/[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 1.10 - (show annotations) (download) (as text)
Mon Feb 25 16:50:56 2008 UTC (17 years, 4 months ago) by dimitri
Branch: MAIN
Changes since 1.9: +13 -3 lines
File MIME type: application/x-tex
addded preliminary abstract

1 % $Header: /u/gcmpack/MITgcm_contrib/articles/ceaice/ceaice.tex,v 1.9 2008/01/21 08:06:00 mlosch Exp $
2 % $Name: $
3 \documentclass[12pt]{article}
4
5 \usepackage[]{graphicx}
6 \usepackage{subfigure}
7
8 \usepackage[round,comma]{natbib}
9 \bibliographystyle{bib/agu04}
10
11 \usepackage{amsmath,amssymb}
12 \newcommand\bmmax{10} \newcommand\hmmax{10}
13 \usepackage{bm}
14
15 % math abbreviations
16 \newcommand{\vek}[1]{\ensuremath{\mathbf{#1}}}
17 \newcommand{\mat}[1]{\ensuremath{\mathbf{#1}}}
18 \newcommand{\vtau}{\bm{{\tau}}}
19
20 \newcommand{\degree}{\ensuremath{^\circ}}
21 \newcommand{\degC}{\,\ensuremath{\degree}C}
22 \newcommand{\degE}{\ensuremath{\degree}\,E}
23 \newcommand{\degS}{\ensuremath{\degree}\,S}
24 \newcommand{\degN}{\ensuremath{\degree}\,N}
25 \newcommand{\degW}{\ensuremath{\degree}\,W}
26
27 % cross reference scheme
28 \newcommand{\reffig}[1]{Figure~\ref{fig:#1}}
29 \newcommand{\reftab}[1]{Table~\ref{tab:#1}}
30 \newcommand{\refapp}[1]{Appendix~\ref{app:#1}}
31 \newcommand{\refsec}[1]{Section~\ref{sec:#1}}
32 \newcommand{\refeq}[1]{\,(\ref{eq:#1})}
33 \newcommand{\refeqs}[2]{\,(\ref{eq:#1})--(\ref{eq:#2})}
34
35 \newlength{\stdfigwidth}\setlength{\stdfigwidth}{20pc}
36 %\newlength{\stdfigwidth}\setlength{\stdfigwidth}{\columnwidth}
37 \newlength{\mediumfigwidth}\setlength{\mediumfigwidth}{39pc}
38 %\newlength{\widefigwidth}\setlength{\widefigwidth}{39pc}
39 \newlength{\widefigwidth}\setlength{\widefigwidth}{\textwidth}
40 \newcommand{\fpath}{figs}
41
42 % commenting scheme
43 \newcommand{\ml}[1]{\textsf{\slshape #1}}
44
45 \title{A Dynamic-Thermodynamic Sea ice Model for Ocean Climate
46 Estimation on an Arakawa C-Grid}
47
48 \author{Martin Losch, Dimitris Menemenlis, Patrick Heimbach, \\
49 Jean-Michel Campin, and Chris Hill}
50 \begin{document}
51
52 \maketitle
53
54 \begin{abstract}
55
56 As part of ongoing efforts to obtain a best possible synthesis of most
57 available, global-scale, ocean and sea ice data, dynamic and thermodynamic
58 sea-ice model components have been incorporated in the Massachusetts Institute
59 of Technology general circulation model (MITgcm). Sea-ice dynamics use either
60 a visco-plastic rheology solved with a line successive relaxation (LSR)
61 technique, reformulated on an Arakawa C-grid in order to match the oceanic and
62 atmospheric grids of the MITgcm, and modified to permit efficient and accurate
63 automatic differentiation of the coupled ocean and sea-ice model
64 configurations.
65
66 \end{abstract}
67
68 \section{Introduction}
69 \label{sec:intro}
70
71 more blabla
72
73 \section{Model}
74 \label{sec:model}
75
76 Traditionally, probably for historical reasons and the ease of
77 treating the Coriolis term, most standard sea-ice models are
78 discretized on Arakawa-B-grids \citep[e.g.,][]{hibler79, harder99,
79 kreyscher00, zhang98, hunke97}. From the perspective of coupling a
80 sea ice-model to a C-grid ocean model, the exchange of fluxes of heat
81 and fresh-water pose no difficulty for a B-grid sea-ice model
82 \citep[e.g.,][]{timmermann02a}. However, surface stress is defined at
83 velocities points and thus needs to be interpolated between a B-grid
84 sea-ice model and a C-grid ocean model. While the smoothing implicitly
85 associated with this interpolation may mask grid scale noise, it may
86 in two-way coupling lead to a computational mode as will be shown. By
87 choosing a C-grid for the sea-ice model, we circumvene this difficulty
88 altogether and render the stress coupling as consistent as the
89 buoyancy coupling.
90
91 A further advantage of the C-grid formulation is apparent in narrow
92 straits. In the limit of only one grid cell between coasts there is no
93 flux allowed for a B-grid (with no-slip lateral boundary counditions),
94 whereas the C-grid formulation allows a flux of sea-ice through this
95 passage for all types of lateral boundary conditions. We (will)
96 demonstrate this effect in the Candian archipelago.
