the maximum compressive stress and twice the maximum shear stress. In the pure plastic regime, the stress principal components should lie on the edge of an elliptical yield curve (Fig. \ref{fig_yield}). In the viscous regime, they are within the ellipse. The ice strength $P$ determines the plastic failure criterion and connects the momentum equation with the state of the sea ice. $P$ is not well constrained and must be parameterized. The heuristic option of \cite{hibler_1979} was here adopted as a reference formulation: 260 260 \begin{linenomath} 261 261 \begin{align} … … 294 294 % Parameters 295 295 $P*$ (rn\_pstar) & ice strength thickness param. & 20000 & N/m2 & - \\ 296 $C$ (rn\_crhg) & ice strength concentration param. & 20 & - & \citep{ Hibler79} \\297 $H^*$ (rn\_hstar) & maximum ridged ice thickness param. & 25 & m & \citep{ Lipscombetal07} \\298 $p$ (rn\_por\_rdg) & porosity of new ridges & 0.3 & - & \citep{ Lepparantaetal95} \\296 $C$ (rn\_crhg) & ice strength concentration param. & 20 & - & \citep{hibler_1979} \\ 297 $H^*$ (rn\_hstar) & maximum ridged ice thickness param. & 25 & m & \citep{lipscomb_2007} \\ 298 $p$ (rn\_por\_rdg) & porosity of new ridges & 0.3 & - & \citep{lepp_ranta_1995} \\ 299 299 $amax$ (rn\_amax) & maximum ice concentration & 0.999 & - & -\\ 300 300 $h_0$ (rn\_hnewice) & thickness of newly formed ice & 0.1 & m & - \\ … … 311 311 312 312 313 Transport connects the horizontal velocity fields and the rest of the ice properties. LIM assumes that the ice properties in the different thickness categories are transported at the same velocity. The scheme of \cite{ Prather86}, based on the conservation of 0, 1$^{st}$ and 2$^{nd}$ order moments in $x-$ and $y-$directions, is used, with some numerical diffusion if desired. Whereas this scheme is accurate, nearly conservative, it is also quite expensive since, for each advected field, five moments need to be advected, which proves CPU consuming, in particular when multiple categories are used. Other solutions are currently explored.314 315 The dissipation of energy associated with plastic failure under convergence and shear is accomplished by rafting (overriding of two ice plates) and ridging (breaking of an ice plate and subsequent piling of the broken ice blocks into pressure ridges). Thin ice preferentially rafts whereas thick ice preferentially ridges \citep{ TuhkuriLensu02}. Because observations of these processes are limited, their representation in LIM is rather heuristic. The amount of ice that rafts/ridges depends on the strain rate tensor invariants (shear and divergence) as in \citep{FlatoHibler95}, while the ice categories involved are determined by a participation function favouring thin ice \citep{Lipscombetal07}. The thickness of ice being deformed ($h'$) determines whether ice rafts ($h'<$ 0.75 m) or ridges ($h'>$ 0.75 m), following \cite{Haapala00}. The deformed ice thickness is $2h'$ after rafting, and is distributed between $2h'$ and $2 \sqrt{H^*h'}$ after ridging, where $H^* = 25$ m \citep{Lipscombetal07}. Newly ridged ice is highly porous, effectively trapping seawater. To represent this, a prescribed volume fraction (30\%) of newly ridged ice \citep{Lepparantaetal95} incorporates mass, salt and heat are extracted from the ocean. Hence, in contrast with other models, the net thermodynamic ice production during convergence is not zero in LIM, since mass is added to sea ice during ridging. Consequently, simulated new ridges have high temperature and salinity as observed \citep{Hoyland02}. A fraction of snow (50 \%) falls into the ocean during deformation.313 Transport connects the horizontal velocity fields and the rest of the ice properties. LIM assumes that the ice properties in the different thickness categories are transported at the same velocity. The scheme of \cite{prather_1986}, based on the conservation of 0, 1$^{st}$ and 2$^{nd}$ order moments in $x-$ and $y-$directions, is used, with some numerical diffusion if desired. Whereas this scheme is accurate, nearly conservative, it is also quite expensive since, for each advected field, five moments need to be advected, which proves CPU consuming, in particular when multiple categories are used. Other solutions are currently explored. 