/[MITgcm]/MITgcm_contrib/articles/ceaice/ceaice.tex
ViewVC logotype

Contents of /MITgcm_contrib/articles/ceaice/ceaice.tex

Parent Directory Parent Directory | Revision Log Revision Log | View Revision Graph Revision Graph


Revision 1.5 - (show annotations) (download) (as text)
Mon Jan 14 15:46:54 2008 UTC (17 years, 6 months ago) by mlosch
Branch: MAIN
Changes since 1.4: +68 -27 lines
File MIME type: application/x-tex
saving todays works:
- write something about the downstream island experiments relating to Hunke (2001)
- fix a few paths and usepackage statements
- change the model description slightly to incorporate limiting
  schemes

1 \documentclass[12pt]{article}
2
3 \usepackage[]{graphicx}
4 \usepackage{subfigure}
5
6 \usepackage[round,comma]{natbib}
7 \bibliographystyle{bib/agu04}
8
9 \usepackage{amsmath,amssymb}
10 \newcommand\bmmax{10} \newcommand\hmmax{10}
11 \usepackage{bm}
12
13 % math abbreviations
14 \newcommand{\vek}[1]{\ensuremath{\mathbf{#1}}}
15 \newcommand{\mat}[1]{\ensuremath{\mathbf{#1}}}
16 \newcommand{\vtau}{\bm{{\tau}}}
17
18 \newcommand{\degree}{\ensuremath{^\circ}}
19 \newcommand{\degC}{\,\ensuremath{\degree}C}
20 \newcommand{\degE}{\ensuremath{\degree}\,E}
21 \newcommand{\degS}{\ensuremath{\degree}\,S}
22 \newcommand{\degN}{\ensuremath{\degree}\,N}
23 \newcommand{\degW}{\ensuremath{\degree}\,W}
24
25 % cross reference scheme
26 \newcommand{\reffig}[1]{Figure~\ref{fig:#1}}
27 \newcommand{\reftab}[1]{Table~\ref{tab:#1}}
28 \newcommand{\refapp}[1]{Appendix~\ref{app:#1}}
29 \newcommand{\refsec}[1]{Section~\ref{sec:#1}}
30 \newcommand{\refeq}[1]{\,(\ref{eq:#1})}
31 \newcommand{\refeqs}[2]{\,(\ref{eq:#1})--(\ref{eq:#2})}
32
33 \newlength{\stdfigwidth}\setlength{\stdfigwidth}{20pc}
34 %\newlength{\stdfigwidth}\setlength{\stdfigwidth}{\columnwidth}
35 \newlength{\mediumfigwidth}\setlength{\mediumfigwidth}{39pc}
36 %\newlength{\widefigwidth}\setlength{\widefigwidth}{39pc}
37 \newlength{\widefigwidth}\setlength{\widefigwidth}{\textwidth}
38 \newcommand{\fpath}{figs}
39
40 % commenting scheme
41 \newcommand{\ml}[1]{\textsf{\slshape #1}}
42
43 \title{A Dynamic-Thermodynamic Sea ice Model for Ocean Climate
44 Estimation on an Arakawa C-Grid}
45
46 \author{Martin Losch, Dimitris Menemenlis, Patrick Heimbach, \\
47 Jean-Michel Campin, and Chris Hill}
48 \begin{document}
49
50 \maketitle
51
52 \begin{abstract}
53 Some blabla
54 \end{abstract}
55
56 \section{Introduction}
57 \label{sec:intro}
58
59 more blabla
60
61 \section{Model}
62 \label{sec:model}
63
64 Traditionally, probably for historical reasons and the ease of
65 treating the Coriolis term, most standard sea-ice models are
66 discretized on Arakawa-B-grids \citep[e.g.,][]{hibler79, harder99,
67 kreyscher00, zhang98, hunke97}. From the perspective of coupling a
68 sea ice-model to a C-grid ocean model, the exchange of fluxes of heat
69 and fresh-water pose no difficulty for a B-grid sea-ice model
70 \citep[e.g.,][]{timmermann02a}. However, surface stress is defined at
71 velocities points and thus needs to be interpolated between a B-grid
72 sea-ice model and a C-grid ocean model. While the smoothing implicitly
73 associated with this interpolation may mask grid scale noise, it may
74 in two-way coupling lead to a computational mode as will be shown. By
75 choosing a C-grid for the sea-ice model, we circumvene this difficulty
76 altogether and render the stress coupling as consistent as the
77 buoyancy coupling.
78
79 A further advantage of the C-grid formulation is apparent in narrow
80 straits. In the limit of only one grid cell between coasts there is no
81 flux allowed for a B-grid (with no-slip lateral boundary counditions),
82 whereas the C-grid formulation allows a flux of sea-ice through this
83 passage for all types of lateral boundary conditions. We (will)
84 demonstrate this effect in the Candian archipelago.