97
98 \subsection{Dynamics}
99 \label{sec:dynamics}
100
101 The momentum equations of the sea-ice model are standard with
102 \begin{equation}
103 \label{eq:momseaice}
104 m \frac{D\vek{u}}{Dt} = -mf\vek{k}\times\vek{u} + \vtau_{air} +
105 \vtau_{ocean} - m \nabla{\phi(0)} + \vek{F},
106 \end{equation}
107 where $\vek{u} = u\vek{i}+v\vek{j}$ is the ice velocity vectory, $m$
108 the ice mass per unit area, $f$ the Coriolis parameter, $g$ is the
109 gravity accelation, $\nabla\phi$ is the gradient (tilt) of the sea
110 surface height potential beneath the ice. $\phi$ is the sum of
111 atmpheric pressure $p_{a}$ and loading due to ice and snow
112 $(m_{i}+m_{s})g$. $\vtau_{air}$ and $\vtau_{ocean}$ are the wind and
113 ice-ocean stresses, respectively. $\vek{F}$ is the interaction force
114 and $\vek{i}$, $\vek{j}$, and $\vek{k}$ are the unit vectors in the
115 $x$, $y$, and $z$ directions. Advection of sea-ice momentum is
116 neglected. The wind and ice-ocean stress terms are given by
117 \begin{align*}
118 \vtau_{air} =& \rho_{air} |\vek{U}_{air}|R_{air}(\vek{U}_{air}) \\
119 \vtau_{ocean} =& \rho_{ocean} |\vek{U}_{ocean}-\vek{u}|
120 R_{ocean}(\vek{U}_{ocean}-\vek{u}), \\
121 \end{align*}
122 where $\vek{U}_{air/ocean}$ are the surface winds of the atmosphere
123 and surface currents of the ocean, respectively. $C_{air/ocean}$ are
124 air and ocean drag coefficients, $\rho_{air/ocean}$ reference
125 densities, and $R_{air/ocean}$ rotation matrices that act on the
126 wind/current vectors. $\vek{F} = \nabla\cdot\sigma$ is the divergence
127 of the interal stress tensor $\sigma_{ij}$.
128
129 For an isotropic system this stress tensor can be related to the ice
130 strain rate and strength by a nonlinear viscous-plastic (VP)
131 constitutive law \citep{hibler79, zhang98}:
132 \begin{equation}
133 \label{eq:vpequation}
134 \sigma_{ij}=2\eta(\dot{\epsilon}_{ij},P)\dot{\epsilon}_{ij}
135 + \left[\zeta(\dot{\epsilon}_{ij},P) -
136 \eta(\dot{\epsilon}_{ij},P)\right]\dot{\epsilon}_{kk}\delta_{ij}
137 - \frac{P}{2}\delta_{ij}.
138 \end{equation}
139 The ice strain rate is given by
140 \begin{equation*}
141 \dot{\epsilon}_{ij} = \frac{1}{2}\left(
142 \frac{\partial{u_{i}}}{\partial{x_{j}}} +
143 \frac{\partial{u_{j}}}{\partial{x_{i}}}\right).
144 \end{equation*}
145 The maximum ice pressure $P_{\max}$, a measure of ice strength, depends on
146 both thickness $h$ and compactness (concentration) $c$:
147 \begin{equation}
148 P_{\max} = P^{*}c\,h\,e^{[C^{*}\cdot(1-c)]},
149 \label{eq:icestrength}
150 \end{equation}
151 with the constants $P^{*}$ and $C^{*}$. The nonlinear bulk and shear
152 viscosities $\eta$ and $\zeta$ are functions of ice strain rate
153 invariants and ice strength such that the principal components of the
154 stress lie on an elliptical yield curve with the ratio of major to
155 minor axis $e$ equal to $2$; they are given by:
156 \begin{align*}
157 \zeta =& \min\left(\frac{P_{\max}}{2\max(\Delta,\Delta_{\min})},
158 \zeta_{\max}\right) \\
159 \eta =& \frac{\zeta}{e^2} \\
160 \intertext{with the abbreviation}
161 \Delta = & \left[
162 \left(\dot{\epsilon}_{11}^2+\dot{\epsilon}_{22}^2\right)
163 (1+e^{-2}) + 4e^{-2}\dot{\epsilon}_{12}^2 +
164 2\dot{\epsilon}_{11}\dot{\epsilon}_{22} (1-e^{-2})
165 \right]^{-\frac{1}{2}}
166 \end{align*}
167 The bulk viscosities are bounded above by imposing both a minimum
168 $\Delta_{\min}=10^{-11}\text{\,s}^{-1}$ (for numerical reasons) and a
169 maximum $\zeta_{\max} = P_{\max}/\Delta^*$, where
170 $\Delta^*=(5\times10^{12}/2\times10^4)\text{\,s}^{-1}$. For stress
171 tensor compuation the replacement pressure $P = 2\,\Delta\zeta$
172 \citep{hibler95} is used so that the stress state always lies on the
173 elliptic yield curve by definition.
174
175 In the so-called truncated ellipse method the shear viscosity $\eta$
176 is capped to suppress any tensile stress \citep{hibler97, geiger98}:
177 \begin{equation}
178 \label{eq:etatem}
179 \eta = \min(\frac{\zeta}{e^2}
180 \frac{\frac{P}{2}-\zeta(\dot{\epsilon}_{11}+\dot{\epsilon}_{22})}
181 {\sqrt{(\dot{\epsilon}_{11}+\dot{\epsilon}_{22})^2
182 +4\dot{\epsilon}_{12}^2}}
183 \end{equation}
184
185 In the current implementation, the VP-model is integrated with the
186 semi-implicit line successive over relaxation (LSOR)-solver of
187 \citet{zhang98}, which allows for long time steps that, in our case,
188 is limited by the explicit treatment of the Coriolis term. The
189 explicit treatment of the Coriolis term does not represent a severe
190 limitation because it restricts the time step to approximately the
191 same length as in the ocean model where the Coriolis term is also
192 treated explicitly.