314 315 The dissipation of energy associated with plastic failure under convergence and shear is accomplished by rafting (overriding of two ice plates) and ridging (breaking of an ice plate and subsequent piling of the broken ice blocks into pressure ridges). Thin ice preferentially rafts whereas thick ice preferentially ridges \citep{tuhkuri_2002}. Because observations of these processes are limited, their representation in LIM is rather heuristic. The amount of ice that rafts/ridges depends on the strain rate tensor invariants (shear and divergence) as in \citep{flato_1995}, while the ice categories involved are determined by a participation function favouring thin ice \citep{lipscomb_2007}. The thickness of ice being deformed ($h'$) determines whether ice rafts ($h'<$ 0.75 m) or ridges ($h'>$ 0.75 m), following \cite{haapala_2000}. The deformed ice thickness is $2h'$ after rafting, and is distributed between $2h'$ and $2 \sqrt{H^*h'}$ after ridging, where $H^* = 25$ m \citep{lipscomb_2007}. Newly ridged ice is highly porous, effectively trapping seawater. To represent this, a prescribed volume fraction (30\%) of newly ridged ice \citep{lepp_ranta_1995} incorporates mass, salt and heat are extracted from the ocean. Hence, in contrast with other models, the net thermodynamic ice production during convergence is not zero in LIM, since mass is added to sea ice during ridging. Consequently, simulated new ridges have high temperature and salinity as observed \citep{h_yland_2002}. A fraction of snow (50 \%) falls into the ocean during deformation. 316 316 317 317 \section{Ice thermodynamics} … … 319 319 320 320 \subsection{Transport in thickness space} 321 Transport in thickness space describes how vertical growth and melt moves ice state variables among the different thicknesses at a velocity $f(h)$, the net ice growth/melt rate, which needs to be first computed. In discretized form, this term moves ice properties between neighbouring categories. The linear remapping scheme of \cite{ Lipscomb01} is used. This scheme is semi-lagrangian, second-order, is less diffusive and converges faster than other options.321 Transport in thickness space describes how vertical growth and melt moves ice state variables among the different thicknesses at a velocity $f(h)$, the net ice growth/melt rate, which needs to be first computed. In discretized form, this term moves ice properties between neighbouring categories. The linear remapping scheme of \cite{lipscomb_2001} is used. This scheme is semi-lagrangian, second-order, is less diffusive and converges faster than other options. 322 322 323 323 \subsection{Thermodynamic source and sink terms} … … 391 391 %-------------------------------------------------------------------------------------------------------------------- 392 392 393 Based on these, brine fraction reduces to $\phi = -\mu S/T$ (see Fig. \ref{fig_thermal_properties}), where $\mu$ relates the freezing point of brine to salinity, and one can derive the specific enthalpy $q_m(S,T)$, defined as the energy required to warm and melt a unit control volume of sea ice at temperature $T$ (in Celsius) and salinity $S$ until 0$^\circ$ C, taken as a reference zero-energy level \citep{ BitzLipscomb99,Schmidtetal04}:393 Based on these, brine fraction reduces to $\phi = -\mu S/T$ (see Fig. \ref{fig_thermal_properties}), where $\mu$ relates the freezing point of brine to salinity, and one can derive the specific enthalpy $q_m(S,T)$, defined as the energy required to warm and melt a unit control volume of sea ice at temperature $T$ (in Celsius) and salinity $S$ until 0$^\circ$ C, taken as a reference zero-energy level \citep{bitz_1999,schmidt_2004}: 394 394 \begin{linenomath} 395 395 \begin{align} … … 399 399 where $c_i$ is pure ice specific heat, $L$ is latent heat of fusion at 0$^\circ$C, and $c_w$ is water specific heat. The first term expresses the warming of solid ice. The second term expresses internal change in brine fraction, which is often the largest because the Stefan number ($c_i T/L$) is generally small. The last term gives the warming of the remaining water from $T_{fr} = -\mu S$ until 0$^\circ$C. Similar, but simpler and linear expressions for snow and water can be derived. 