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

Annotation of /MITgcm_contrib/articles/ceaice/ceaice.tex

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


Revision 1.3 - (hide annotations) (download) (as text)
Wed Nov 7 17:26:13 2007 UTC (17 years, 8 months ago) by dimitri
Branch: MAIN
Changes since 1.2: +0 -6 lines
File MIME type: application/x-tex
my first mod to ceaice.tex - don't hold your breath guys!

1 dimitri 1.1 \documentclass[12pt]{article}
2     \usepackage{epsfig}
3     \usepackage{graphics}
4     \usepackage{subfigure}
5    
6     \usepackage[round,comma]{natbib}
7 dimitri 1.2 \bibliographystyle{bib/agu04}
8 dimitri 1.1
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}{.}
39    
40     \title{A Dynamic-Thermodynamic Sea ice Model for Ocean Climate
41     Estimation on an Arakawa C-Grid}
42    
43     \author{Martin Losch, Dimitris Menemenlis, Patrick Heimbach, \\
44     Jean-Michel Campin, and Chris Hill}
45     \begin{document}
46    
47     \maketitle
48    
49     \begin{abstract}
50     Some blabla
51     \end{abstract}
52    
53     \section{Introduction}
54     \label{sec:intro}
55    
56     more blabla
57    
58     \section{Model}
59     \label{sec:model}
60    
61     Traditionally, probably for historical reasons and the ease of
62     treating the Coriolis term, most standard sea-ice models are
63     discretized on Arakawa-B-grids \citep[e.g.,][]{hibler79, harder99,
64     kreyscher00, zhang98, hunke97}. From the perspective of coupling a
65     sea ice-model to a C-grid ocean model, the exchange of fluxes of heat
66     and fresh-water pose no difficulty for a B-grid sea-ice model
67     \citep[e.g.,][]{timmermann02a}. However, surface stress is defined at
68     velocities points and thus needs to be interpolated between a B-grid
69     sea-ice model and a C-grid ocean model. While the smoothing implicitly
70     associated with this interpolation may mask grid scale noise, it may
71     in two-way coupling lead to a computational mode as will be shown. By
72     choosing a C-grid for the sea-ice model, we circumvene this difficulty
73     altogether and render the stress coupling as consistent as the
74     buoyancy coupling.
75    
76     A further advantage of the C-grid formulation is apparent in narrow
77     straits. In the limit of only one grid cell between coasts there is no
78     flux allowed for a B-grid (with no-slip lateral boundary counditions),
79     whereas the C-grid formulation allows a flux of sea-ice through this
80     passage for all types of lateral boundary conditions. We (will)
81     demonstrate this effect in the Candian archipelago.
82    
83     \subsection{Dynamics}
84     \label{sec:dynamics}
85    
86     The momentum equations of the sea-ice model are standard with
87     \begin{equation}
88     \label{eq:momseaice}
89     m \frac{D\vek{u}}{Dt} = -mf\vek{k}\times\vek{u} + \vtau_{air} +
90     \vtau_{ocean} - m \nabla{\phi(0)} + \vek{F},
91     \end{equation}
92     where $\vek{u} = u\vek{i}+v\vek{j}$ is the ice velocity vectory, $m$
93     the ice mass per unit area, $f$ the Coriolis parameter, $g$ is the
94     gravity accelation, $\nabla\phi$ is the gradient (tilt) of the sea
95     surface height potential beneath the ice. $\phi$ is the sum of
96     atmpheric pressure $p_{a}$ and loading due to ice and snow
97     $(m_{i}+m_{s})g$. $\vtau_{air}$ and $\vtau_{ocean}$ are the wind and
98     ice-ocean stresses, respectively. $\vek{F}$ is the interaction force
99     and $\vek{i}$, $\vek{j}$, and $\vek{k}$ are the unit vectors in the
100     $x$, $y$, and $z$ directions. Advection of sea-ice momentum is
101     neglected. The wind and ice-ocean stress terms are given by
102     \begin{align*}
103     \vtau_{air} =& \rho_{air} |\vek{U}_{air}|R_{air}(\vek{U}_{air}) \\
104     \vtau_{ocean} =& \rho_{ocean} |\vek{U}_{ocean}-\vek{u}|
105     R_{ocean}(\vek{U}_{ocean}-\vek{u}), \\
106     \end{align*}
107     where $\vek{U}_{air/ocean}$ are the surface winds of the atmosphere
108     and surface currents of the ocean, respectively. $C_{air/ocean}$ are
109     air and ocean drag coefficients, $\rho_{air/ocean}$ reference
110     densities, and $R_{air/ocean}$ rotation matrices that act on the
111     wind/current vectors. $\vek{F} = \nabla\cdot\sigma$ is the divergence
112     of the interal stress tensor $\sigma_{ij}$.