193
194 \citet{hunke97}'s introduced an elastic contribution to the strain
195 rate elatic-viscous-plastic in order to regularize
196 Eq.\refeq{vpequation} in such a way that the resulting
197 elatic-viscous-plastic (EVP) and VP models are identical at steady
198 state,
199 \begin{equation}
200 \label{eq:evpequation}
201 \frac{1}{E}\frac{\partial\sigma_{ij}}{\partial{t}} +
202 \frac{1}{2\eta}\sigma_{ij}
203 + \frac{\eta - \zeta}{4\zeta\eta}\sigma_{kk}\delta_{ij}
204 + \frac{P}{4\zeta}\delta_{ij}
205 = \dot{\epsilon}_{ij}.
206 \end{equation}
207 %In the EVP model, equations for the components of the stress tensor
208 %$\sigma_{ij}$ are solved explicitly. Both model formulations will be
209 %used and compared the present sea-ice model study.
210 The EVP-model uses an explicit time stepping scheme with a short
211 timestep. According to the recommendation of \citet{hunke97}, the
212 EVP-model is stepped forward in time 120 times within the physical
213 ocean model time step (although this parameter is under debate), to
214 allow for elastic waves to disappear. Because the scheme does not
215 require a matrix inversion it is fast in spite of the small timestep
216 \citep{hunke97}. For completeness, we repeat the equations for the
217 components of the stress tensor $\sigma_{1} =
218 \sigma_{11}+\sigma_{22}$, $\sigma_{2}= \sigma_{11}-\sigma_{22}$, and
219 $\sigma_{12}$. Introducing the divergence $D_D =
220 \dot{\epsilon}_{11}+\dot{\epsilon}_{22}$, and the horizontal tension
221 and shearing strain rates, $D_T =
222 \dot{\epsilon}_{11}-\dot{\epsilon}_{22}$ and $D_S =
223 2\dot{\epsilon}_{12}$, respectively and using the above abbreviations,
224 the equations can be written as:
225 \begin{align}
226 \label{eq:evpstresstensor1}
227 \frac{\partial\sigma_{1}}{\partial{t}} + \frac{\sigma_{1}}{2T} +
228 \frac{P}{2T} &= \frac{P}{2T\Delta} D_D \\
229 \label{eq:evpstresstensor2}
230 \frac{\partial\sigma_{2}}{\partial{t}} + \frac{\sigma_{2} e^{2}}{2T}
231 &= \frac{P}{2T\Delta} D_T \\
232 \label{eq:evpstresstensor12}
233 \frac{\partial\sigma_{12}}{\partial{t}} + \frac{\sigma_{12} e^{2}}{2T}
234 &= \frac{P}{4T\Delta} D_S
235 \end{align}
236 Here, the elastic parameter $E$ is redefined in terms of a damping timescale
237 $T$ for elastic waves \[E=\frac{\zeta}{T}.\]
238 $T=E_{0}\Delta{t}$ with the tunable parameter $E_0<1$ and
239 the external (long) timestep $\Delta{t}$. \citet{hunke97} recommend
240 $E_{0} = \frac{1}{3}$.
241
242 For details of the spatial discretization, the reader is referred to
243 \citet{zhang98, zhang03}. Our discretization differs only (but
244 importantly) in the underlying grid, namely the Arakawa C-grid, but is
245 otherwise straightforward. The EVP model in particular is discretized
246 naturally on the C-grid with $\sigma_{1}$ and $\sigma_{2}$ on the
247 center points and $\sigma_{12}$ on the corner (or vorticity) points of
248 the grid. With this choice all derivatives are discretized as central
249 differences and averaging is only involved in computing $\Delta$ and
250 $P$ at vorticity points.
251
252 For a general curvilinear grid, one needs in principle to take metric
253 terms into account that arise in the transformation a curvilinear grid
254 on the sphere. However, for now we can neglect these metric terms
255 because they are very small on the cubed sphere grids used in this
256 paper; in particular, only near the edges of the cubed sphere grid, we
257 expect them to be non-zero, but these edges are at approximately
258 35\degS\ or 35\degN\ which are never covered by sea-ice in our
259 simulations. Everywhere else the coordinate system is locally nearly
260 cartesian. However, for last-glacial-maximum or snowball-earth-like
261 simulations the question of metric terms needs to be reconsidered.
262 Either, one includes these terms as in \citet{zhang03}, or one finds a
263 vector-invariant formulation fo the sea-ice internal stress term that
264 does not require any metric terms, as it is done in the ocean dynamics
265 of the MITgcm \citep{adcroft04:_cubed_sphere}.
266
267 Moving sea ice exerts a stress on the ocean which is the opposite of
268 the stress $\vtau_{ocean}$ in Eq.\refeq{momseaice}. This stess is
269 applied directly to the surface layer of the ocean model. An
270 alternative ocean stress formulation is given by \citet{hibler87}.
271 Rather than applying $\vtau_{ocean}$ directly, the stress is derived
272 from integrating over the ice thickness to the bottom of the oceanic
273 surface layer. In the resulting equation for the \emph{combined}
274 ocean-ice momentum, the interfacial stress cancels and the total
275 stress appears as the sum of windstress and divergence of internal ice
276 stresses: $\delta(z) (\vtau_{air} + \vek{F})/\rho_0$, \citep[see also
277 Eq.\,2 of][]{hibler87}. The disadvantage of this formulation is that
278 now the velocity in the surface layer of the ocean that is used to
279 advect tracers, is really an average over the ocean surface
280 velocity and the ice velocity leading to an inconsistency as the ice
281 temperature and salinity are different from the oceanic variables.
282
283 Sea ice distributions are characterized by sharp gradients and edges.
284 For this reason, we employ a positive 3rd-order advection scheme
285 \citep{hundsdorfer94} for the thermodynamic variables discussed in the
286 next section.