400 400 401 The second overarching aspect is that all growth and melt processes must be calculated consistently with the enthalpy formulation. Energetics of phase transitions are handled using the formalism of \cite{ Schmidtetal04}. For each phase transition, initial and final states (temperature and salinity) are defined, and the ice-to-ocean mass flux to the ice $F_m$ (kg/s) relates to the energy gain or loss $\Delta Q$ through:401 The second overarching aspect is that all growth and melt processes must be calculated consistently with the enthalpy formulation. Energetics of phase transitions are handled using the formalism of \cite{schmidt_2004}. For each phase transition, initial and final states (temperature and salinity) are defined, and the ice-to-ocean mass flux to the ice $F_m$ (kg/s) relates to the energy gain or loss $\Delta Q$ through: 402 402 \begin{linenomath} 403 403 \begin{align} … … 418 418 \item the sensible heat flux from the ocean to the sea ice ($A.F_w$) 419 419 \end{itemize} 420 Other contributions are not assumed not to contribute. The ocean-to-ice sensible heat flux is formulated the bulk formula of \citep{ McPhee92}.420 Other contributions are not assumed not to contribute. The ocean-to-ice sensible heat flux is formulated the bulk formula of \citep{mcphee_1992}. 421 421 422 422 % Ice growth 423 If $B^{opw}$ is such that the SST would decrease below the freezing point, the remainder of the heat is used to form new ice. The heat loss is converted into a volume of new ice $v_0$. The thickness $h_0$ of the new ice grown during a sea ice time step depends on unresolved small-scale currents and waves and is prescribed. The fraction $a_0=v_0/h_0$ is computed accordingly. The salinity of this new ice $S_0$ is given by the salinity-thickness empirical relationship of \cite{ Kovacs96}. The temperature assumed for this new ice is the local freezing point. If by contrast $B^{opw}$ is positive and there still is ice in the grid cell, then $B^{opw}$ is directly redirected to bottom melting. This argument follows from \cite{MaykutMcPhee95}, who found that most of solar heat absorbed in the surface waters is converted into melting. In practise, this prevents the SST to be above freezing as long ice is present.423 If $B^{opw}$ is such that the SST would decrease below the freezing point, the remainder of the heat is used to form new ice. The heat loss is converted into a volume of new ice $v_0$. The thickness $h_0$ of the new ice grown during a sea ice time step depends on unresolved small-scale currents and waves and is prescribed. The fraction $a_0=v_0/h_0$ is computed accordingly. The salinity of this new ice $S_0$ is given by the salinity-thickness empirical relationship of \cite{kovacs_1996}. The temperature assumed for this new ice is the local freezing point. If by contrast $B^{opw}$ is positive and there still is ice in the grid cell, then $B^{opw}$ is directly redirected to bottom melting. This argument follows from \cite{maykut_1995}, who found that most of solar heat absorbed in the surface waters is converted into melting. In practise, this prevents the SST to be above freezing as long ice is present. 424 424 425 425 $B^{opw}$ can be seen as a predictor of the heat budget of the first ocean level. As such, it only helps to compute new ice formation and the extra bottom melt in summer, but is not part of the conservation of heat in the model. To ensure heat conservation, the heat effectively contributing to changing sea ice is removed from the non-solar flux sent to the ocean. This includes: (i) the heat loss used for ice formation, (ii) the heat gain used to melt ice, and (iii) the sensible heat given by the ocean to the ice. Finally, because ice dynamics are not able to maintain the small amount of open water that is observed, a maximum ice fraction ($amax, <1$) is prescribed. … … 435 435 \end{align} 436 436 \end{linenomath} 437 which state that the local change in enthalpy is given by the divergence of the vertical conduction ($F_c=-k(S,T) \partial T / \partial z$) and radiation ($F_r$) fluxes. $\rho$ is the density of ice or snow. Re-expressed as a function of temperature, this becomes the heat diffusion equation. This equation is non-linear in $T$, because of $q$ and $k$, and its main specificity is that internal melting requires large amounts of energy near the freezing point. The thermal conductivity is formulated following \cite{ Pringleetal07}, empirically accounting for the reduction of thermal conductivity at large brine fractions.437 which state that the local change in enthalpy is given by the divergence of the vertical conduction ($F_c=-k(S,T) \partial T / \partial z$) and radiation ($F_r$) fluxes. $\rho$ is the density of ice or snow. Re-expressed as a function of temperature, this becomes the heat diffusion equation. This equation is non-linear in $T$, because of $q$ and $k$, and its main specificity is that internal melting requires large amounts of energy near the freezing point. The thermal conductivity is formulated following \cite{pringle_2007}, empirically accounting for the reduction of thermal conductivity at large brine fractions. 