85
86 \subsection{Dynamics}
87 \label{sec:dynamics}
88
89 The momentum equations of the sea-ice model are standard with
90 \begin{equation}
91 \label{eq:momseaice}
92 m \frac{D\vek{u}}{Dt} = -mf\vek{k}\times\vek{u} + \vtau_{air} +
93 \vtau_{ocean} - m \nabla{\phi(0)} + \vek{F},
94 \end{equation}
95 where $\vek{u} = u\vek{i}+v\vek{j}$ is the ice velocity vectory, $m$
96 the ice mass per unit area, $f$ the Coriolis parameter, $g$ is the
97 gravity accelation, $\nabla\phi$ is the gradient (tilt) of the sea
98 surface height potential beneath the ice. $\phi$ is the sum of
99 atmpheric pressure $p_{a}$ and loading due to ice and snow
100 $(m_{i}+m_{s})g$. $\vtau_{air}$ and $\vtau_{ocean}$ are the wind and
101 ice-ocean stresses, respectively. $\vek{F}$ is the interaction force
102 and $\vek{i}$, $\vek{j}$, and $\vek{k}$ are the unit vectors in the
103 $x$, $y$, and $z$ directions. Advection of sea-ice momentum is
104 neglected. The wind and ice-ocean stress terms are given by
105 \begin{align*}
106 \vtau_{air} =& \rho_{air} |\vek{U}_{air}|R_{air}(\vek{U}_{air}) \\
107 \vtau_{ocean} =& \rho_{ocean} |\vek{U}_{ocean}-\vek{u}|
108 R_{ocean}(\vek{U}_{ocean}-\vek{u}), \\
109 \end{align*}
110 where $\vek{U}_{air/ocean}$ are the surface winds of the atmosphere
111 and surface currents of the ocean, respectively. $C_{air/ocean}$ are
112 air and ocean drag coefficients, $\rho_{air/ocean}$ reference
113 densities, and $R_{air/ocean}$ rotation matrices that act on the
114 wind/current vectors. $\vek{F} = \nabla\cdot\sigma$ is the divergence
115 of the interal stress tensor $\sigma_{ij}$.
116
117 For an isotropic system this stress tensor can be related to the ice
118 strain rate and strength by a nonlinear viscous-plastic (VP)
119 constitutive law \citep{hibler79, zhang98}:
120 \begin{equation}
121 \label{eq:vpequation}
122 \sigma_{ij}=2\eta(\dot{\epsilon}_{ij},P)\dot{\epsilon}_{ij}
123 + \left[\zeta(\dot{\epsilon}_{ij},P) -
124 \eta(\dot{\epsilon}_{ij},P)\right]\dot{\epsilon}_{kk}\delta_{ij}
125 - \frac{P}{2}\delta_{ij}.
126 \end{equation}
127 The ice strain rate is given by
128 \begin{equation*}
129 \dot{\epsilon}_{ij} = \frac{1}{2}\left(
130 \frac{\partial{u_{i}}}{\partial{x_{j}}} +
131 \frac{\partial{u_{j}}}{\partial{x_{i}}}\right).
132 \end{equation*}
133 The maximum ice pressure $P_{\max}$, a measure of ice strength, depends on
134 both thickness $h$ and compactness (concentration) $c$:
135 \begin{equation}
136 P_{\max} = P^{*}c\,h\,e^{[C^{*}\cdot(1-c)]},
137 \label{icestrength}
138 \end{equation}
139 with the constants $P^{*}$ and $C^{*}$. The nonlinear bulk and shear
140 viscosities $\eta$ and $\zeta$ are functions of ice strain rate
141 invariants and ice strength such that the principal components of the
142 stress lie on an elliptical yield curve with the ratio of major to
143 minor axis $e$ equal to $2$; they are given by:
144 \begin{align*}
145 \zeta =& \min\left(\frac{P_{\max}}{2\max(\Delta,\Delta_{\min})},
146 \zeta_{\max}\right) \\
147 \eta =& \frac{\zeta}{e^2} \\
148 \intertext{with the abbreviation}
149 \Delta = & \left[
150 \left(\dot{\epsilon}_{11}^2+\dot{\epsilon}_{22}^2\right)
151 (1+e^{-2}) + 4e^{-2}\dot{\epsilon}_{12}^2 +
152 2\dot{\epsilon}_{11}\dot{\epsilon}_{22} (1-e^{-2})
153 \right]^{-\frac{1}{2}}
154 \end{align*}
155 The bulk viscosities are bounded above by imposing both a minimum
156 $\Delta_{\min}=10^{-11}\text{\,s}^{-1}$ (for numerical reasons) and a
157 maximum $\zeta_{\max} = P_{\max}/\Delta^*$, where
158 $\Delta^*=(5\times10^{12}/2\times10^4)\text{\,s}^{-1}$. For stress
159 tensor compuation the replacement pressure $P = 2\,\Delta\zeta$
160 \citep{hibler95} is used so that the stress state always lies on the
161 elliptic yield curve by definition.
162
163 In the current implementation, the VP-model is integrated with the
164 semi-implicit line successive over relaxation (LSOR)-solver of
165 \citet{zhang98}, which allows for long time steps that, in our case,
166 is limited by the explicit treatment of the Coriolis term. The
167 explicit treatment of the Coriolis term does not represent a severe
168 limitation because it restricts the time step to approximately the
169 same length as in the ocean model where the Coriolis term is also
170 treated explicitly.
171
172 \citet{hunke97}'s introduced an elastic contribution to the strain
173 rate elatic-viscous-plastic in order to regularize
174 Eq.\refeq{vpequation} in such a way that the resulting
175 elatic-viscous-plastic (EVP) and VP models are identical at steady
176 state,
177 \begin{equation}
178 \label{eq:evpequation}
179 \frac{1}{E}\frac{\partial\sigma_{ij}}{\partial{t}} +
180 \frac{1}{2\eta}\sigma_{ij}
181 + \frac{\eta - \zeta}{4\zeta\eta}\sigma_{kk}\delta_{ij}
182 + \frac{P}{4\zeta}\delta_{ij}
183 = \dot{\epsilon}_{ij}.