287
288 \subparagraph{boundary conditions: no-slip, free-slip, half-slip}
289
290 \begin{itemize}
291 \item transition from existing B-Grid to C-Grid
292 \item boundary conditions: no-slip, free-slip, half-slip
293 \item fancy (multi dimensional) advection schemes
294 \item VP vs.\ EVP \citep{hunke97}
295 \item ocean stress formulation \citep{hibler87}
296 \end{itemize}
297
298 \subsection{Thermodynamics}
299 \label{sec:thermodynamics}
300
301 In the original formulation the sea ice model \citep{menemenlis05}
302 uses simple thermodynamics following the appendix of
303 \citet{semtner76}. This formulation does not allow storage of heat
304 (heat capacity of ice is zero, and this type of model is often refered
305 to as a ``zero-layer'' model). Upward heat flux is parameterized
306 assuming a linear temperature profile and together with a constant ice
307 conductivity. It is expressed as $(K/h)(T_{w}-T_{0})$, where $K$ is
308 the ice conductivity, $h$ the ice thickness, and $T_{w}-T_{0}$ the
309 difference between water and ice surface temperatures. The surface
310 heat budget is computed in a similar way to that of
311 \citet{parkinson79} and \citet{manabe79}.
312
313 There is considerable doubt about the reliability of such a simple
314 thermodynamic model---\citet{semtner84} found significant errors in
315 phase (one month lead) and amplitude ($\approx$50\%\,overestimate) in
316 such models---, so that today many sea ice models employ more complex
317 thermodynamics. A popular thermodynamics model of \citet{winton00} is
318 based on the 3-layer model of \citet{semtner76} and treats brine
319 content by means of enthalphy conservation. This model requires in
320 addition to ice-thickness and compactness (fractional area) additional
321 state variables to be advected by ice velocities, namely enthalphy of
322 the two ice layers and the thickness of the overlying snow layer.
323
324 \section{Funnel Experiments}
325 \label{sec:funnel}
326
327 For a first/detailed comparison between the different variants of the
328 MIT sea ice model an idealized geometry of a periodic channel,
329 1000\,km long and 500\,m wide on a non-rotating plane, with converging
330 walls forming a symmetric funnel and a narrow strait of 40\,km width
331 is used. The horizontal resolution is 5\,km throughout the domain
332 making the narrow strait 8 grid points wide. The ice model is
333 initialized with a complete ice cover of 50\,cm uniform thickness. The
334 ice model is driven by a constant along channel eastward ocean current
335 of 25\,cm/s that does not see the walls in the domain. All other
336 ice-ocean-atmosphere interactions are turned off, in particular there
337 is no feedback of ice dynamics on the ocean current. All thermodynamic
338 processes are turned off so that ice thickness variations are only
339 caused by convergent or divergent ice flow. Ice volume (effective
340 thickness) and concentration are advected with a third-order scheme
341 with a flux limiter \citep{hundsdorfer94} to avoid undershoots. This
342 scheme is unconditionally stable and does not require additional
343 diffusion. The time step used here is 1\,h.
344
345 \reffig{funnelf0} compares the dynamic fields ice concentration $c$,
346 effective thickness $h_{eff} = h\cdot{c}$, and velocities $(u,v)$ for
347 five different cases at steady state (after 10\,years of integration):
348 \begin{description}
349 \item[B-LSRns:] LSR solver with no-slip boundary conditions on a B-grid,
350 \item[C-LSRns:] LSR solver with no-slip boundary conditions on a C-grid,
351 \item[C-LSRfs:] LSR solver with free-slip boundary conditions on a C-grid,
352 \item[C-EVPns:] EVP solver with no-slip boundary conditions on a C-grid,
353 \item[C-EVPfs:] EVP solver with free-slip boundary conditions on a C-grid,
354 \end{description}
355 \ml{[We have not implemented the EVP solver on a B-grid.]}
356 \begin{figure*}[htbp]
357 \includegraphics[width=\widefigwidth]{\fpath/all_086280}
358 \caption{Ice concentration, effective thickness [m], and ice
359 velocities [m/s]
360 for 5 different numerical solutions.}
361 \label{fig:funnelf0}
362 \end{figure*}
363 At a first glance, the solutions look similar. This is encouraging as
364 the details of discretization and numerics should not affect the
365 solutions to first order. In all cases the ice-ocean stress pushes the
366 ice cover eastwards, where it converges in the funnel. In the narrow
367 channel the ice moves quickly (nearly free drift) and leaves the
368 channel as narrow band.
369
370 A close look reveals interesting differences between the B- and C-grid
371 results. The zonal velocity in the narrow channel is nearly the free
372 drift velocity ( = ocean velocity) of 25\,cm/s for the C-grid
373 solutions, regardless of the boundary conditions, while it is just
374 above 20\,cm/s for the B-grid solution. The ice accelerates to
375 25\,cm/s after it exits the channel. Concentrating on the solutions
376 B-LSRns and C-LSRns, the ice volume (effective thickness) along the
377 boundaries in the narrow channel is larger in the B-grid case although
378 the ice concentration is reduces in the C-grid case. The combined
379 effect leads to a larger actual ice thickness at smaller
380 concentrations in the C-grid case. However, since the effective
381 thickness determines the ice strength $P$ in Eq\refeq{icestrength},
382 the ice strength and thus the bulk and shear viscosities are larger in
383 the B-grid case leading to more horizontal friction. This circumstance
384 might explain why the no-slip boundary conditions in the B-grid case
385 appear to be more effective in reducing the flow within the narrow
386 channel, than in the C-grid case. Further, the viscosities are also
387 sensitive to details of the velocity gradients. Via $\Delta$, these
388 gradients enter the viscosities in the denominator so that large
389 gradients tend to reduce the viscosities. This again favors more flow
390 along the boundaries in the C-grid case: larger velocities
391 (\reffig{funnelf0}) on grid points that are closer to the boundary by
392 a factor $\frac{1}{2}$ than in the B-grid case because of the stagger
393 nature of the C-grid lead numerically to larger tangential gradients
394 across the boundary; these in turn make the viscosities smaller for
395 less tangential friction and allow more tangential flow along the
396 boundaries.