113    
114     For an isotropic system this stress tensor can be related to the ice
115     strain rate and strength by a nonlinear viscous-plastic (VP)
116     constitutive law \citep{hibler79, zhang98}:
117     \begin{equation}
118     \label{eq:vpequation}
119     \sigma_{ij}=2\eta(\dot{\epsilon}_{ij},P)\dot{\epsilon}_{ij}
120     + \left[\zeta(\dot{\epsilon}_{ij},P) -
121     \eta(\dot{\epsilon}_{ij},P)\right]\dot{\epsilon}_{kk}\delta_{ij}
122     - \frac{P}{2}\delta_{ij}.
123     \end{equation}
124     The ice strain rate is given by
125     \begin{equation*}
126     \dot{\epsilon}_{ij} = \frac{1}{2}\left(
127     \frac{\partial{u_{i}}}{\partial{x_{j}}} +
128     \frac{\partial{u_{j}}}{\partial{x_{i}}}\right).
129     \end{equation*}
130     The pressure $P$, a measure of ice strength, depends on both thickness
131     $h$ and compactness (concentration) $c$: \[P =
132     P^{*}c\,h\,e^{[C^{*}\cdot(1-c)]},\] with the constants $P^{*}$ and
133     $C^{*}$. The nonlinear bulk and shear viscosities $\eta$ and $\zeta$
134     are functions of ice strain rate invariants and ice strength such that
135     the principal components of the stress lie on an elliptical yield
136     curve with the ratio of major to minor axis $e$ equal to $2$; they are
137     given by:
138     \begin{align*}
139     \zeta =& \frac{P}{2\Delta} \\
140     \eta =& \frac{P}{2\Delta{e}^2} \\
141     \intertext{with the abbreviation}
142     \Delta = & \left[
143     \left(\dot{\epsilon}_{11}^2+\dot{\epsilon}_{22}^2\right)
144     (1+e^{-2}) + 4e^{-2}\dot{\epsilon}_{12}^2 +
145     2\dot{\epsilon}_{11}\dot{\epsilon}_{22} (1-e^{-2})
146     \right]^{-\frac{1}{2}}
147     \end{align*}
148     In the current implementation, the VP-model is integrated with the
149     semi-implicit line successive over relaxation (LSOR)-solver of
150     \citet{zhang98}, which allows for long time steps that, in our case,
151     is limited by the explicit treatment of the Coriolis term. The
152     explicit treatment of the Coriolis term does not represent a severe
153     limitation because it restricts the time step to approximately the
154     same length as in the ocean model where the Coriolis term is also
155     treated explicitly.
156    
157     \citet{hunke97}'s introduced an elastic contribution to the strain
158     rate elatic-viscous-plastic in order to regularize
159     Eq.\refeq{vpequation} in such a way that the resulting
160     elatic-viscous-plastic (EVP) and VP models are identical at steady
161     state,
162     \begin{equation}
163     \label{eq:evpequation}
164     \frac{1}{E}\frac{\partial\sigma_{ij}}{\partial{t}} +
165     \frac{1}{2\eta}\sigma_{ij}
166     + \frac{\eta - \zeta}{4\zeta\eta}\sigma_{kk}\delta_{ij}
167     + \frac{P}{4\zeta}\delta_{ij}
168     = \dot{\epsilon}_{ij}.
169     \end{equation}
170     %In the EVP model, equations for the components of the stress tensor
171     %$\sigma_{ij}$ are solved explicitly. Both model formulations will be
172     %used and compared the present sea-ice model study.
