Ocean modelling - Centre of Excellence for Climate System Science
Ocean modelling - Centre of Excellence for Climate System Science Ocean modelling - Centre of Excellence for Climate System Science
Ocean Modelling: An Introduction for Everybody Stephanie Waterman Climate Change Research Centre & ARC Centre of Excellence for Climate System Science University of New South Wales [www.nasa.gov/topics/earth/features/perpetual-ocean.html]
- Page 3 and 4: [www.nasa.gov/topics/earth/features
- Page 5 and 6: What is an ocean model ? a represen
- Page 7 and 8: There are many types of ocean model
- Page 9 and 10: Ocean vs. Atmospheric Models I 1. l
- Page 11 and 12: Ocean vs. Atmospheric Models III 3.
- Page 13 and 14: Ocean vs. Atmospheric Models V 5. t
- Page 15 and 16: Ocean Model Ingredients I The$hydro
- Page 17 and 18: Ocean Model Practicalities I 1. the
- Page 19 and 20: 1. the horizontal grid regular grid
- Page 21 and 22: 1. the horizontal grid irregular gr
- Page 23 and 24: 1. the horizontal grid Ocean Model
- Page 25 and 26: 2. the vertical grid Ocean Model Pr
- Page 27 and 28: 2. the vertical grid Ocean Model Pr
- Page 29 and 30: 3. model resolution Ocean Model Pra
- Page 31 and 32: Ocean Model Practicalities III 3. m
- Page 33 and 34: 4. parameterizations Ocean Model Pr
- Page 35 and 36: 4. parameterizations Parameteriza)o
- Page 37 and 38: 4. parameterizations Ocean Model Pr
- Page 39 and 40: Sources of Error Errors in ocean mo
- Page 41 and 42: That’s nice, but what is needed f
- Page 43 and 44: Want to know more? • MIT Open Cou
<strong>Ocean</strong> Modelling:<br />
An Introduction <strong>for</strong> Everybody<br />
Stephanie Waterman<br />
<strong>Climate</strong> Change Research <strong>Centre</strong><br />
& ARC <strong>Centre</strong> <strong>of</strong> <strong>Excellence</strong> <strong>for</strong> <strong>Climate</strong> <strong>System</strong> <strong>Science</strong><br />
University <strong>of</strong> New South Wales<br />
[www.nasa.gov/topics/earth/features/perpetual-ocean.html]
[www.nasa.gov/topics/earth/features/perpetual-ocean.html]
<strong>Ocean</strong> Modelling:<br />
An Introduction <strong>for</strong> Everybody<br />
Stephanie Waterman<br />
<strong>Climate</strong> Change Research <strong>Centre</strong><br />
& ARC <strong>Centre</strong> <strong>of</strong> <strong>Excellence</strong> <strong>for</strong> <strong>Climate</strong> <strong>System</strong> <strong>Science</strong><br />
University <strong>of</strong> New South Wales<br />
[www.nasa.gov/topics/earth/features/perpetual-ocean.html]
What is an ocean model ?<br />
a representation<br />
in the <strong>for</strong>m <strong>of</strong><br />
equations / computer code<br />
describing<br />
physical processes<br />
<strong>of</strong> our understanding <strong>of</strong> how the ocean works.
What is an ocean model ?<br />
a representation<br />
in the <strong>for</strong>m <strong>of</strong><br />
equations / computer code<br />
describing<br />
physical processes<br />
<strong>of</strong> our understanding <strong>of</strong> how the ocean works.<br />
•the exchange <strong>of</strong> energy, mass, and momentum between the ocean<br />
and external sources (e.g. radiation, evaporation, precipitation, river run<strong>of</strong>f, wind<br />
energy that creates waves or currents etc. etc.)<br />
•ocean movement/dynamics including horizontal advection and vertical convection;<br />
and<br />
•3-dimensional mixing and dissipation processes at scales from molecular to<br />
ocean basin<br />
• ...
There are many types <strong>of</strong> ocean models...<br />
integration time<br />
number <strong>of</strong><br />
processes<br />
Earth Models <strong>of</strong><br />
Intermediate Complexity<br />
(EMICs)<br />
Global <strong>Climate</strong> Models or<br />
General Circulation Models<br />
(GCMs)<br />
conceptual or<br />
process models<br />
detail <strong>of</strong> description
There are many ways to use ocean models...<br />
• to consider future climate scenarios<br />
• to make operational now-casts and <strong>for</strong>ecasts<br />
• to experimentally investigate hypotheses <strong>for</strong><br />
the workings <strong>of</strong> ocean and climate system<br />
phenomena<br />
• to mechanistically interpret ocean observations<br />
• ...