397
398 The above argument can also be invoked to explain the small
399 differences between the free-slip and no-slip solutions on the C-grid.
400 Because of the non-linearities in the ice viscosities, in particular
401 along the boundaries, the no-slip boundary conditions have only a small
402 impact on the solution.
403
404 The difference between LSR and EVP solutions is largest in the
405 effective thickness and meridional velocity fields. The EVP velocity
406 fields appears to be a little noisy. This noise has been address by
407 \citet{hunke01}. It can be dealt with by reducing EVP's internal time
408 step (increasing the number of iterations along with the computational
409 cost) or by regularizing the bulk and shear viscosities. We revisit
410 the latter option by reproducing some of the results of
411 \citet{hunke01}, namely the experiment described in her section~4, for
412 our C-grid no-slip cases: in a square domain with a few islands the
413 ice model is initialized with constant ice thickness and linearly
414 increasing ice concentration to the east. The model dynamics are
415 forced with a constant anticyclonic ocean gyre and by variable
416 atmospheric wind whose mean direction is diagnonal to the north-east
417 corner of the domain; ice volume and concentration are held constant
418 (no thermodynamics and no advection by ice velocity).
419 \reffig{hunke01} shows the ice velocity field, its divergence, and the
420 bulk viscosity $\zeta$ for the cases C-LRSns and C-EVPns, and for a
421 C-EVPns case, where \citet{hunke01}'s regularization has been
422 implemented; compare to Fig.\,4 in \citet{hunke01}. The regularization
423 contraint limits ice strength and viscosities as a function of damping
424 time scale, resolution and EVP-time step, effectively allowing the
425 elastic waves to damp out more quickly \citep{hunke01}.
426 \begin{figure*}[htbp]
427 \includegraphics[width=\widefigwidth]{\fpath/hun12days}
428 \caption{Ice flow, divergence and bulk viscosities of three
429 experiments with \citet{hunke01}'s test case: C-LSRns (top),
430 C-EVPns (middle), and C-EVPns with damping described in
431 \citet{hunke01} (bottom).}
432 \label{fig:hunke01}
433 \end{figure*}
434
435 In the far right (``east'') side of the domain the ice concentration
436 is close to one and the ice should be nearly rigid. The applied wind
437 tends to push ice toward the upper right corner. Because the highly
438 compact ice is confined by the boundary, it resists any further
439 compression and exhibits little motion in the rigid region on the
440 right hand side. The C-LSRns solution (top row) allows high
441 viscosities in the rigid region suppressing nearly all flow.
442 \citet{hunke01}'s regularization for the C-EVPns solution (bottom row)
443 clearly suppresses the noise present in $\nabla\cdot\vek{u}$ and
444 $\log_{10}\zeta$ in the
445 unregularized case (middle row), at the cost of reduced viscosities.
446 These reduced viscosities lead to small but finite ice velocities
447 which in turn can have a strong effect on solutions in the limit of
448 nearly rigid regimes (arching and blocking, not shown).
449
450 \ml{[Say something about performance? This is tricky, as the
451 perfomance depends strongly on the configuration. A run with slowly
452 changing forcing is favorable for LSR, because then only very few
453 iterations are required for convergences while EVP uses its fixed
454 number of internal timesteps. If the forcing in changing fast, LSR
455 needs far more iterations while EVP still uses the fixed number of
456 internal timesteps. I have produces runs where for slow forcing LSR
457 is much faster than EVP and for fast forcing, LSR is much slower
458 than EVP. EVP is certainly more efficient in terms of vectorization
459 and MFLOPS on our SX8, but is that a criterion?]}
460
461 \subsection{C-grid}
462 \begin{itemize}
463 \item no-slip vs. free-slip for both lsr and evp;
464 "diagnostics" to look at and use for comparison
465 \begin{itemize}
466 \item ice transport through Fram Strait/Denmark Strait/Davis
467 Strait/Bering strait (these are general)
468 \item ice transport through narrow passages, e.g.\ Nares-Strait
469 \end{itemize}
470 \item compare different advection schemes (if lsr turns out to be more
471 effective, then with lsr otherwise I prefer evp), eg.
472 \begin{itemize}
473 \item default 2nd-order with diff1=0.002
474 \item 1st-order upwind with diff1=0.
475 \item DST3FL (SEAICEadvScheme=33 with diff1=0., should work, works for me)
476 \item 2nd-order wit flux limiter (SEAICEadvScheme=77 with diff1=0.)
477 \end{itemize}
478 That should be enough. Here, total ice mass and location of ice edge
479 is interesting. However, this comparison can be done in an idealized
480 domain, may not require full Arctic Domain?
481 \item
482 Do a little study on the parameters of LSR and EVP
483 \begin{enumerate}
484 \item convergence of LSR, how many iterations do you need to get a
485 true elliptic yield curve
486 \item vary deltaTevp and the relaxation parameter for EVP and see when
487 the EVP solution breaks down (relative to the forcing time scale).
488 For this, it is essential that the evp solver gives use "stripeless"
489 solutions, that is your dtevp = 1sec solutions/or 10sec solutions
490 with SEAICE\_evpDampC = 615.