173     The EVP-model uses an explicit time stepping scheme with a short
174     timestep. According to the recommendation of \citet{hunke97}, the
175     EVP-model is stepped forward in time 120 times within the physical
176     ocean model time step (although this parameter is under debate), to
177     allow for elastic waves to disappear. Because the scheme does not
178     require a matrix inversion it is fast in spite of the small timestep
179     \citep{hunke97}. For completeness, we repeat the equations for the
180     components of the stress tensor $\sigma_{1} =
181     \sigma_{11}+\sigma_{22}$, $\sigma_{2}= \sigma_{11}-\sigma_{22}$, and
182     $\sigma_{12}$. Introducing the divergence $D_D =
183     \dot{\epsilon}_{11}+\dot{\epsilon}_{22}$, and the horizontal tension
184     and shearing strain rates, $D_T =
185     \dot{\epsilon}_{11}-\dot{\epsilon}_{22}$ and $D_S =
186     2\dot{\epsilon}_{12}$, respectively and using the above abbreviations,
187     the equations can be written as:
188     \begin{align}
189     \label{eq:evpstresstensor1}
190     \frac{\partial\sigma_{1}}{\partial{t}} + \frac{\sigma_{1}}{2T} +
191     \frac{P}{2T} &= \frac{P}{2T\Delta} D_D \\
192     \label{eq:evpstresstensor2}
193     \frac{\partial\sigma_{2}}{\partial{t}} + \frac{\sigma_{2} e^{2}}{2T}
194     &= \frac{P}{2T\Delta} D_T \\
195     \label{eq:evpstresstensor12}
196     \frac{\partial\sigma_{12}}{\partial{t}} + \frac{\sigma_{12} e^{2}}{2T}
197     &= \frac{P}{4T\Delta} D_S
198     \end{align}
199     Here, the elastic parameter $E$ is redefined in terms of a damping timescale
200     $T$ for elastic waves \[E=\frac{\zeta}{T}.\]
201     $T=E_{0}\Delta{t}$ with the tunable parameter $E_0<1$ and
202     the external (long) timestep $\Delta{t}$. \citet{hunke97} recommend
203     $E_{0} = \frac{1}{3}$.
204    
205     For details of the spatial discretization, the reader is referred to
206     \citet{zhang98, zhang03}. Our discretization differs only (but
207     importantly) in the underlying grid, namely the Arakawa C-grid, but is
208     otherwise straightforward. The EVP model in particular is discretized
209     naturally on the C-grid with $\sigma_{1}$ and $\sigma_{2}$ on the
210     center points and $\sigma_{12}$ on the corner (or vorticity) points of
211     the grid. With this choice all derivatives are discretized as central
212     differences and averaging is only involved in computing $\Delta$ and
213     $P$ at vorticity points.
214    
215     For a general curvilinear grid, one needs in principle to take metric
216     terms into account that arise in the transformation a curvilinear grid
217     on the sphere. However, for now we can neglect these metric terms
218     because they are very small on the cubed sphere grids used in this
219     paper; in particular, only near the edges of the cubed sphere grid, we
220     expect them to be non-zero, but these edges are at approximately
221     35\degS\ or 35\degN\ which are never covered by sea-ice in our
222     simulations. Everywhere else the coordinate system is locally nearly
223     cartesian. However, for last-glacial-maximum or snowball-earth-like
224     simulations the question of metric terms needs to be reconsidered.
225     Either, one includes these terms as in \citet{zhang03}, or one finds a
226     vector-invariant formulation fo the sea-ice internal stress term that
227     does not require any metric terms, as it is done in the ocean dynamics
228     of the MITgcm \citep{adcroft04:_cubed_sphere}.
229    
230     Moving sea ice exerts a stress on the ocean which is the opposite of
231     the stress $\vtau_{ocean}$ in Eq.\refeq{momseaice}. This stess is
232     applied directly to the surface layer of the ocean model. An
233     alternative ocean stress formulation is given by \citet{hibler87}.