<strong>Ocean</strong> vs. Atmospheric Models I<br />
1. liquid vs. gas<br />
• air is a compressible gas; seawater is a<br />
(nearly) incompressible liquid<br />
• this relationship requires a fundamentally<br />
different equation <strong>of</strong> state:<br />
- atmosphere: ideal gas law (easy!)<br />
- ocean: density = fn(temperature,<br />
salinity,pressure) (hard!)<br />
• BUT (in most applications) we can assume<br />
incompressibility (so water into a box =<br />
water out)
<strong>Ocean</strong> vs. Atmospheric Models II<br />
2. salinty vs. humidity<br />
• seawater contains dissolved chemicals<br />
known collectively as “salinity”<br />
• ocean models must account <strong>for</strong> the<br />
effects <strong>of</strong> salinity on density in an<br />
analogous way that atmospheric<br />
models must account <strong>for</strong> humidity<br />
[http://en.wikipedia.org/wiki/File:IAPSO_Standard_Seawater.jpg#filelinks]
<strong>Ocean</strong> vs. Atmospheric Models III<br />
3. vertical structure<br />
• the vertical structure <strong>of</strong> the ocean and<br />
atmosphere share both similarities and<br />
differences<br />
• both have a well-mixed layer near the surface<br />
where most <strong>of</strong> the heating and cooling occurs<br />
• BUT mixing occurs (to some degree) throughout<br />
the atmosphere while the ocean is much more<br />
stratified below its thin mixed layer<br />
• because the ocean is so stratified, ocean<br />
models can make assumptions about the<br />
dominance <strong>of</strong> horizontal processes over those<br />
in the vertical<br />
• ocean modellers can benefit from vertical<br />
coordinate systems that exploit the fact that<br />
much <strong>of</strong> the action occurs in the near-surface<br />
mixed layer<br />
Temperature Pr<strong>of</strong>ile: Seafloor to Stratosphere<br />
[source: The COMET program]
<strong>Ocean</strong> vs. Atmospheric Models IV<br />
4. horizontal structure<br />
• the atmosphere blankets the earth in a<br />
laterally continuous layer; it is pierced by<br />
(relatively) small mountain ranges<br />
• the ocean is bounded on 5 <strong>of</strong> 6 sides by<br />
complex topography:<br />
- a series <strong>of</strong> irregularly shaped basins<br />
- fringed by narrow continental shelves<br />
- bottom bathymetry plays on 0(1) role<br />
• there are lateral boundary conditions:<br />
freshwater run-<strong>of</strong>f from the continents<br />
that alters density and currents along the<br />
coast<br />
• the need to resolve both horizontal and<br />
vertical processes along ocean margins<br />
plays a critical role the choice <strong>of</strong><br />
horizontal and vertical co-ordinate<br />
systems<br />
[courtesy <strong>of</strong> Maxim Nikurashin]
<strong>Ocean</strong> vs. Atmospheric Models V<br />
5. time and length scales <strong>of</strong> motion<br />
• Steve told you that atmospheric models<br />
are more expensive than ocean models<br />
<strong>for</strong> a given grid size<br />
• BUT the energy-containing (geostrophic)<br />
scales in the atmosphere are much<br />
larger than those in the ocean<br />
• this means we need to model the ocean<br />
at a finer resolution to resolve the same<br />
“types” <strong>of</strong> features<br />
• also because the ocean interior is almost<br />
adiabatic (i.e. along-density surfaces)<br />
whereas the atmosphere is (relatively)<br />
well-mixed, the equilibrium timescale <strong>of</strong><br />
the ocean is much s l o w e r<br />
• to spin up a ocean model from rest:<br />
1000 + years integration !!!<br />
The space and time scales <strong>of</strong> various ocean phenomena.<br />
Figure 3. The approximate space and time scales <strong>of</strong> phenomena <strong>of</strong> interest that could be investigated from<br />
altimetric measurements <strong>of</strong> ocean topography with adequate spatial and temporal resolution. The dashed lines<br />
indicate the approximate lower bounds <strong>of</strong> the space and time scales that can be resolved [source: in SSH fields Chelton constructed (ed) 2001]
<strong>Ocean</strong> vs. Atmospheric Models<br />
<strong>Ocean</strong> modellers have it both easy and hard<br />
compared to their atmospheric<br />
counterparts:<br />
• the ocean is (nearly) incompressible:<br />
water in ~ water out<br />
• the ocean is strongly stratified:<br />
horizontal processes dominate over<br />
vertical ones<br />
• no change <strong>of</strong> state <strong>of</strong> seawater: just<br />
<strong>for</strong>m ice when T < -1.8 o C<br />
BUT<br />
• domain geometry is complex<br />
• lateral boundary conditions are required<br />
and poorly constrained<br />
• the eddies are smaller<br />
• the spin-up time is longer<br />
• there are fewer observations <strong>for</strong><br />
validation<br />
[courtesy <strong>of</strong> Maxim Nikurashin]
<strong>Ocean</strong> Model Ingredients I<br />
The$hydrosta,c$equa,ons$<br />
1. the primitive (hydrostatic) equations<br />
<br />
du<br />