184 \end{equation}
185 %In the EVP model, equations for the components of the stress tensor
186 %$\sigma_{ij}$ are solved explicitly. Both model formulations will be
187 %used and compared the present sea-ice model study.
188 The EVP-model uses an explicit time stepping scheme with a short
189 timestep. According to the recommendation of \citet{hunke97}, the
190 EVP-model is stepped forward in time 120 times within the physical
191 ocean model time step (although this parameter is under debate), to
192 allow for elastic waves to disappear. Because the scheme does not
193 require a matrix inversion it is fast in spite of the small timestep
194 \citep{hunke97}. For completeness, we repeat the equations for the
195 components of the stress tensor $\sigma_{1} =
196 \sigma_{11}+\sigma_{22}$, $\sigma_{2}= \sigma_{11}-\sigma_{22}$, and
197 $\sigma_{12}$. Introducing the divergence $D_D =
198 \dot{\epsilon}_{11}+\dot{\epsilon}_{22}$, and the horizontal tension
199 and shearing strain rates, $D_T =
200 \dot{\epsilon}_{11}-\dot{\epsilon}_{22}$ and $D_S =
201 2\dot{\epsilon}_{12}$, respectively and using the above abbreviations,
202 the equations can be written as:
203 \begin{align}
204 \label{eq:evpstresstensor1}
205 \frac{\partial\sigma_{1}}{\partial{t}} + \frac{\sigma_{1}}{2T} +
206 \frac{P}{2T} &= \frac{P}{2T\Delta} D_D \\
207 \label{eq:evpstresstensor2}
208 \frac{\partial\sigma_{2}}{\partial{t}} + \frac{\sigma_{2} e^{2}}{2T}
209 &= \frac{P}{2T\Delta} D_T \\
210 \label{eq:evpstresstensor12}
211 \frac{\partial\sigma_{12}}{\partial{t}} + \frac{\sigma_{12} e^{2}}{2T}
212 &= \frac{P}{4T\Delta} D_S
213 \end{align}
214 Here, the elastic parameter $E$ is redefined in terms of a damping timescale
215 $T$ for elastic waves \[E=\frac{\zeta}{T}.\]
216 $T=E_{0}\Delta{t}$ with the tunable parameter $E_0<1$ and
217 the external (long) timestep $\Delta{t}$. \citet{hunke97} recommend
218 $E_{0} = \frac{1}{3}$.
219
220 For details of the spatial discretization, the reader is referred to
221 \citet{zhang98, zhang03}. Our discretization differs only (but
222 importantly) in the underlying grid, namely the Arakawa C-grid, but is
223 otherwise straightforward. The EVP model in particular is discretized
224 naturally on the C-grid with $\sigma_{1}$ and $\sigma_{2}$ on the
225 center points and $\sigma_{12}$ on the corner (or vorticity) points of
226 the grid. With this choice all derivatives are discretized as central
227 differences and averaging is only involved in computing $\Delta$ and
228 $P$ at vorticity points.
229
230 For a general curvilinear grid, one needs in principle to take metric
231 terms into account that arise in the transformation a curvilinear grid
232 on the sphere. However, for now we can neglect these metric terms
233 because they are very small on the cubed sphere grids used in this
234 paper; in particular, only near the edges of the cubed sphere grid, we
235 expect them to be non-zero, but these edges are at approximately
236 35\degS\ or 35\degN\ which are never covered by sea-ice in our
237 simulations. Everywhere else the coordinate system is locally nearly
238 cartesian. However, for last-glacial-maximum or snowball-earth-like
239 simulations the question of metric terms needs to be reconsidered.
240 Either, one includes these terms as in \citet{zhang03}, or one finds a
241 vector-invariant formulation fo the sea-ice internal stress term that
242 does not require any metric terms, as it is done in the ocean dynamics
243 of the MITgcm \citep{adcroft04:_cubed_sphere}.
244
245 Moving sea ice exerts a stress on the ocean which is the opposite of
246 the stress $\vtau_{ocean}$ in Eq.\refeq{momseaice}. This stess is
247 applied directly to the surface layer of the ocean model. An
248 alternative ocean stress formulation is given by \citet{hibler87}.
249 Rather than applying $\vtau_{ocean}$ directly, the stress is derived
250 from integrating over the ice thickness to the bottom of the oceanic
251 surface layer. In the resulting equation for the \emph{combined}
252 ocean-ice momentum, the interfacial stress cancels and the total
253 stress appears as the sum of windstress and divergence of internal ice
254 stresses: $\delta(z) (\vtau_{air} + \vek{F})/\rho_0$, \citep[see also
255 Eq.\,2 of][]{hibler87}. The disadvantage of this formulation is that
256 now the velocity in the surface layer of the ocean that is used to
257 advect tracers, is really an average over the ocean surface
258 velocity and the ice velocity leading to an inconsistency as the ice
259 temperature and salinity are different from the oceanic variables.
260
261 Sea ice distributions are characterized by sharp gradients and edges.
262 For this reason, we employ a positive 3rd-order advection scheme
263 \citep{hundsdorfer94} for the thermodynamic variables discussed in the
264 next section.