234     Rather than applying $\vtau_{ocean}$ directly, the stress is derived
235     from integrating over the ice thickness to the bottom of the oceanic
236     surface layer. In the resulting equation for the \emph{combined}
237     ocean-ice momentum, the interfacial stress cancels and the total
238     stress appears as the sum of windstress and divergence of internal ice
239     stresses: $\delta(z) (\vtau_{air} + \vek{F})/\rho_0$, \citep[see also
240     Eq.\,2 of][]{hibler87}. The disadvantage of this formulation is that
241     now the velocity in the surface layer of the ocean that is used to
242     advect tracers, is really an average over the ocean surface
243     velocity and the ice velocity leading to an inconsistency as the ice
244     temperature and salinity are different from the oceanic variables.
245    
246     Sea ice distributions are characterized by sharp gradients and edges.
247     For this reason, we employ a positive 3rd-order advection scheme
248     \citep{hundsdorfer94} for the thermodynamic variables discussed in the
249     next section.
250    
251     \subparagraph{boundary conditions: no-slip, free-slip, half-slip}
252    
253     \begin{itemize}
254     \item transition from existing B-Grid to C-Grid
255     \item boundary conditions: no-slip, free-slip, half-slip
256     \item fancy (multi dimensional) advection schemes
257     \item VP vs.\ EVP \citep{hunke97}
258     \item ocean stress formulation \citep{hibler87}
259     \end{itemize}
260    
261     \subsection{Thermodynamics}
262     \label{sec:thermodynamics}
263    
264     In the original formulation the sea ice model \citep{menemenlis05}
265     uses simple thermodynamics following the appendix of
266     \citet{semtner76}. This formulation does not allow storage of heat
267     (heat capacity of ice is zero, and this type of model is often refered
268     to as a ``zero-layer'' model). Upward heat flux is parameterized
269     assuming a linear temperature profile and together with a constant ice
270     conductivity. It is expressed as $(K/h)(T_{w}-T_{0})$, where $K$ is
271     the ice conductivity, $h$ the ice thickness, and $T_{w}-T_{0}$ the
272     difference between water and ice surface temperatures. The surface
273     heat budget is computed in a similar way to that of
274     \citet{parkinson79} and \citet{manabe79}.
275    
276     There is considerable doubt about the reliability of such a simple
277     thermodynamic model---\citet{semtner84} found significant errors in
278     phase (one month lead) and amplitude ($\approx$50\%\,overestimate) in
279     such models---, so that today many sea ice models employ more complex
280     thermodynamics. A popular thermodynamics model of \citet{winton00} is
281     based on the 3-layer model of \citet{semtner76} and treats brine
282     content by means of enthalphy conservation. This model requires in
283     addition to ice-thickness and compactness (fractional area) additional
284     state variables to be advected by ice velocities, namely enthalphy of
285     the two ice layers and the thickness of the overlying snow layer.
286    
287     \section{Funnel Experiments}
288     \label{sec:funnel}
289    
290     \begin{itemize}
291     \item B-grid LSR no-slip
292     \item C-grid LSR no-slip
293     \item C-grid LSR slip
294     \item C-grid EVP no-slip
295     \item C-grid EVP slip
296     \end{itemize}
297    
298     \subsection{B-grid vs.\ C-grid}
299     Comparison between:
300     \begin{itemize}
301     \item B-grid, lsr, no-slip
302     \item C-grid, lsr, no-slip
303     \item C-grid, evp, no-slip
304     \end{itemize}
305     all without ice-ocean stress, because ice-ocean stress does not work
306     for B-grid.
307    
308     \subsection{C-grid}
309     \begin{itemize}
310     \item no-slip vs. free-slip for both lsr and evp;
311     "diagnostics" to look at and use for comparison
312     \begin{itemize}
313     \item ice transport through Fram Strait/Denmark Strait/Davis
314     Strait/Bering strait (these are general)
315     \item ice transport through narrow passages, e.g.\ Nares-Strait
316     \end{itemize}
317     \item compare different advection schemes (if lsr turns out to be more
318     effective, then with lsr otherwise I prefer evp), eg.
319     \begin{itemize}
320     \item default 2nd-order with diff1=0.002
321     \item 1st-order upwind with diff1=0.