dt<br />
∇P<br />
<br />
+ fk × u = − + F + ν∇<br />
ρ<br />
∂P<br />
∂z<br />
=<br />
− gρ<br />
∂ρ<br />
∂ρ<br />
& ∂w<br />
#<br />
+ u. ∇<br />
H<br />
ρ + w + ρ$<br />
∇<br />
H<br />
. u + ! =<br />
∂t<br />
∂z<br />
% ∂z<br />
"<br />
dT<br />
dt<br />
dS<br />
dt<br />
= κ∇<br />
2<br />
= κ ∇<br />
T<br />
S<br />
+<br />
+<br />
Q T<br />
2<br />
S<br />
Q S<br />
2<br />
<br />
u<br />
0<br />
Horizontal$momentum$<br />
Hydrosta,c$balance$<br />
Conserva,on$<strong>of</strong>$mass$<br />
Conserva,on$<strong>of</strong>$heat$<br />
Conserva,on$<strong>of</strong>$salinity$<br />
ρ = ρ( P,<br />
S,<br />
T<br />
)<br />
Nonlinear$equa,on$<strong>of</strong>$state$<br />
7$coupled$equa,ons$in$7$unknowns:$u,$v,$w,$P,$T,$S,$ρ; all$varying$in$3$spa,al$direc,ons$a<br />
= 7 coupled equations in 7 unknowns: u, v, w, P, T, S,
<strong>Ocean</strong> Model Ingredients cont.<br />
2. boundary conditions<br />
- basin geometry<br />
- bottom topography<br />
- an atmosphere on top<br />
3. initial conditions (maybe from climatology i.e. the mean<br />
state or a previous already spun up model run)<br />
- initial temperature and salinity fields<br />
- initial velocity fields<br />
4. <strong>for</strong>cing fields<br />
- shortwave radiation, long-wave radiation, latent heat and sensible heat<br />
at the surface<br />
- evaporation and precipitation at the surface<br />
- land surface run-<strong>of</strong>f at the margins<br />
- winds<br />
- tides
<strong>Ocean</strong> Model Practicalities I<br />
1. the horizontal grid<br />
Regular Grid<br />
• regular grids consist <strong>of</strong><br />
regularly spaced lines<br />
• on a spherical earth CAN’T<br />
have both uni<strong>for</strong>m grid<br />
spacing AND straight lines<br />
• in practice grids tend to be<br />
curvilinear and their internal<br />
spacing tends to vary<br />
• regular lat-lon grids also<br />
have a problem at the poles<br />
where grid lines converge
<strong>Ocean</strong> Model Practicalities I<br />
1. the horizontal grid<br />
Tri-polar Grid<br />
• a cleaver solution: the<br />
tripolar grid = a circular<br />
grid laid over the Arctic<br />
polar region with two poles<br />
positioned over land<br />
[Schopf 2005 after Murray 1996]
1. the horizontal grid<br />
regular grids:<br />
• are computationally efficient<br />
• have (relatively) straight<strong>for</strong>ward<br />
analysis algorithms<br />
• have benefited from decades <strong>of</strong><br />
research experience<br />
BUT<br />
• (<strong>for</strong> a given latitude) have a<br />
fixed resolution: to increase<br />
resolution near the edge <strong>of</strong> an<br />
ocean basin (where you want<br />
it!) requires an increase <strong>of</strong><br />
resolution everywhere including<br />
out in the middle <strong>of</strong> the ocean<br />
(where you don’t!)<br />
<strong>Ocean</strong> Model Practicalities I
1. the horizontal grid<br />
<strong>Ocean</strong> Model Practicalities I<br />
• irregular grids are designed<br />
to give you more freedom to<br />
put spatial resolution where<br />
you want it<br />
• a common scheme is<br />
composed as a series <strong>of</strong><br />
triangles = “finite elements”<br />
• by varying the triangle size<br />
we can construct a nonuni<strong>for</strong>m<br />
horizontal resolution<br />
over the computational<br />
domain<br />
[Gorman et al, 2006]
1. the horizontal grid<br />
irregular grids:<br />
<strong>Ocean</strong> Model Practicalities I<br />
• are efficient owing to the fact that resolution<br />
can be tailored to need as a function <strong>of</strong> space<br />
• can accurately represent highly irregular<br />
coastlines and topography (no need <strong>for</strong><br />
“staircase” coasts and topography!)<br />
BUT<br />
• have unresolved issues with flows dominated<br />
by geostrophy and advection (characteristics<br />
<strong>of</strong> large-scale ocean flows!)<br />
• have spatially variable resolution-dependent<br />
physics (e.g. viscosity and diffusivity<br />
coefficients should really be resolution<br />
dependent!)<br />
• have spatially variable spurious diapycnal<br />
mixing associated with the numerical scheme<br />
<strong>for</strong> advection<br />
(a)<br />
(b)<br />
[http://www.ocean-modeling.org]
1. the horizontal grid<br />
irregular grids:<br />
<strong>Ocean</strong> Model Practicalities I<br />
• have issues related to the matching <strong>of</strong> low<br />
resolution regions (say with eddy<br />
parameterizations employed) to highly<br />
resolved regions<br />
• tools to analyze are immature<br />
• are computationally expensive: low-order finite<br />
element models are about an order <strong>of</strong><br />
magnitude more expensive than finite<br />
difference models per degree <strong>of</strong> freedom<br />
(a)<br />
• while common in coastal and estuarine<br />
<strong>modelling</strong>, irregular grids are uncommon<br />
in large-scale ocean <strong>modelling</strong><br />
• CLIVAR’s Working Group on <strong>Ocean</strong> Model<br />