491 \end{enumerate}
492 \end{itemize}
493
494 \section{Forward sensitivity experiments}
495 \label{sec:forward}
496
497 A second series of forward sensitivity experiments have been carried out on an
498 Arctic Ocean domain with open boundaries. Once again the objective is to
499 compare the old B-grid LSR dynamic solver with the new C-grid LSR and EVP
500 solvers. One additional experiment is carried out to illustrate the
501 differences between the two main options for sea ice thermodynamics in the MITgcm.
502
503 \subsection{Arctic Domain with Open Boundaries}
504 \label{sec:arctic}
505
506 The Arctic domain of integration is illustrated in Fig.~\ref{???}. It
507 is carved out from, and obtains open boundary conditions from, the
508 global cubed-sphere configuration of the Estimating the Circulation
509 and Climate of the Ocean, Phase II (ECCO2) project
510 \citet{menemenlis05}. The domain size is 420 by 384 grid boxes
511 horizontally with mean horizontal grid spacing of 18 km.
512
513 There are 50 vertical levels ranging in thickness from 10 m near the surface
514 to approximately 450 m at a maximum model depth of 6150 m. Bathymetry is from
515 the National Geophysical Data Center (NGDC) 2-minute gridded global relief
516 data (ETOPO2) and the model employs the partial-cell formulation of
517 \citet{adcroft97:_shaved_cells}, which permits accurate representation of the bathymetry. The
518 model is integrated in a volume-conserving configuration using a finite volume
519 discretization with C-grid staggering of the prognostic variables. In the
520 ocean, the non-linear equation of state of \citet{jackett95}. The ocean model is
521 coupled to a sea-ice model described hereinabove.
522
523 This particular ECCO2 simulation is initialized from rest using the
524 January temperature and salinity distribution from the World Ocean
525 Atlas 2001 (WOA01) [Conkright et al., 2002] and it is integrated for
526 32 years prior to the 1996--2001 period discussed in the study. Surface
527 boundary conditions are from the National Centers for Environmental
528 Prediction and the National Center for Atmospheric Research
529 (NCEP/NCAR) atmospheric reanalysis [Kistler et al., 2001]. Six-hourly
530 surface winds, temperature, humidity, downward short- and long-wave
531 radiations, and precipitation are converted to heat, freshwater, and
532 wind stress fluxes using the \citet{large81, large82} bulk formulae.
533 Shortwave radiation decays exponentially as per Paulson and Simpson
534 [1977]. Additionally the time-mean river run-off from Large and Nurser
535 [2001] is applied and there is a relaxation to the monthly-mean
536 climatological sea surface salinity values from WOA01 with a
537 relaxation time scale of 3 months. Vertical mixing follows
538 \citet{large94} with background vertical diffusivity of
539 $1.5\times10^{-5}\text{\,m$^{2}$\,s$^{-1}$}$ and viscosity of
540 $10^{-3}\text{\,m$^{2}$\,s$^{-1}$}$. A third order, direct-space-time
541 advection scheme with flux limiter is employed \citep{hundsdorfer94}
542 and there is no explicit horizontal diffusivity. Horizontal viscosity
543 follows \citet{lei96} but
544 modified to sense the divergent flow as per Fox-Kemper and Menemenlis
545 [in press]. Shortwave radiation decays exponentially as per Paulson
546 and Simpson [1977]. Additionally, the time-mean runoff of Large and
547 Nurser [2001] is applied near the coastline and, where there is open
548 water, there is a relaxation to monthly-mean WOA01 sea surface
549 salinity with a time constant of 45 days.
550
551 Open water, dry
552 ice, wet ice, dry snow, and wet snow albedo are, respectively, 0.15, 0.85,
553 0.76, 0.94, and 0.8.
554
555 \begin{itemize}
556 \item Configuration
557 \item OBCS from cube
558 \item forcing
559 \item 1/2 and full resolution
560 \item with a few JFM figs from C-grid LSR no slip
561 ice transport through Canadian Archipelago
562 thickness distribution
563 ice velocity and transport
564 \end{itemize}
565
566 \begin{itemize}
567 \item Arctic configuration
568 \item ice transport through straits and near boundaries
569 \item focus on narrow straits in the Canadian Archipelago
570 \end{itemize}
571
572 \begin{itemize}
573 \item B-grid LSR no-slip
574 \item C-grid LSR no-slip
575 \item C-grid LSR slip
576 \item C-grid EVP no-slip
577 \item C-grid EVP slip
578 \item C-grid LSR + TEM (truncated ellipse method, no tensile stress, new flag)
579 \item C-grid LSR no-slip + Winton
580 \item speed-performance-accuracy (small)
581 ice transport through Canadian Archipelago differences
582 thickness distribution differences
583 ice velocity and transport differences
584 \end{itemize}
585
586 We anticipate small differences between the different models due to:
587 \begin{itemize}
588 \item advection schemes: along the ice-edge and regions with large
589 gradients
590 \item C-grid: less transport through narrow straits for no slip
591 conditons, more for free slip
592 \item VP vs.\ EVP: speed performance, accuracy?
593 \item ocean stress: different water mass properties beneath the ice
594 \end{itemize}
595
596 \section{Adjoint sensiivities of the MITsim}
597
598 \subsection{The adjoint of MITsim}
599
600 The ability to generate tangent linear and adjoint model components
601 of the MITsim has been a main design task.
602 For the ocean the adjoint capability has proven to be an
603 invaluable tool for sensitivity analysis as well as state estimation.
604 In short, the adjoint enables very efficient computation of the gradient
605 of scalar-valued model diagnostics (called cost function or objective function)
606 with respect to many model "variables".
607 These variables can be two- or three-dimensional fields of initial
608 conditions, model parameters such as mixing coefficients, or
609 time-varying surface or lateral (open) boundary conditions.
610 When combined, these variables span a potentially high-dimensional
611 (e.g. O(10$^8$)) so-called control space. Performing parameter perturbations
612 to assess model sensitivities quickly becomes prohibitive at these scales.