322     \item DST3FL (SEAICEadvScheme=33 with diff1=0., should work, works for me)
323     \item 2nd-order wit flux limiter (SEAICEadvScheme=77 with diff1=0.)
324     \end{itemize}
325     That should be enough. Here, total ice mass and location of ice edge
326     is interesting. However, this comparison can be done in an idealized
327     domain, may not require full Arctic Domain?
328     \item
329     Do a little study on the parameters of LSR and EVP
330     \begin{enumerate}
331     \item convergence of LSR, how many iterations do you need to get a
332     true elliptic yield curve
333     \item vary deltaTevp and the relaxation parameter for EVP and see when
334     the EVP solution breaks down (relative to the forcing time scale).
335     For this, it is essential that the evp solver gives use "stripeless"
336     solutions, that is your dtevp = 1sec solutions/or 10sec solutions
337     with SEAICE\_evpDampC = 615.
338     \end{enumerate}
339     \end{itemize}
340    
341     \section{Forward sensitivity experiments}
342     \label{sec:forward}
343    
344     A second series of forward sensitivity experiments have been carried out on an
345     Arctic Ocean domain with open boundaries. Once again the objective is to
346     compare the old B-grid LSR dynamic solver with the new C-grid LSR and EVP
347     solvers. One additional experiment is carried out to illustrate the
348     differences between the two main options for sea ice thermodynamics in the MITgcm.
349    
350     \subsection{Arctic Domain with Open Boundaries}
351     \label{sec:arctic}
352    
353     The Arctic domain of integration is illustrated in Fig.~\ref{???}. It is
354     carved out from, and obtains open boundary conditions from, the global
355     cubed-sphere configuration of the Estimating the Circulation and Climate of
356     the Ocean, Phase II (ECCO2) project \cite{men05a}. The domain size is 420 by
357     384 grid boxes horizontally with mean horizontal grid spacing of 18 km.
358    
359     There are 50 vertical levels ranging in thickness from 10 m near the surface
360     to approximately 450 m at a maximum model depth of 6150 m. Bathymetry is from
361     the National Geophysical Data Center (NGDC) 2-minute gridded global relief
362     data (ETOPO2) and the model employs the partial-cell formulation of
363     \cite{adc97}, which permits accurate representation of the bathymetry. The
364     model is integrated in a volume-conserving configuration using a finite volume
365     discretization with C-grid staggering of the prognostic variables. In the
366     ocean, the non-linear equation of state of \cite{jac95}. The ocean model is
367     coupled to a sea-ice model described hereinabove.
368    
369     This particular ECCO2 simulation is initialized from rest using the January
370     temperature and salinity distribution from the World Ocean Atlas 2001 (WOA01)
371     [Conkright et al., 2002] and it is integrated for 32 years prior to the
372     1996-2001 period discussed in the study. Surface boundary conditions are from
373     the National Centers for Environmental Prediction and the National Center for
374     Atmospheric Research (NCEP/NCAR) atmospheric reanalysis [Kistler et al.,
375     2001]. Six-hourly surface winds, temperature, humidity, downward short- and
376     long-wave radiations, and precipitation are converted to heat, freshwater, and
377     wind stress fluxes using the Large and Pond [1981, 1982] bulk
378     formulae. Shortwave radiation decays exponentially as per Paulson and Simpson
379     [1977]. Additionally the time-mean river run-off from Large and Nurser [2001]
380     is applied and there is a relaxation to the monthly-mean climatological sea
381     surface salinity values from WOA01 with a relaxation time scale of 3
382     months. Vertical mixing follows Large et al. [1994] with background vertical
383     diffusivity of 1.5 × 10-5 m2 s-1 and viscosity of 10-3 m2 s-1. A third order,
384     direct-space-time advection scheme with flux limiter is employed and there is
385     no explicit horizontal diffusivity. Horizontal viscosity follows Leith [1996]
386     but modified to sense the divergent flow as per Fox-Kemper and Menemenlis [in
387     press]. Shortwave radiation decays exponentially as per Paulson and Simpson
388     [1977]. Additionally, the time-mean runoff of Large and Nurser [2001] is
389     applied near the coastline and, where there is open water, there is a
390     relaxation to monthly-mean WOA01 sea surface salinity with a time constant of
391     45 days.