Development (WGOMD) however anticipate<br />
that a coupled climate model using an<br />
unstructured mesh ocean will emerge<br />
within a decade<br />
(b)<br />
[http://www.ocean-modeling.org]
1. the horizontal grid<br />
<strong>Ocean</strong> Model Practicalities I<br />
• adaptive grids =<br />
unstructured meshes with a<br />
dynamically changing<br />
resolution responding to the<br />
nature <strong>of</strong> flow in time<br />
• very cool but very much in<br />
its infancy <strong>for</strong> ocean<br />
<strong>modelling</strong> applications<br />
• “there will inevitably come a<br />
point at which the additional<br />
overhead <strong>of</strong> uni<strong>for</strong>m mesh<br />
refinement where it is not<br />
required is greater than the<br />
overhead associated with the<br />
use <strong>of</strong> finite elements and<br />
mesh adaptivity” - ICOM<br />
model developer<br />
[Brito-Parada et al, 2012]
2. the vertical grid<br />
<strong>Ocean</strong> Model Practicalities II<br />
Discre)za)on:*v<br />
• the choice <strong>of</strong> the vertical co-ordinate system is<br />
loaded because:<br />
- the oceans are <strong>for</strong>ced at the surface; most <strong>of</strong><br />
the “action” occurs there<br />
- the oceans are strongly stratified<br />
- the oceans are ~ adiabatic in the interior<br />
Discre)za)on:*ver)cal*coordi<br />
z<br />
Height Terrai<br />
• Simple<br />
• Bounda<br />
• No pressure gradient<br />
errors<br />
resolut<br />
Discre)za)on:*ver)cal*coordinate*<br />
- there is complex bottom bathymetry to deal • Spurious diapycnal<br />
z<br />
σ<br />
with<br />
fluxes<br />
z<br />
• as a consequence there exist a number Height <strong>of</strong><br />
Terrain following x<br />
approaches to choose from • Simple<br />
• Boundary layer<br />
• No pressure gradient resolution (BBL)<br />
errors<br />
z<br />
Height<br />
• Spurious diapycnal<br />
σ fluxes<br />
Terrain following<br />
MOM,*POP,*MITgcm,*OPA*<br />
• Pressure gradient<br />
error<br />
ρ<br />
• Spurious diapycnal<br />
fluxes Isopycnal<br />
σ<br />
• Pressu<br />
ρ error<br />
• Spuriou<br />
Iso<br />
fluxes<br />
• Simple<br />
ROMS<br />
• “Exactly<br />
• No reso<br />
unstrati<br />
[courtesy <strong>of</strong> Sonya Legg]
2. the vertical grid<br />
<strong>Ocean</strong> Model Practicalities II<br />
1. absolute depth / z-coordinate system<br />
• based on a series <strong>of</strong> depth levels<br />
• common to add vertical resolution near<br />
the surface by decreasing the spacing<br />
between the levels in the upper ocean<br />
relative to the deep<br />
z<br />
Discre)za)on:*v<br />
Discre)za)o<br />
Discre)za)on:*ver)cal*coordi<br />
Height Terrai<br />
• Simple<br />
• Bounda<br />
• No pressure gradient<br />
errors<br />
resolut<br />
• ADVANTAGES: simple to set up;<br />
computationally efficient; there are no<br />
pressure gradient<br />
Discre)za)on:*ver)cal*coordinate*<br />
errors<br />
• Spurious diapycnal<br />
z<br />
σ<br />
• DISADVANTAGES: increased vertical<br />
fluxes<br />
z<br />
resolution near the lateral boundaries Height z Terrain following x<br />
(i.e. on the continental slopes) requires • Simple<br />
• MOM,*POP,*MITgcm,*OPA*<br />
Boundary layer<br />
the addition <strong>of</strong> grid cells throughout • No pressure gradient resolution<br />
Height<br />
(BBL)<br />
the basin; spurious diapycnal mixing errors<br />
associated with the numerical advection<br />
• Simple • Pressure gradient<br />
scheme<br />
• Spurious diapycnal<br />
z<br />
σ<br />
error<br />
fluxes<br />
ρ<br />
e.g. MOM (ocean model in ACCESS)<br />
• Spurious diapycnal<br />
Height Terrain following fluxes Isopycnal<br />
σ<br />
• Pressu<br />
ρ error<br />
• Spuriou<br />
Iso<br />
fluxes<br />
• Simple<br />
ROMS<br />
• “Exactly<br />
• No reso<br />
unstrati<br />
•<br />
• No pressure gradient<br />
[courtesy <strong>of</strong> Sonya Legg]
2. the vertical grid<br />
<strong>Ocean</strong> Model Practicalities II<br />
Discre)za)on:*v<br />
Discre)za)on:*ver)cal*co<br />
2. “terrain following” / σ-coordinate system<br />
• based on the fractional depth, scaled<br />
from 0 to 1:<br />
- 0.01 level is 1% <strong>of</strong> the depth <strong>of</strong> the ocean<br />
- 0.5 level is exactly half the depth <strong>of</strong> the ocean<br />
- 0.99 level is at 99% <strong>of</strong> the depth <strong>of</strong> the ocean<br />
Discre)za)on:*ver)cal*coordi<br />
z<br />
Height Terrai<br />
• Simple<br />
• Bounda<br />
• No pressure gradient<br />
errors<br />
resolut<br />
• ADVANTAGES: mimics the bathymetry<br />
and allows high resolution near the sea<br />
• Pressu<br />
floor regardless<br />
Discre)za)on:*ver)cal*coordinate*<br />
<strong>of</strong> depth or proximity<br />
• Spurious diapycnal<br />
z<br />
σ<br />
ρ error<br />
to land<br />
fluxes<br />
z<br />
• Spuriou<br />
• DISADVANTAGES: z<br />
pressure gradient Height σ Terrain following x Iso<br />
fluxes<br />
errors; issues with spurious diapycnal • Simple<br />