265
266 \subparagraph{boundary conditions: no-slip, free-slip, half-slip}
267
268 \begin{itemize}
269 \item transition from existing B-Grid to C-Grid
270 \item boundary conditions: no-slip, free-slip, half-slip
271 \item fancy (multi dimensional) advection schemes
272 \item VP vs.\ EVP \citep{hunke97}
273 \item ocean stress formulation \citep{hibler87}
274 \end{itemize}
275
276 \subsection{Thermodynamics}
277 \label{sec:thermodynamics}
278
279 In the original formulation the sea ice model \citep{menemenlis05}
280 uses simple thermodynamics following the appendix of
281 \citet{semtner76}. This formulation does not allow storage of heat
282 (heat capacity of ice is zero, and this type of model is often refered
283 to as a ``zero-layer'' model). Upward heat flux is parameterized
284 assuming a linear temperature profile and together with a constant ice
285 conductivity. It is expressed as $(K/h)(T_{w}-T_{0})$, where $K$ is
286 the ice conductivity, $h$ the ice thickness, and $T_{w}-T_{0}$ the
287 difference between water and ice surface temperatures. The surface
288 heat budget is computed in a similar way to that of
289 \citet{parkinson79} and \citet{manabe79}.
290
291 There is considerable doubt about the reliability of such a simple
292 thermodynamic model---\citet{semtner84} found significant errors in
293 phase (one month lead) and amplitude ($\approx$50\%\,overestimate) in
294 such models---, so that today many sea ice models employ more complex
295 thermodynamics. A popular thermodynamics model of \citet{winton00} is
296 based on the 3-layer model of \citet{semtner76} and treats brine
297 content by means of enthalphy conservation. This model requires in
298 addition to ice-thickness and compactness (fractional area) additional
299 state variables to be advected by ice velocities, namely enthalphy of
300 the two ice layers and the thickness of the overlying snow layer.
301
302 \section{Funnel Experiments}
303 \label{sec:funnel}
304
305 For a first/detailed comparison between the different variants of the
306 MIT sea ice model an idealized geometry of a periodic channel,
307 1000\,km long and 500\,m wide on a non-rotating plane, with converging
308 walls forming a symmetric funnel and a narrow strait of 40\,km width
309 is used. The horizontal resolution is 5\,km throughout the domain
310 making the narrow strait 8 grid points wide. The ice model is
311 initialized with a complete ice cover of 50\,cm uniform thickness. The
312 ice model is driven by a constant along channel eastward ocean current
313 of 25\,cm/s that does not see the walls in the domain. All other
314 ice-ocean-atmosphere interactions are turned off, in particular there
315 is no feedback of ice dynamics on the ocean current. All thermodynamic
316 processes are turned off so that ice thickness variations are only
317 caused by convergent or divergent ice flow. Ice volume (effective
318 thickness) and concentration are advected with a third-order scheme
319 with a flux limiter \citep{hundsdorfer94} to avoid undershoots. This
320 scheme is unconditionally stable and does not require additional
321 diffusion. The time step used here is 1\,h.
322
323 \reffig{funnelf0} compares the dynamic fields ice concentration $c$,
324 effective thickness $h_{eff} = h\cdot{c}$, and velocities $(u,v)$ for
325 five different cases at steady state (after 10\,years of integration):
326 \begin{description}
327 \item[B-LSRns:] LSR solver with no-slip boundary conditions on a B-grid,
328 \item[C-LSRns:] LSR solver with no-slip boundary conditions on a C-grid,
329 \item[C-LSRfs:] LSR solver with free-slip boundary conditions on a C-grid,
330 \item[C-EVPns:] EVP solver with no-slip boundary conditions on a C-grid,
331 \item[C-EVPfs:] EVP solver with free-slip boundary conditions on a C-grid,
332 \end{description}
333 \ml{[We have not implemented the EVP solver on a B-grid.]}
334 \begin{figure*}[htbp]
335 \includegraphics[width=\widefigwidth]{\fpath/all_086280}
336 \caption{Ice concentration, effective thickness [m], and ice
337 velocities [m/s]
338 for 5 different numerical solutions.}
339 \label{fig:funnelf0}
340 \end{figure*}
341 At a first glance, the solutions look similar. This is encouraging as
342 the details of discretization and numerics should not affect the
343 solutions to first order. In all cases the ice-ocean stress pushes the
344 ice cover eastwards, where it converges in the funnel. In the narrow
345 channel the ice moves quickly (nearly free drift) and leaves the
346 channel as narrow band.
347
348 A close look reveals interesting differences between the B- and C-grid
349 results. The zonal velocity in the narrow channel is nearly the free
350 drift velocity ( = ocean velocity) of 25\,cm/s for the C-grid
351 solutions, regardless of the boundary conditions, while it is just
352 above 20\,cm/s for the B-grid solution. The ice accelerates to
353 25\,cm/s after it exits the channel. Concentrating on the solutions
354 B-LSRns and C-LSRns, the ice volume (effective thickness) along the
355 boundaries in the narrow channel is larger in the B-grid case although
356 the ice concentration is reduces in the C-grid case. The combined
357 effect leads to a larger actual ice thickness at smaller
358 concentrations in the C-grid case. However, since the effective
359 thickness determines the ice strength $P$ in Eq\refeq{icestrength},
360 the ice strength and thus the bulk and shear viscosities are larger in
361 the B-grid case leading to more horizontal friction. This circumstance
362 might explain why the no-slip boundary conditions in the B-grid case
363 appear to be more effective in reducing the flow within the narrow
364 channel, than in the C-grid case. Further, the viscosities are also
365 sensitive to details of the velocity gradients. Via $\Delta$, these
366 gradients enter the viscosities in the denominator so that large
367 gradients tend to reduce the viscosities. This again favors more flow
368 along the boundaries in the C-grid case: larger velocities
369 (\reffig{funnelf0}) on grid points that are closer to the boundary by
370 a factor $\frac{1}{2}$ than in the B-grid case because of the stagger
371 nature of the C-grid lead numerically to larger tangential gradients
372 across the boundary; these in turn make the viscosities smaller for
373 less tangential friction and allow more tangential flow along the
374 boundaries.