392    
393     Open water, dry
394     ice, wet ice, dry snow, and wet snow albedo are, respectively, 0.15, 0.85,
395     0.76, 0.94, and 0.8.
396    
397     \begin{itemize}
398     \item Configuration
399     \item OBCS from cube
400     \item forcing
401     \item 1/2 and full resolution
402     \item with a few JFM figs from C-grid LSR no slip
403     ice transport through Canadian Archipelago
404     thickness distribution
405     ice velocity and transport
406     \end{itemize}
407    
408     \begin{itemize}
409     \item Arctic configuration
410     \item ice transport through straits and near boundaries
411     \item focus on narrow straits in the Canadian Archipelago
412     \end{itemize}
413    
414     \begin{itemize}
415     \item B-grid LSR no-slip
416     \item C-grid LSR no-slip
417     \item C-grid LSR slip
418     \item C-grid EVP no-slip
419     \item C-grid EVP slip
420     \item C-grid LSR no-slip + Winton
421     \item speed-performance-accuracy (small)
422     ice transport through Canadian Archipelago differences
423     thickness distribution differences
424     ice velocity and transport differences
425     \end{itemize}
426    
427     We anticipate small differences between the different models due to:
428     \begin{itemize}
429     \item advection schemes: along the ice-edge and regions with large
430     gradients
431     \item C-grid: more transport through narrow straits for no slip
432     conditons, less for free slip
433     \item VP vs.\ EVP: speed performance, accuracy?
434     \item ocean stress: different water mass properties beneath the ice
435     \end{itemize}
436    
437     \section{Adjoint sensiivities of the MITsim}
438    
439     \subsection{The adjoint of MITsim}
440    
441     The ability to generate tangent linear and adjoint model components
442     of the MITsim has been a main design task.
443     For the ocean the adjoint capability has proven to be an
444     invaluable tool for sensitivity analysis as well as state estimation.
445     In short, the adjoint enables very efficient computation of the gradient
446     of scalar-valued model diagnostics (called cost function or objective function)
447     with respect to many model "variables".
448     These variables can be two- or three-dimensional fields of initial
449     conditions, model parameters such as mixing coefficients, or
450     time-varying surface or lateral (open) boundary conditions.
451     When combined, these variables span a potentially high-dimensional
452     (e.g. O(10$^8$)) so-called control space. Performing parameter perturbations
453     to assess model sensitivities quickly becomes prohibitive at these scales.
454     Alternatively, (time-varying) sensitivities of the objective function
455     to any element of the control space can be computed very efficiently in
456     one single adjoint
457     model integration, provided an efficient adjoint model is available.
458    
459     [REFERENCES]
460    
461    
462     The adjoint operator (ADM) is the transpose of the tangent linear operator (TLM)
463     of the full (in general nonlinear) forward model, i.e. the MITsim.
464     The TLM maps perturbations of elements of the control space
465     (e.g. initial ice thickness distribution)
466     via the model Jacobian
467     to a perturbation in the objective function
468     (e.g. sea-ice export at the end of the integration interval).
469     \textit{Tangent} linearity ensures that the derivatives are evaluated
470     with respect to the underlying model trajectory at each point in time.
471     This is crucial for nonlinear trajectories and the presence of different
472     regimes (e.g. effect of the seaice growth term at or away from the
473     freezing point of the ocean surface).
474     Ensuring tangent linearity can be easily achieved by integrating
475     the full model in sync with the TLM to provide the underlying model state.
476     Ensuring \textit{tangent} adjoints is equally crucial, but much more
477     difficult to achieve because of the reverse nature of the integration:
478     the adjoint accumulates sensitivities backward in time,
479     starting from a unit perturbation of the objective function.
480     The adjoint model requires the model state in reverse order.
481     This presents one of the major complications in deriving an
482     exact, i.e. \textit{tangent} adjoint model.
483    
484     Following closely the development and maintenance of TLM and ADM
485     components of the MITgcm we have relied heavily on the
486     autmomatic differentiation (AD) tool
487     "Transformation of Algorithms in Fortran" (TAF)
488     developed by Fastopt (Giering and Kaminski, 1998)
489     to derive TLM and ADM code of the MITsim.