• MOM,*POP,*MITgcm,*OPA*<br />
Boundary layer • Simple<br />
ROMS<br />
mixing coming from the numerical • No pressure gradient resolution (BBL) • “Exactly<br />
Height<br />
advection scheme<br />
errors<br />
Terrain following • No reso<br />
• Simple<br />
• Pressure gradient unstrati<br />
• Spurious<br />
•<br />
diapycnal<br />
Boundary z<br />
σ<br />
error layer<br />
e.g. ROMS, POM<br />
fluxes<br />
ρ<br />
• No pressure gradient<br />
• Spurious diapycnal<br />
Height<br />
resolution Terrain following fluxes Isopycnal (BBL)<br />
[courtesy <strong>of</strong> Sonya Legg]<br />
σ<br />
•
2. the vertical grid<br />
<strong>Ocean</strong> Model Practicalities II<br />
• vertical grid defined by density surfaces<br />
• exploits the fact that below the mixed<br />
layer, ocean currents generally flow<br />
along surfaces <strong>of</strong> equal density (flow is<br />
“adiabatic”)<br />
z<br />
Discre)za)on:*v<br />
on:*ver)cal*coordinate*<br />
3. density (ρ) / “isopycnal” coordinate<br />
system (“layered models”)<br />
Discre)za)on:*ver)cal*coordi<br />
Height Terrai<br />
• Simple<br />
• Bounda<br />
• No pressure gradient<br />
errors<br />
resolut<br />
• ADVANTAGES: simple, “exactly<br />
Discre)za)on:*ver)cal*coordinate*<br />
isopycnal” (no spurious diapycnal mixing!)<br />
• Spurious diapycnal<br />
z<br />
σ<br />
fluxes<br />
z<br />
• DISADVANTAGES: σ<br />
per<strong>for</strong>m poorly ρ Height Terrain following x<br />
where the ocean is less stratified • (e.g. Simple in<br />
• Boundary layer<br />
shallow water); no resolution in an<br />
• No pressure gradient resolution (BBL)<br />
unstratified Terrain fluid; no following<br />
mixed layer unless Isopycnal<br />
errors<br />
you tack one on; issues with entrainment<br />
• Pressure gradient<br />
• Spurious • diapycnal Simple<br />
z<br />
σ<br />
error<br />
fluxes<br />
ρ<br />
e.g. GOLD, MICOM<br />
resolution (BBL)<br />
• Spurious diapycnal<br />
Height Terrain following fluxes Isopycnal<br />
• Boundary layer<br />
nt<br />
MOM,*POP,*MITgcm,*OPA*<br />
• “Exactly” Adiabatic<br />
z<br />
σ<br />
x<br />
• Pressu<br />
ρ error<br />
• Spuriou<br />
Iso<br />
fluxes<br />
• Simple<br />
ROMS<br />
• “Exactly<br />
• No reso<br />
unstrati<br />
[courtesy <strong>of</strong> Sonya Legg]
2. the vertical grid<br />
4. hybrid coordinate systems<br />
<strong>Ocean</strong> Model Practicalities II<br />
• seek to optimize per<strong>for</strong>mance by<br />
combining the best suited system in<br />
different regions based on the<br />
dominant processes at work<br />
• HYCOM = z coordinates in the<br />
surface mixed layer and density<br />
coordinates in the stratified ocean<br />
below<br />
• here the vertical coordinates<br />
evolve in both time and space as<br />
the depth <strong>of</strong> the mixed layer<br />
changes. ADVANTAGE: dynamically<br />
optimized coordinate system gives<br />
improved results; DISADVANTAGE:<br />
high computational cost<br />
[courtesy <strong>of</strong> Eric Chassingnet]
3. model resolution<br />
<strong>Ocean</strong> Model Practicalities III<br />
What are we trying to resolve?!?<br />
• like the atmosphere the ocean has<br />
“multiple scale variability” = a<br />
broadband <strong>of</strong> time and length scales<br />
that are important/on which motions<br />
exhibit variation<br />
• further, processes include coupling<br />
across scales (“non-local interactions”)<br />
and these scale interactions can have<br />
important effects<br />
• the range <strong>of</strong> important time and space<br />
scales is immense: molecular (mm and<br />
seconds) to basin scale (10 000 km and<br />
1000 years)<br />
• multiple scale variability and scale<br />
interactions make ocean dynamics<br />
fascinating but very challenging to<br />
model!<br />
The space and time scales <strong>of</strong> various ocean processes.<br />
[adapted from: Chelton (ed) 2001 and Dickey and Chang 2001]
3. model resolution<br />
<strong>Ocean</strong> Model Practicalities III<br />
• general principles <strong>of</strong> resolution are the<br />
same <strong>for</strong> both atmospheric and ocean<br />
models<br />
• there are different rules <strong>of</strong> thumb: one<br />
is that it takes 5 grid points to<br />
accurately define a feature without<br />
aliasing<br />
• this means 1/8° global resolution with<br />
an average horizontal grid cell <strong>of</strong> 14<br />
km can accurately depict only features<br />
larger than 56 km<br />
• models with variable grid spacing have<br />
variable resolution - beware <strong>of</strong><br />
resolution-dependent physics!<br />
• resolution is not cheap - because <strong>of</strong> the<br />
CFL* condition, as we shrink the<br />
horizontal grid spacing we must add<br />
vertical layers and decrease the time<br />
step<br />
“every halving <strong>of</strong> the grid spacing<br />
requires roughly ten times as many<br />
computations”<br />
* no transport faster than one grid cell per time step!