375
376 The above argument can also be invoked to explain the small
377 differences between the free-slip and no-slip solutions on the C-grid.
378 Because of the non-linearities in the ice viscosities, in particular
379 along the boundaries, the no-slip boundary conditions has only a small
380 impact on the solution.
381
382 The difference between LSR and EVP solutions is largest in the
383 effective thickness and meridional velocity fields. The velocity field
384 appears to be a little noisy. This noise has been address by
385 \citet{hunke01}. It can be dealt with by reducing EVP's internal time
386 step (increasing the number of iterations along with the computational
387 cost) or by regularizing the bulk and shear viscosities. We revisit
388 the latter option by reproducing some of the results of
389 \citet{hunke01}, namely the experiment described in her section~4, for
390 our C-grid no-slip cases: in a square domain with a few islands the
391 ice model is initialized with constant ice thickness and linearly
392 increasing ice concentration to the east. The model dynamics are
393 forced with a constant anticyclonic ocean gyre and variable
394 atmospheric wind whose mean directed diagnonally to the north-east
395 corner of the domain; ice volume and concentration are held constant
396 (no advection by ice velocity). \reffig{hunke01} shows the ice
397 velocity field, its divergence, and the bulk viscosity $\zeta$ for the
398 cases C-LRSns and C-EVPns, and for a C-EVPns case, where
399 \citet{hunke01}'s regularization has been implemented; compare to
400 Fig.\,4 in \citet{hunke01}. The regularization contraint limits ice
401 strength and viscosities as a function of damping time scale,
402 resolution and EVP-time step, effectively allowing the elastic waves to
403 damp out more quickly \citep{hunke01}.
404 \begin{figure*}[htbp]
405 \includegraphics[width=\widefigwidth]{\fpath/hun12days}
406 \caption{Hunke's test case.}
407 \label{fig:hunke01}
408 \end{figure*}
409
410 In the far right (``east'') side of the domain the ice concentration
411 is close to one and the ice should be nearly rigid. The applied wind
412 tends to push ice toward the upper right corner. Because the highly
413 compact ice is confinded by the boundary, it resists any further
414 compression and exhibits little motion in the rigid region on the
415 right hand side. The C-LSRns solution (top row) allows high
416 viscosities in the rigid region suppressing nearly all flow.
417 \citet{hunke01}'s regularization for the C-EVPns solution (bottom row)
418 clearly suppresses the noise present in $\nabla\cdot\vek{u}$ in the
419 unregularized case (middle row), at the cost of reduced viscosities
420 These reduced viscosities lead to small but finite ice velocities
421 which in turn can have a strong effect on solutions in the limit of
422 nearly rigid regimes (arching and blocking, not shown).
423
424
425 %\begin{itemize}
426 %\item B-grid LSR no-slip
427 %\item C-grid LSR no-slip
428 %\item C-grid LSR slip
429 %\item C-grid EVP no-slip
430 %\item C-grid EVP slip
431 %\end{itemize}
432
433 %\subsection{B-grid vs.\ C-grid}
434 %Comparison between:
435 %\begin{itemize}
436 %\item B-grid, lsr, no-slip
437 %\item C-grid, lsr, no-slip
438 %\item C-grid, evp, no-slip
439 %\end{itemize}
440 %all without ice-ocean stress, because ice-ocean stress does not work
441 %for B-grid.
442
443 \subsection{C-grid}
444 \begin{itemize}
445 \item no-slip vs. free-slip for both lsr and evp;
446 "diagnostics" to look at and use for comparison
447 \begin{itemize}
448 \item ice transport through Fram Strait/Denmark Strait/Davis
449 Strait/Bering strait (these are general)
450 \item ice transport through narrow passages, e.g.\ Nares-Strait
451 \end{itemize}
452 \item compare different advection schemes (if lsr turns out to be more
453 effective, then with lsr otherwise I prefer evp), eg.
454 \begin{itemize}
455 \item default 2nd-order with diff1=0.002
456 \item 1st-order upwind with diff1=0.
457 \item DST3FL (SEAICEadvScheme=33 with diff1=0., should work, works for me)
458 \item 2nd-order wit flux limiter (SEAICEadvScheme=77 with diff1=0.)
459 \end{itemize}
460 That should be enough. Here, total ice mass and location of ice edge
461 is interesting. However, this comparison can be done in an idealized
462 domain, may not require full Arctic Domain?
463 \item
464 Do a little study on the parameters of LSR and EVP
465 \begin{enumerate}
466 \item convergence of LSR, how many iterations do you need to get a
467 true elliptic yield curve
468 \item vary deltaTevp and the relaxation parameter for EVP and see when
469 the EVP solution breaks down (relative to the forcing time scale).
470 For this, it is essential that the evp solver gives use "stripeless"
471 solutions, that is your dtevp = 1sec solutions/or 10sec solutions
472 with SEAICE\_evpDampC = 615.
473 \end{enumerate}
474 \end{itemize}
475
476 \section{Forward sensitivity experiments}
477 \label{sec:forward}
478
479 A second series of forward sensitivity experiments have been carried out on an
480 Arctic Ocean domain with open boundaries. Once again the objective is to
481 compare the old B-grid LSR dynamic solver with the new C-grid LSR and EVP
482 solvers. One additional experiment is carried out to illustrate the
483 differences between the two main options for sea ice thermodynamics in the MITgcm.