490     Briefly, the nonlinear parent model is fed to the AD tool which produces
491     derivative code for the specified control space and objective function.
492     Following this approach has (apart from its evident success)
493     several advantages:
494     (1) the adjoint model is the exact adjoint operator of the parent model,
495     (2) the adjoint model can be kept up to date with respect to ongoing
496     development of the parent model, and adjustments to the parent model
497     to extend the automatically generated adjoint are incremental changes
498     only, rather than extensive re-developments,
499     (3) the parallel structure of the parent model is preserved
500     by the adjoint model, ensuring efficient use in high performance
501     computing environments.
502    
503     Some initial code adjustments are required to support dependency analysis
504     of the flow reversal and certain language limitations which may lead
505     to irreducible flow graphs (e.g. GOTO statements).
506     The problem of providing the required model state in reverse order
507     at the time of evaluating nonlinear or conditional
508     derivatives is solved via balancing
509     storing vs. recomputation of the model state in a multi-level
510     checkpointing loop.
511     Again, an initial code adjustment is required to support TAFs
512     checkpointing capability.
513     The code adjustments are sufficiently simply so as not to cause
514     major limitations to the full nonlinear parent model.
515     Once in place, an adjoint model of a new model configuration
516     may be derived in about 10 minutes.
517    
518     [HIGHLIGHT COUPLED NATURE OF THE ADJOINT!]
519    
520     \subsection{Special considerations}
521    
522     * growth term(?)
523    
524     * small active denominators
525    
526     * dynamic solver (implicit function theorem)
527    
528     * approximate adjoints
529    
530    
531     \subsection{An example: sensitivities of sea-ice export through Fram Strait}
532    
533     We demonstrate the power of the adjoint method
534     in the context of investigating sea-ice export sensitivities through Fram Strait
535     (for details of this study see Heimbach et al., 2007).
536     The domain chosen is a coarsened version of the Arctic face of the
537     high-resolution cubed-sphere configuration of the ECCO2 project
538     (see Menemenlis et al. 2005). It covers the entire Arctic,
539     extends into the North Pacific such as to cover the entire
540     ice-covered regions, and comprises parts of the North Atlantic
541     down to XXN to enable analysis of remote influences of the
542     North Atlantic current to sea-ice variability and export.
543     The horizontal resolution varies between XX and YY km
544     with 50 unevenly spaced vertical levels.
545     The adjoint models run efficiently on 80 processors
546     (benchmarks have been performed both on an SGI Altix as well as an
547     IBM SP5 at NASA/ARC).
548    
549     Following a 1-year spinup, the model has been integrated for four years
550     between 1992 and 1995.
551     It is forced using realistic 6-hourly NCEP/NCAR atmospheric state variables.
552     Over the open ocean these are converted into
553     air-sea fluxes via the bulk formulae of Large and Yeager (2004).
554     Derivation of air-sea fluxes in the presence of sea-ice is handled
555     by the ice model as described in Section XXX.
556     The objective function chosen is sea-ice export through Fram Strait
557     computed for December 1995
558     The adjoint model computes sensitivities to sea-ice export back in time
559     from 1995 to 1992 along this trajectory.
560     In principle all adjoint model variable (i.e. Lagrange multipliers)
561     of the coupled ocean/sea-ice model
562     are available to analyze the transient sensitivity behaviour
563     of the ocean and sea-ice state.
564     Over the open ocean, the adjoint of the bulk formula scheme
565     computes sensitivities to the time-varying atmospheric state.
566     Over ice-covered parts, the sea-ice adjoint converts
567     surface ocean sensitivities to atmospheric sensitivities.
568    
569     Fig. XXX(a--d) depict sensitivities of sea-ice export through Fram Strait
570     in December 1995 to changes in sea-ice thickness
571     12, 24, 36, 48 months back in time.
572     Corresponding sensitivities to ocean surface temperature are
573     depicted in Fig. XXX(a--d).
574     The main characteristics is consistency with expected advection
575     of sea-ice over the relevant time scales considered.