613 Alternatively, (time-varying) sensitivities of the objective function
614 to any element of the control space can be computed very efficiently in
615 one single adjoint
616 model integration, provided an efficient adjoint model is available.
617
618 [REFERENCES]
619
620
621 The adjoint operator (ADM) is the transpose of the tangent linear operator (TLM)
622 of the full (in general nonlinear) forward model, i.e. the MITsim.
623 The TLM maps perturbations of elements of the control space
624 (e.g. initial ice thickness distribution)
625 via the model Jacobian
626 to a perturbation in the objective function
627 (e.g. sea-ice export at the end of the integration interval).
628 \textit{Tangent} linearity ensures that the derivatives are evaluated
629 with respect to the underlying model trajectory at each point in time.
630 This is crucial for nonlinear trajectories and the presence of different
631 regimes (e.g. effect of the seaice growth term at or away from the
632 freezing point of the ocean surface).
633 Ensuring tangent linearity can be easily achieved by integrating
634 the full model in sync with the TLM to provide the underlying model state.
635 Ensuring \textit{tangent} adjoints is equally crucial, but much more
636 difficult to achieve because of the reverse nature of the integration:
637 the adjoint accumulates sensitivities backward in time,
638 starting from a unit perturbation of the objective function.
639 The adjoint model requires the model state in reverse order.
640 This presents one of the major complications in deriving an
641 exact, i.e. \textit{tangent} adjoint model.
642
643 Following closely the development and maintenance of TLM and ADM
644 components of the MITgcm we have relied heavily on the
645 autmomatic differentiation (AD) tool
646 "Transformation of Algorithms in Fortran" (TAF)
647 developed by Fastopt (Giering and Kaminski, 1998)
648 to derive TLM and ADM code of the MITsim.
649 Briefly, the nonlinear parent model is fed to the AD tool which produces
650 derivative code for the specified control space and objective function.
651 Following this approach has (apart from its evident success)
652 several advantages:
653 (1) the adjoint model is the exact adjoint operator of the parent model,
654 (2) the adjoint model can be kept up to date with respect to ongoing
655 development of the parent model, and adjustments to the parent model
656 to extend the automatically generated adjoint are incremental changes
657 only, rather than extensive re-developments,
658 (3) the parallel structure of the parent model is preserved
659 by the adjoint model, ensuring efficient use in high performance
660 computing environments.
661
662 Some initial code adjustments are required to support dependency analysis
663 of the flow reversal and certain language limitations which may lead
664 to irreducible flow graphs (e.g. GOTO statements).
665 The problem of providing the required model state in reverse order
666 at the time of evaluating nonlinear or conditional
667 derivatives is solved via balancing
668 storing vs. recomputation of the model state in a multi-level
669 checkpointing loop.
670 Again, an initial code adjustment is required to support TAFs
671 checkpointing capability.
672 The code adjustments are sufficiently simple so as not to cause
673 major limitations to the full nonlinear parent model.
674 Once in place, an adjoint model of a new model configuration
675 may be derived in about 10 minutes.
676
677 [HIGHLIGHT COUPLED NATURE OF THE ADJOINT!]
678
679 \subsection{Special considerations}
680
681 * growth term(?)
682
683 * small active denominators
684
685 * dynamic solver (implicit function theorem)
686
687 * approximate adjoints
688
689
690 \subsection{An example: sensitivities of sea-ice export through Fram Strait}
691
692 We demonstrate the power of the adjoint method
693 in the context of investigating sea-ice export sensitivities through Fram Strait
694 (for details of this study see Heimbach et al., 2007).
695 %\citep[for details of this study see][]{heimbach07}. %Heimbach et al., 2007).
696 The domain chosen is a coarsened version of the Arctic face of the
697 high-resolution cubed-sphere configuration of the ECCO2 project
698 \citep[see][]{menemenlis05}. It covers the entire Arctic,
699 extends into the North Pacific such as to cover the entire
700 ice-covered regions, and comprises parts of the North Atlantic
701 down to XXN to enable analysis of remote influences of the
702 North Atlantic current to sea-ice variability and export.
703 The horizontal resolution varies between XX and YY km
704 with 50 unevenly spaced vertical levels.
705 The adjoint models run efficiently on 80 processors
706 (benchmarks have been performed both on an SGI Altix as well as an
707 IBM SP5 at NASA/ARC).
708
709 Following a 1-year spinup, the model has been integrated for four
710 years between 1992 and 1995. It is forced using realistic 6-hourly
711 NCEP/NCAR atmospheric state variables. Over the open ocean these are
712 converted into air-sea fluxes via the bulk formulae of
713 \citet{large04}. Derivation of air-sea fluxes in the presence of
714 sea-ice is handled by the ice model as described in \refsec{model}.
715 The objective function chosen is sea-ice export through Fram Strait
716 computed for December 1995. The adjoint model computes sensitivities
717 to sea-ice export back in time from 1995 to 1992 along this
718 trajectory. In principle all adjoint model variable (i.e., Lagrange
719 multipliers) of the coupled ocean/sea-ice model are available to
720 analyze the transient sensitivity behaviour of the ocean and sea-ice
721 state. Over the open ocean, the adjoint of the bulk formula scheme
722 computes sensitivities to the time-varying atmospheric state. Over
723 ice-covered parts, the sea-ice adjoint converts surface ocean
724 sensitivities to atmospheric sensitivities.