484
485 \subsection{Arctic Domain with Open Boundaries}
486 \label{sec:arctic}
487
488 The Arctic domain of integration is illustrated in Fig.~\ref{???}. It is
489 carved out from, and obtains open boundary conditions from, the global
490 cubed-sphere configuration of the Estimating the Circulation and Climate of
491 the Ocean, Phase II (ECCO2) project \cite{men05a}. The domain size is 420 by
492 384 grid boxes horizontally with mean horizontal grid spacing of 18 km.
493
494 There are 50 vertical levels ranging in thickness from 10 m near the surface
495 to approximately 450 m at a maximum model depth of 6150 m. Bathymetry is from
496 the National Geophysical Data Center (NGDC) 2-minute gridded global relief
497 data (ETOPO2) and the model employs the partial-cell formulation of
498 \cite{adc97}, which permits accurate representation of the bathymetry. The
499 model is integrated in a volume-conserving configuration using a finite volume
500 discretization with C-grid staggering of the prognostic variables. In the
501 ocean, the non-linear equation of state of \cite{jac95}. The ocean model is
502 coupled to a sea-ice model described hereinabove.
503
504 This particular ECCO2 simulation is initialized from rest using the January
505 temperature and salinity distribution from the World Ocean Atlas 2001 (WOA01)
506 [Conkright et al., 2002] and it is integrated for 32 years prior to the
507 1996-2001 period discussed in the study. Surface boundary conditions are from
508 the National Centers for Environmental Prediction and the National Center for
509 Atmospheric Research (NCEP/NCAR) atmospheric reanalysis [Kistler et al.,
510 2001]. Six-hourly surface winds, temperature, humidity, downward short- and
511 long-wave radiations, and precipitation are converted to heat, freshwater, and
512 wind stress fluxes using the Large and Pond [1981, 1982] bulk
513 formulae. Shortwave radiation decays exponentially as per Paulson and Simpson
514 [1977]. Additionally the time-mean river run-off from Large and Nurser [2001]
515 is applied and there is a relaxation to the monthly-mean climatological sea
516 surface salinity values from WOA01 with a relaxation time scale of 3
517 months. Vertical mixing follows Large et al. [1994] with background vertical
518 diffusivity of 1.5 × 10-5 m2 s-1 and viscosity of 10-3 m2 s-1. A third order,
519 direct-space-time advection scheme with flux limiter is employed and there is
520 no explicit horizontal diffusivity. Horizontal viscosity follows Leith [1996]
521 but modified to sense the divergent flow as per Fox-Kemper and Menemenlis [in
522 press]. Shortwave radiation decays exponentially as per Paulson and Simpson
523 [1977]. Additionally, the time-mean runoff of Large and Nurser [2001] is
524 applied near the coastline and, where there is open water, there is a
525 relaxation to monthly-mean WOA01 sea surface salinity with a time constant of
526 45 days.
527
528 Open water, dry
529 ice, wet ice, dry snow, and wet snow albedo are, respectively, 0.15, 0.85,
530 0.76, 0.94, and 0.8.
531
532 \begin{itemize}
533 \item Configuration
534 \item OBCS from cube
535 \item forcing
536 \item 1/2 and full resolution
537 \item with a few JFM figs from C-grid LSR no slip
538 ice transport through Canadian Archipelago
539 thickness distribution
540 ice velocity and transport
541 \end{itemize}
542
543 \begin{itemize}
544 \item Arctic configuration
545 \item ice transport through straits and near boundaries
546 \item focus on narrow straits in the Canadian Archipelago
547 \end{itemize}
548
549 \begin{itemize}
550 \item B-grid LSR no-slip
551 \item C-grid LSR no-slip
552 \item C-grid LSR slip
553 \item C-grid EVP no-slip
554 \item C-grid EVP slip
555 \item C-grid LSR no-slip + Winton
556 \item speed-performance-accuracy (small)
557 ice transport through Canadian Archipelago differences
558 thickness distribution differences
559 ice velocity and transport differences
560 \end{itemize}
561
562 We anticipate small differences between the different models due to:
563 \begin{itemize}
564 \item advection schemes: along the ice-edge and regions with large
565 gradients
566 \item C-grid: more transport through narrow straits for no slip
567 conditons, less for free slip
568 \item VP vs.\ EVP: speed performance, accuracy?
569 \item ocean stress: different water mass properties beneath the ice
570 \end{itemize}
571
572 \section{Adjoint sensiivities of the MITsim}
573
574 \subsection{The adjoint of MITsim}
575
576 The ability to generate tangent linear and adjoint model components
577 of the MITsim has been a main design task.
578 For the ocean the adjoint capability has proven to be an
579 invaluable tool for sensitivity analysis as well as state estimation.
580 In short, the adjoint enables very efficient computation of the gradient
581 of scalar-valued model diagnostics (called cost function or objective function)
582 with respect to many model "variables".
583 These variables can be two- or three-dimensional fields of initial
584 conditions, model parameters such as mixing coefficients, or
585 time-varying surface or lateral (open) boundary conditions.
586 When combined, these variables span a potentially high-dimensional
587 (e.g. O(10$^8$)) so-called control space. Performing parameter perturbations
588 to assess model sensitivities quickly becomes prohibitive at these scales.
589 Alternatively, (time-varying) sensitivities of the objective function
590 to any element of the control space can be computed very efficiently in
591 one single adjoint
592 model integration, provided an efficient adjoint model is available.