576     The general positive pattern means that an increase in
577     sea-ice thickness at location $(x,y)$ and time $t$ will increase
578     sea-ice export through Fram Strait at time $T_e$.
579     Largest distances from Fram Strait indicate fastest sea-ice advection
580     over the time span considered.
581     The ice thickness sensitivities are in close correspondence to
582     ocean surface sentivitites, but of opposite sign.
583     An increase in temperature will incur ice melting, decrease in ice thickness,
584     and therefore decrease in sea-ice export at time $T_e$.
585    
586     The picture is fundamentally different and much more complex
587     for sensitivities to ocean temperatures away from the surface.
588     Fig. XXX (a--d) depicts ice export sensitivities to
589     temperatures at roughly 400 m depth.
590     Primary features are the effect of the heat transport of the North
591     Atlantic current which feeds into the West Spitsbergen current,
592     the circulation around Svalbard, and ...
593    
594     \begin{figure}[t!]
595     \centerline{
596     \subfigure[{\footnotesize -12 months}]
597     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJheff_arc_lev1_tim072_cmax2.0E+02.eps}}
598     %\includegraphics*[width=.3\textwidth]{H_c.bin_res_100_lev1.pdf}
599     %
600     \subfigure[{\footnotesize -24 months}]
601     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJheff_arc_lev1_tim145_cmax2.0E+02.eps}}
602     }
603    
604     \centerline{
605     \subfigure[{\footnotesize
606     -36 months}]
607     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJheff_arc_lev1_tim218_cmax2.0E+02.eps}}
608     %
609     \subfigure[{\footnotesize
610     -48 months}]
611     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJheff_arc_lev1_tim292_cmax2.0E+02.eps}}
612     }
613     \caption{Sensitivity of sea-ice export through Fram Strait in December 2005 to
614     sea-ice thickness at various prior times.
615     \label{fig:4yradjheff}}
616     \end{figure}
617    
618    
619     \begin{figure}[t!]
620     \centerline{
621     \subfigure[{\footnotesize -12 months}]
622     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJtheta_arc_lev1_tim072_cmax5.0E+01.eps}}
623     %\includegraphics*[width=.3\textwidth]{H_c.bin_res_100_lev1.pdf}
624     %
625     \subfigure[{\footnotesize -24 months}]
626     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJtheta_arc_lev1_tim145_cmax5.0E+01.eps}}
627     }
628    
629     \centerline{
630     \subfigure[{\footnotesize
631     -36 months}]
632     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJtheta_arc_lev1_tim218_cmax5.0E+01.eps}}
633     %
634     \subfigure[{\footnotesize
635     -48 months}]
636     {\includegraphics*[width=0.44\linewidth]{figs/run_4yr_ADJtheta_arc_lev1_tim292_cmax5.0E+01.eps}}
637     }
638     \caption{Same as Fig. XXX but for sea surface temperature
639     \label{fig:4yradjthetalev1}}
640     \end{figure}
641    
642    
643    
644     \section{Discussion and conclusion}
645     \label{sec:concl}
646    
647     The story of the paper could be:
648     B-grid ice model + C-grid ocean model does not work properly for us,
649     therefore C-grid ice model with advantages:
650     \begin{enumerate}
651     \item stress coupling simpler (no interpolation required)
652     \item different boundary conditions
653     \item advection schemes carry over trivially from main code
654     \end{enumerate}
655     LSR/EVP solutions are similar with appropriate bcs, evp parameters as
656     a function of forcing time scale (when does VP solution break
657     down). Same for LSR solver, provided that it works (o:
658     Which scheme is more efficient for the resolution/time stepping
659     parameters that we use here. What about other resolutions?
660    
661     \paragraph{Acknowledgements}
662     We thank Jinlun Zhang for providing the original B-grid code and many
663     helpful discussions.
664    
665     \bibliography{bib/journal_abrvs,bib/seaice,bib/genocean,bib/maths,bib/mitgcmuv,bib/fram}
666    
667     \end{document}
668    
669     %%% Local Variables:
670     %%% mode: latex
671     %%% TeX-master: t
672     %%% End:

  ViewVC Help
Powered by ViewVC 1.1.22