438 438 439 439 At the ice base, we assume that the temperature is at the local freezing point. Ice grows or melt if the heat balance between the oceanic sensible heat flux ($F_w$) and internal conduction is negative or positive. … … 447 447 where $Q^{sr}$ and $Q^{ns}$ are the net downwelling atmospheric solar and non-solar flux components. If the solution of this equation without melting gives a surface temperature ($T_{su}$) below 0$^\circ$ C, then there is no melting and the heat available for surface melting $Q_{sum}=0$. Otherwise $T_{su}$ is capped at 0$^\circ$ C and $Q_{sum}$ is calculated as a residual. 448 448 449 \textbf{Radiation}. Radiation contributes to the surface and internal heat budget. The radiative transfer scheme is currently basic, composed of surface albedo, transmission through the ice interior and attenuation with vertical depth. The albedo is computed empirically as a function of ice thickness, snow depth and surface temperature, using a reformulation of the parameterization of \cite{ ShineHenderson85}. When snow is present, all the absorbed radiation is transformed into sensible heat available for conduction or melting. Over snow-free ice, a fraction of solar radiation is transmitted below the surface and attenuates exponentially with depth, until it reaches the base of the ice.449 \textbf{Radiation}. Radiation contributes to the surface and internal heat budget. The radiative transfer scheme is currently basic, composed of surface albedo, transmission through the ice interior and attenuation with vertical depth. The albedo is computed empirically as a function of ice thickness, snow depth and surface temperature, using a reformulation of the parameterization of \cite{shine_1985}. When snow is present, all the absorbed radiation is transformed into sensible heat available for conduction or melting. Over snow-free ice, a fraction of solar radiation is transmitted below the surface and attenuates exponentially with depth, until it reaches the base of the ice. 450 450 451 451 \textbf{Growth and melt processes}. Snow grows from precipitation and loses mass from melting and snow-ice conversion once the snow base is below sea level. Sea ice grows and melts by various means. Ice forms by congelation or melt at the base, can melt at the surface and form from snow-to-ice conversion at the snow-ice interface if the latter is below sea level. Some new ice is also added to the system when seawater is trapped into newly formed pressure ridges. 452 452 453 \textbf{Salt dynamics}. Bulk salinity is empirically parameterized, as a function of salt uptake during growth, gravity drainage and flushing. The shape of the vertical profile depends on the bulk salinity \citep{ Vancoppenolleetal09b}.454 455 \textbf{Single-category parameterizations}. If the single-category representation is adopted, then two parameterizations can be activated, following \citep{ FichefetMaqueda97}. First, the thermal conductivity of both ice and snow is multiplied by a factor $>1$ accounting for the unresolved thin ice, effectively increasing the ice growth rate. Second, to account for the loss of thin ice in summer, the ice concentration is reduced in proportion to the loss of ice thickness. Both parameterizations have been tuned to match the results in multi-category mode.453 \textbf{Salt dynamics}. Bulk salinity is empirically parameterized, as a function of salt uptake during growth, gravity drainage and flushing. The shape of the vertical profile depends on the bulk salinity \citep{vancoppenolle_2009}. 454 455 \textbf{Single-category parameterizations}. If the single-category representation is adopted, then two parameterizations can be activated, following \citep{fichefet_1997}. First, the thermal conductivity of both ice and snow is multiplied by a factor $>1$ accounting for the unresolved thin ice, effectively increasing the ice growth rate. Second, to account for the loss of thin ice in summer, the ice concentration is reduced in proportion to the loss of ice thickness. Both parameterizations have been tuned to match the results in multi-category mode. 456 456 457 457 \end{document}
Note: See TracChangeset
for help on using the changeset viewer.