<strong>Ocean</strong> Model Practicalities III<br />
3. model resolution<br />
• global ocean models <strong>of</strong>ten describe their horizontal<br />
resolution with respect to their ability to “permit” or<br />
“resolve” mesoscale (i.e. Rossby radius scale) eddies<br />
1 o “coarse resolution” model<br />
resolution<br />
≥ 1°<br />
~ 0.5°<br />
≤ 0.2°<br />
lingo<br />
“coarse”<br />
“eddy-permitting”<br />
“eddy-resolving”<br />
meaning<br />
no eddies<br />
some eddies<br />
eddies generate at realistic<br />
strength and rate<br />
• “eddy resolving” does NOT mean all eddies are<br />
resolved or that all eddy effects <strong>of</strong> resolved eddies<br />
are acting !!!<br />
• the spatial resolution <strong>of</strong> the ocean component <strong>of</strong><br />
CMIP5 coupled models is 0.2° to 2°: from “coarse” (no<br />
eddies) to “eddy-permitting” (partially resolved eddy<br />
field)<br />
• the effects <strong>of</strong> eddies need to be parameterized in<br />
coarse models (more later!); what to do in models that<br />
partially resolve the eddy field is an increasingly<br />
important question<br />
0.1 o “eddy-resolving” model<br />
[courtesy <strong>of</strong> Peter Gent]
4. parameterizations<br />
<strong>Ocean</strong> Model Practicalities IV<br />
• processes need to be parameterized in a model<br />
<strong>for</strong> 2 main reasons:<br />
1. we won’t spend the computational resources<br />
required to directly treat them because they<br />
are either too small or too complex;<br />
2. we don’t understand it well enough to be<br />
represented by an equation<br />
Physical processes in overflows/gravity currents<br />
Final properties, transport<br />
and depth <strong>of</strong> overflow product<br />
water depend on all these<br />
processes<br />
• processes commonly parameterized in ocean<br />
models include:<br />
- mesoscale eddy effects<br />
- submesoscale eddy effects<br />
- dense overflows<br />
- coastal processes<br />
- surface mixed layer processes<br />
- friction<br />
- sub-grid scale mixing<br />
- ocean-ice interactions<br />
Geostrophic eddies<br />
Neutral<br />
buoyancy<br />
level<br />
• low res models: the main problem is mesoscale<br />
eddies; high res models: submesoscale eddies<br />
(fronts and filaments), internal wave mixing,<br />
details <strong>of</strong> flow-topography interactions<br />
Entrainment <strong>of</strong><br />
ambient water<br />
Shear<br />
instability<br />
Bottom friction<br />
Downslope descent<br />
Upper ocean flow<br />
Isopycnal coordinate models must explicitly include parameterization <strong>of</strong> entrainment.<br />
Z-coordinate models have difficulties capturing descent without excessive spurious mixing.<br />
All coarse resolution models have difficulties with small-scale topography, e.g. channels.<br />
y<br />
Hydraulic<br />
control<br />
[courtesy <strong>of</strong> Sonya Legg]<br />
z<br />
x<br />
[courtesy <strong>of</strong> Bob Weller]<br />
(Goosse*et*al)*<br />
[Goosse et al.]<br />
CICE*model*fundamental*equa)on:*56/50 =−8.(36)−<br />
6(.,h,0);h=frac)onal*area*covered*by*ice*in*thickness*range*(h,h+<br />
9=rate*<strong>of</strong>*thermodynamic*ice*growth*<br />
:=redistribu)on*by*ridging*<br />
[courtesy <strong>of</strong> Max Nikurahsin]
4. parameterizations<br />
<strong>Ocean</strong> Model Practicalities IV<br />
1. parameterizing mesoscale eddy effects:<br />
• mesoscale eddy effects include:<br />
- mixing along isopycnals<br />
- restratifying / flattening isopycnals<br />
- non-local fluxes: transport by rings, Meddies<br />
- acting as a source <strong>of</strong> small scale noise/<br />
variability<br />
- modifying air-sea fluxes<br />
- cascading energy to different scales<br />
- ...<br />
Labrador Sea:<br />