593
594 [REFERENCES]
595
596
597 The adjoint operator (ADM) is the transpose of the tangent linear operator (TLM)
598 of the full (in general nonlinear) forward model, i.e. the MITsim.
599 The TLM maps perturbations of elements of the control space
600 (e.g. initial ice thickness distribution)
601 via the model Jacobian
602 to a perturbation in the objective function
603 (e.g. sea-ice export at the end of the integration interval).
604 \textit{Tangent} linearity ensures that the derivatives are evaluated
605 with respect to the underlying model trajectory at each point in time.
606 This is crucial for nonlinear trajectories and the presence of different
607 regimes (e.g. effect of the seaice growth term at or away from the
608 freezing point of the ocean surface).
609 Ensuring tangent linearity can be easily achieved by integrating
610 the full model in sync with the TLM to provide the underlying model state.
611 Ensuring \textit{tangent} adjoints is equally crucial, but much more
612 difficult to achieve because of the reverse nature of the integration:
613 the adjoint accumulates sensitivities backward in time,
614 starting from a unit perturbation of the objective function.
615 The adjoint model requires the model state in reverse order.
616 This presents one of the major complications in deriving an
617 exact, i.e. \textit{tangent} adjoint model.
618
619 Following closely the development and maintenance of TLM and ADM
620 components of the MITgcm we have relied heavily on the
621 autmomatic differentiation (AD) tool
622 "Transformation of Algorithms in Fortran" (TAF)
623 developed by Fastopt (Giering and Kaminski, 1998)
624 to derive TLM and ADM code of the MITsim.
625 Briefly, the nonlinear parent model is fed to the AD tool which produces
626 derivative code for the specified control space and objective function.
627 Following this approach has (apart from its evident success)
628 several advantages:
629 (1) the adjoint model is the exact adjoint operator of the parent model,
630 (2) the adjoint model can be kept up to date with respect to ongoing
631 development of the parent model, and adjustments to the parent model
632 to extend the automatically generated adjoint are incremental changes
633 only, rather than extensive re-developments,
634 (3) the parallel structure of the parent model is preserved
635 by the adjoint model, ensuring efficient use in high performance
636 computing environments.
637
638 Some initial code adjustments are required to support dependency analysis
639 of the flow reversal and certain language limitations which may lead
640 to irreducible flow graphs (e.g. GOTO statements).
641 The problem of providing the required model state in reverse order
642 at the time of evaluating nonlinear or conditional
643 derivatives is solved via balancing
644 storing vs. recomputation of the model state in a multi-level
645 checkpointing loop.
646 Again, an initial code adjustment is required to support TAFs
647 checkpointing capability.
648 The code adjustments are sufficiently simply so as not to cause
649 major limitations to the full nonlinear parent model.
650 Once in place, an adjoint model of a new model configuration
651 may be derived in about 10 minutes.
652
653 [HIGHLIGHT COUPLED NATURE OF THE ADJOINT!]
654
655 \subsection{Special considerations}
656
657 * growth term(?)
658
659 * small active denominators
660
661 * dynamic solver (implicit function theorem)
662
663 * approximate adjoints
664
665
666 \subsection{An example: sensitivities of sea-ice export through Fram Strait}
667
668 We demonstrate the power of the adjoint method
669 in the context of investigating sea-ice export sensitivities through Fram Strait
670 (for details of this study see Heimbach et al., 2007).
671 The domain chosen is a coarsened version of the Arctic face of the
672 high-resolution cubed-sphere configuration of the ECCO2 project
673 (see Menemenlis et al. 2005). It covers the entire Arctic,
674 extends into the North Pacific such as to cover the entire
675 ice-covered regions, and comprises parts of the North Atlantic
676 down to XXN to enable analysis of remote influences of the
677 North Atlantic current to sea-ice variability and export.
678 The horizontal resolution varies between XX and YY km
679 with 50 unevenly spaced vertical levels.
680 The adjoint models run efficiently on 80 processors
681 (benchmarks have been performed both on an SGI Altix as well as an
682 IBM SP5 at NASA/ARC).
683
684 Following a 1-year spinup, the model has been integrated for four years
685 between 1992 and 1995.
686 It is forced using realistic 6-hourly NCEP/NCAR atmospheric state variables.
687 Over the open ocean these are converted into
688 air-sea fluxes via the bulk formulae of Large and Yeager (2004).
689 Derivation of air-sea fluxes in the presence of sea-ice is handled
690 by the ice model as described in Section XXX.
691 The objective function chosen is sea-ice export through Fram Strait
692 computed for December 1995
693 The adjoint model computes sensitivities to sea-ice export back in time
694 from 1995 to 1992 along this trajectory.
695 In principle all adjoint model variable (i.e. Lagrange multipliers)
696 of the coupled ocean/sea-ice model
697 are available to analyze the transient sensitivity behaviour
698 of the ocean and sea-ice state.
699 Over the open ocean, the adjoint of the bulk formula scheme
700 computes sensitivities to the time-varying atmospheric state.
701 Over ice-covered parts, the sea-ice adjoint converts
702 surface ocean sensitivities to atmospheric sensitivities.
703
704 Fig. XXX(a--d) depict sensitivities of sea-ice export through Fram Strait
705 in December 1995 to changes in sea-ice thickness
706 12, 24, 36, 48 months back in time.
707 Corresponding sensitivities to ocean surface temperature are
708 depicted in Fig. XXX(a--d).