725
726 \reffig{4yradjheff}(a--d) depict sensitivities of sea-ice export
727 through Fram Strait in December 1995 to changes in sea-ice thickness
728 12, 24, 36, 48 months back in time. Corresponding sensitivities to
729 ocean surface temperature are depicted in
730 \reffig{4yradjthetalev1}(a--d). The main characteristics is
731 consistency with expected advection of sea-ice over the relevant time
732 scales considered. The general positive pattern means that an
733 increase in sea-ice thickness at location $(x,y)$ and time $t$ will
734 increase sea-ice export through Fram Strait at time $T_e$. Largest
735 distances from Fram Strait indicate fastest sea-ice advection over the
736 time span considered. The ice thickness sensitivities are in close
737 correspondence to ocean surface sentivitites, but of opposite sign.
738 An increase in temperature will incur ice melting, decrease in ice
739 thickness, and therefore decrease in sea-ice export at time $T_e$.
740
741 The picture is fundamentally different and much more complex
742 for sensitivities to ocean temperatures away from the surface.
743 \reffig{4yradjthetalev10??}(a--d) depicts ice export sensitivities to
744 temperatures at roughly 400 m depth.
745 Primary features are the effect of the heat transport of the North
746 Atlantic current which feeds into the West Spitsbergen current,
747 the circulation around Svalbard, and ...
748
749 \begin{figure}[t!]
750 \centerline{
751 \subfigure[{\footnotesize -12 months}]
752 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim072_cmax2.0E+02.eps}}
753 %\includegraphics*[width=.3\textwidth]{H_c.bin_res_100_lev1.pdf}
754 %
755 \subfigure[{\footnotesize -24 months}]
756 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim145_cmax2.0E+02.eps}}
757 }
758
759 \centerline{
760 \subfigure[{\footnotesize
761 -36 months}]
762 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim218_cmax2.0E+02.eps}}
763 %
764 \subfigure[{\footnotesize
765 -48 months}]
766 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJheff_arc_lev1_tim292_cmax2.0E+02.eps}}
767 }
768 \caption{Sensitivity of sea-ice export through Fram Strait in December 2005 to
769 sea-ice thickness at various prior times.
770 \label{fig:4yradjheff}}
771 \end{figure}
772
773
774 \begin{figure}[t!]
775 \centerline{
776 \subfigure[{\footnotesize -12 months}]
777 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim072_cmax5.0E+01.eps}}
778 %\includegraphics*[width=.3\textwidth]{H_c.bin_res_100_lev1.pdf}
779 %
780 \subfigure[{\footnotesize -24 months}]
781 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim145_cmax5.0E+01.eps}}
782 }
783
784 \centerline{
785 \subfigure[{\footnotesize
786 -36 months}]
787 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim218_cmax5.0E+01.eps}}
788 %
789 \subfigure[{\footnotesize
790 -48 months}]
791 {\includegraphics*[width=0.44\linewidth]{\fpath/run_4yr_ADJtheta_arc_lev1_tim292_cmax5.0E+01.eps}}
792 }
793 \caption{Same as \reffig{4yradjheff} but for sea surface temperature
794 \label{fig:4yradjthetalev1}}
795 \end{figure}
796
797
798
799 \section{Discussion and conclusion}
800 \label{sec:concl}
801
802 The story of the paper could be:
803 B-grid ice model + C-grid ocean model does not work properly for us,
804 therefore C-grid ice model with advantages:
805 \begin{enumerate}
806 \item stress coupling simpler (no interpolation required)
807 \item different boundary conditions
808 \item advection schemes carry over trivially from main code
809 \end{enumerate}
810 LSR/EVP solutions are similar with appropriate bcs, evp parameters as
811 a function of forcing time scale (when does VP solution break
812 down). Same for LSR solver, provided that it works (o:
813 Which scheme is more efficient for the resolution/time stepping
814 parameters that we use here. What about other resolutions?
815
816 \paragraph{Acknowledgements}
817 We thank Jinlun Zhang for providing the original B-grid code and many
818 helpful discussions. ML thanks Elizabeth Hunke for multiple explanations.
819
820 \bibliography{bib/journal_abrvs,bib/seaice,bib/genocean,bib/maths,bib/mitgcmuv,bib/fram}
821
822 \end{document}
823
824 %%% Local Variables:
825 %%% mode: latex
826 %%% TeX-master: t
827 %%% End:
828
829
830 A Dynamic-Thermodynamic Sea ice Model for Ocean Climate
831 Estimation on an Arakawa C-Grid
832
833 Introduction
834
835 Ice Model:
836 Dynamics formulation.
837 B-C, LSR, EVP, no-slip, slip
838 parallellization
839 Thermodynamics formulation.
840 0-layer Hibler salinity + snow
841 3-layer Winton
842
843 Idealized tests
844 Funnel Experiments
845 Downstream Island tests
846 B-grid LSR no-slip
847 C-grid LSR no-slip
848 C-grid LSR slip
849 C-grid EVP no-slip
850 C-grid EVP slip
851
852 Arctic Setup
853 Configuration
854 OBCS from cube
855 forcing
856 1/2 and full resolution
857 with a few JFM figs from C-grid LSR no slip
858 ice transport through Canadian Archipelago
859 thickness distribution
860 ice velocity and transport
861
862 Arctic forward sensitivity experiments
863 B-grid LSR no-slip
864 C-grid LSR no-slip
865 C-grid LSR slip
866 C-grid EVP no-slip
867 C-grid EVP slip
868 C-grid LSR no-slip + Winton
869 speed-performance-accuracy (small)
870 ice transport through Canadian Archipelago differences
871 thickness distribution differences
872 ice velocity and transport differences
873
874 Adjoint sensitivity experiment on 1/2-res setup
875 Sensitivity of sea ice volume flow through Fram Strait
876 *** Sensitivity of sea ice volume flow through Canadian Archipelago
877
878 Summary and conluding remarks

  ViewVC Help
Powered by ViewVC 1.1.22