Evidence <strong>of</strong> mesoscale eddies<br />
Labrador Sea<br />
Greenland<br />
• critical eddy parameterizations in<br />
coarse-resolution models are:<br />
1. Along-isopycnal diffusion (Redi)<br />
2. Bolus transport (Gent and<br />
McWilliams or “GM”)<br />
Sea Surface Temperature (10th July 1992)<br />
Sea surface temperature showing the eddy field in the Labrador Sea.<br />
[Jourdain et al. 2008]
1.*Mixing*tracers*along*isop<br />
<strong>Ocean</strong><br />
isopycnals.*<br />
Model Practicalities IV<br />
1. Along-isopycnal diffusion (Redi)<br />
= mixing <strong>of</strong> tracers (temperature,<br />
salinty etc.) along density surfaces<br />
2.Bolus transport (Gent and<br />
McWilliams or “GM”)<br />
A*tracer*is*diffused*along*isopycnals,* & 1+<br />
S<br />
isopycnals.*<br />
1<br />
− S<br />
4. parameterizations − ∇ . u<br />
' τ ' = ∇.<br />
κ K ∇τ<br />
where* K<br />
ρ Redi<br />
=<br />
1. parameterizing mesoscale eddy effects:<br />
$ $$ Redi<br />
+<br />
2 x<br />
1 | S |<br />
% Sx<br />
− ∇. u<br />
' τ ' = ∇.<br />
κ K ∇τ<br />
whe<br />
• mesoscale eddy effects include:<br />
ρ Redi<br />
- mixing along isopycnals<br />
S xσ<br />
x<br />
= −∂ / ∂<br />
end result <strong>of</strong><br />
z<br />
- restratifying / flattening isopycnals<br />
mixing by<br />
isopycnal large-scale<br />
- non-local fluxes: transport by rings, Meddies<br />
eddies along<br />
diffusivity (resolved)<br />
- are a source <strong>of</strong> small scale noise/variability<br />
tracer gradient<br />
density surfaces<br />
- modify air-sea fluxes<br />
& 1 0 S #<br />
- cascade energy to different scales<br />
where<br />
$<br />
x<br />
!<br />
- ...<br />
If*|S|≪1* K = $ 0 1 S !<br />
Redi<br />
y<br />
• critical eddy parameterizations in<br />
$<br />
2 !<br />
coarse-resolution models are:<br />
% S S | S |<br />
x y "<br />
If*|S|≪1* K =<br />
eddy<br />
diffusivity<br />
components <strong>of</strong><br />
the (resolved)<br />
isopycnal<br />
slope<br />
(the*components<br />
Re di<br />
[see Redi, 1982]<br />
[courtesy <strong>of</strong> Sonya Legg]
4. parameterizations<br />
Parameteriza)on<br />
<strong>Ocean</strong> Model Practicalities IV<br />
1. parameterizing mesoscale eddy effects:<br />
• mesoscale eddy effects include:<br />
- mixing along isopycnals<br />
- restratifying / flattening isopycnals<br />
- non-local fluxes: transport by rings, Meddies<br />
- are a source <strong>of</strong> small scale noise/variability<br />
- modify air-sea fluxes<br />
- cascade energy to different scales<br />
- ...<br />
• critical eddy parameterizations in<br />
coarse-resolution models are:<br />
1. Along-isopycnal diffusion (Redi)<br />
2.Bolus transport (Gent and<br />
McWilliams or “GM”)<br />
= represents the advective or transport effect<br />
<strong>of</strong> eddies by means <strong>of</strong> a “bolus” velocity<br />
2.*Bolus*transport*(Gent*and<br />
2.*Bolus*transport*(Gent*an<br />
<br />
<br />
− ∇. u'<br />
' τ ''<br />
=<br />
−∇.<br />
. τ u<br />
Parameteriza)ons*<strong>of</strong>*mesoscale<br />
u * & − ∂ z(<br />
κ GMS<br />
x)<br />
#<br />
<br />
x <br />
− ∇. u'<br />
τ ' = −∇.<br />
τ u *<br />
“bolus velocity”<br />
Where*3 ↑∗ *is*the*bolus*velocity*<br />
& u *#<br />
& − ∂<br />
z<br />
( κGM<br />
Sx<br />
) #<br />
<br />
. $<br />
τ<br />
!<br />
u <br />
$<br />
!<br />
u*<br />
= $ v*<br />
! =<br />
* $<br />
= − ∇ ..<br />
∂<br />
z<br />
( κGM<br />
S<br />
y<br />
) !<br />
$ ! $<br />
!<br />
% w*<br />
" % ∂<br />
x<br />
( κGM<br />
Sx<br />
) + ∂<br />
y<br />
( κGM<br />
S<br />
y<br />
)"<br />
− ∇ . τ u * = ∇ .<br />
( )<br />
κ ∇ τ<br />
K GM GM<br />
*<br />
Where*3<br />
Where*3 effect <strong>of</strong> eddies<br />
↑∗<br />
↑∗<br />
*is*the*bolus*velocity*<br />
*is*the*bolus*velocity*<br />