709 The main characteristics is consistency with expected advection
710 of sea-ice over the relevant time scales considered.
711 The general positive pattern means that an increase in
712 sea-ice thickness at location $(x,y)$ and time $t$ will increase
713 sea-ice export through Fram Strait at time $T_e$.
714 Largest distances from Fram Strait indicate fastest sea-ice advection
715 over the time span considered.
716 The ice thickness sensitivities are in close correspondence to
717 ocean surface sentivitites, but of opposite sign.
718 An increase in temperature will incur ice melting, decrease in ice thickness,
719 and therefore decrease in sea-ice export at time $T_e$.
720
721 The picture is fundamentally different and much more complex
722 for sensitivities to ocean temperatures away from the surface.
723 Fig. XXX (a--d) depicts ice export sensitivities to
724 temperatures at roughly 400 m depth.
725 Primary features are the effect of the heat transport of the North
726 Atlantic current which feeds into the West Spitsbergen current,
727 the circulation around Svalbard, and ...
728
729 \begin{figure}[t!]
730 \centerline{
731 \subfigure[{\footnotesize -12 months}]
732 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim072_cmax2.0E+02.eps}}
733 %\includegraphics*[width=.3\textwidth]{H_c.bin_res_100_lev1.pdf}
734 %
735 \subfigure[{\footnotesize -24 months}]
736 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim145_cmax2.0E+02.eps}}
737 }
738
739 \centerline{
740 \subfigure[{\footnotesize
741 -36 months}]
742 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim218_cmax2.0E+02.eps}}
743 %
744 \subfigure[{\footnotesize
745 -48 months}]
746 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim292_cmax2.0E+02.eps}}
747 }
748 \caption{Sensitivity of sea-ice export through Fram Strait in December 2005 to
749 sea-ice thickness at various prior times.
750 \label{fig:4yradjheff}}
751 \end{figure}
752
753
754 \begin{figure}[t!]
755 \centerline{
756 \subfigure[{\footnotesize -12 months}]
757 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim072_cmax5.0E+01.eps}}
758 %\includegraphics*[width=.3\textwidth]{H_c.bin_res_100_lev1.pdf}
759 %
760 \subfigure[{\footnotesize -24 months}]
761 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim145_cmax5.0E+01.eps}}
762 }
763
764 \centerline{
765 \subfigure[{\footnotesize
766 -36 months}]
767 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim218_cmax5.0E+01.eps}}
768 %
769 \subfigure[{\footnotesize
770 -48 months}]
771 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim292_cmax5.0E+01.eps}}
772 }
773 \caption{Same as Fig. XXX but for sea surface temperature
774 \label{fig:4yradjthetalev1}}
775 \end{figure}
776
777
778
779 \section{Discussion and conclusion}
780 \label{sec:concl}
781
782 The story of the paper could be:
783 B-grid ice model + C-grid ocean model does not work properly for us,
784 therefore C-grid ice model with advantages:
785 \begin{enumerate}
786 \item stress coupling simpler (no interpolation required)
787 \item different boundary conditions
788 \item advection schemes carry over trivially from main code
789 \end{enumerate}
790 LSR/EVP solutions are similar with appropriate bcs, evp parameters as
791 a function of forcing time scale (when does VP solution break
792 down). Same for LSR solver, provided that it works (o:
793 Which scheme is more efficient for the resolution/time stepping
794 parameters that we use here. What about other resolutions?
795
796 \paragraph{Acknowledgements}
797 We thank Jinlun Zhang for providing the original B-grid code and many
798 helpful discussions.
799
800 %\bibliography{bib/journal_abrvs,bib/seaice,bib/genocean,bib/maths,bib/mitgcmuv,bib/fram}
801 \bibliography{journal_abrvs,seaice,genocean,maths,mitgcmuv,bib/fram}
802
803 \end{document}
804
805 %%% Local Variables:
806 %%% mode: latex
807 %%% TeX-master: t
808 %%% End:
809
810
811 A Dynamic-Thermodynamic Sea ice Model for Ocean Climate
812 Estimation on an Arakawa C-Grid
813
814 Introduction
815
816 Ice Model:
817 Dynamics formulation.
818 B-C, LSR, EVP, no-slip, slip
819 parallellization
820 Thermodynamics formulation.
821 0-layer Hibler salinity + snow
822 3-layer Winton
823
824 Idealized tests
825 Funnel Experiments
826 Downstream Island tests
827 B-grid LSR no-slip
828 C-grid LSR no-slip
829 C-grid LSR slip
830 C-grid EVP no-slip
831 C-grid EVP slip
832
833 Arctic Setup
834 Configuration
835 OBCS from cube
836 forcing
837 1/2 and full resolution
838 with a few JFM figs from C-grid LSR no slip
839 ice transport through Canadian Archipelago
840 thickness distribution
841 ice velocity and transport
842
843 Arctic forward sensitivity experiments
844 B-grid LSR no-slip
845 C-grid LSR no-slip
846 C-grid LSR slip
847 C-grid EVP no-slip
848 C-grid EVP slip
849 C-grid LSR no-slip + Winton
850 speed-performance-accuracy (small)
851 ice transport through Canadian Archipelago differences
852 thickness distribution differences
853 ice velocity and transport differences
854
855 Adjoint sensitivity experiment on 1/2-res setup
856 Sensitivity of sea ice volume flow through Fram Strait
857 *** Sensitivity of sea ice volume flow through Canadian Archipelago
858
859 Summary and conluding remarks

  ViewVC Help
Powered by ViewVC 1.1.22