& u *<br />
# & − ∂ ( κ S ) #<br />
z GM x<br />
<br />
$<br />
!<br />
$<br />
!<br />
!<br />
u<br />
*<br />
=<br />
$<br />
v<br />
*<br />
!<br />
=<br />
$<br />
− ∂<br />
z<br />
z<br />
((<br />
κGM<br />
GM<br />
S<br />
y<br />
y<br />
))<br />
!!<br />
2.*Bolus*transport*(Gent*and*McWilliams)**<br />
$<br />
!<br />
$<br />
!!<br />
%<br />
w<br />
*<br />
"<br />
%<br />
∂<br />
x(<br />
(<br />
κGM<br />
GM<br />
Sx<br />
x<br />
))<br />
+ ∂<br />
y<br />
y<br />
((<br />
κGM<br />
GM<br />
SS<br />
y<br />
y<br />
))<br />
""<br />
−<br />
∇<br />
- mixes “isopycnal thickness” to flatten density<br />
surfaces without mixing density (to mimic baroclinic instability)<br />
<br />
<br />
. u τ τ u<br />
end result <strong>of</strong> mixing<br />
by the “advective”<br />
( ))<br />
components <strong>of</strong><br />
κ the (resolved)<br />
∇<br />
τ<br />
GM K GM<br />
K GM GM<br />
isopycnal slope<br />
“GM<br />
eddy<br />
diffusivity”<br />
& 0 0 − Sx<br />
#<br />
$<br />
!<br />
KGM<br />
= $ 0 0 − S<br />
y !<br />
$<br />
!<br />
% Sx<br />
S<br />
y<br />
0 "<br />
An*an)Osymmetric*tensor*<br />
K<br />
GM<br />
G<br />
z<br />
κ = αL | S | N<br />
2<br />
κ GM =eddy*coefficient$ S<br />
GM<br />
κ GM 2 [see (Visbeck*et*al,*1994)*<br />
GM<br />
κ GM =eddy*coefficient$ κ<br />
Gent and 2<br />
GM =<br />
McWilliams, αL 1990]<br />
| | S<br />
(A<br />
Tend<br />
isopy<br />
dens<br />
An A<br />
L |<br />
[courtesy <strong>of</strong> Sonya Legg]
4. parameterizations<br />
<strong>Ocean</strong> Model Practicalities IV<br />
1. parameterizing mesoscale eddy effects:<br />
• adding along-isopycnal diffusivity and bolus transport “fixed” the problem <strong>of</strong> widespread<br />
convection and localized it to the few deep water <strong>for</strong>mation sites as observed<br />
constant horizontal diffusivity<br />
with along-isopycnal diffusivity and<br />
bolus transport<br />
Locations <strong>of</strong> deep convection/deep water <strong>for</strong>mation in a 3 o x 3 o global model.<br />
[JDanabasoglu et al.1994]<br />
[courtesy <strong>of</strong> Sonya Legg]
4. parameterizations<br />
<strong>Ocean</strong> Model Practicalities IV<br />
1. parameterizing mesoscale eddy effects:<br />
• adding along-isopycnal diffusivity and bolus transport “fixed” the problem <strong>of</strong> widespread<br />
convection and localized it to the few deep water <strong>for</strong>mation sites as observed<br />
There once was an ocean model called MOM,<br />
constant horizontal diffusivity<br />
with along-isopycnal diffusivity and<br />
bolus transport<br />
which occasionally used to bomb,<br />
but eddy advection, and much less convection,<br />
turned it into a stable NCOM*.<br />
* Navy Coastal <strong>Ocean</strong> Model.<br />
Locations <strong>of</strong> deep convection/deep water <strong>for</strong>mation in a 3 o x 3 o global model.<br />
[JDanabasoglu et al.1994]<br />
[courtesy <strong>of</strong> Sonya Legg]
4. parameterizations<br />
2. vertical mixing schemes<br />
<strong>Ocean</strong> Model Practicalities IV<br />
Mixed*layer*<br />
Eddy*diffusivity*<br />
w ' τ ' ( d)<br />
w ' τ ' ( d)<br />
Mixed*layer*<br />
ver)cal*fluxes*<br />
Eddy*diffusivity*<br />
& ∂<br />
τ $ #<br />
$<br />
% ∂z<br />
= −κ<br />
!<br />
τ<br />
−γ<br />
τ<br />
DownOgradient*<br />
% ∂z<br />
"<br />
1994)*<br />
& ∂ τ #<br />
= −κ<br />
$ −γ<br />
!<br />
τ<br />
"<br />
end<br />
diffusion*<br />
• when the ocean becomes statically ver)cal*fluxes* result <strong>of</strong> eddy<br />
DownOgradient*<br />
downgradient<br />
vertical<br />
non local<br />
Nondimensional*<br />
unstable ( z > 0 ) vertical over-turning<br />
diffusivity<br />
KOpr<strong>of</strong>ile*parameteriza)on*(Large*et*al,*<br />
mixing by diffusion* shape*func)on*<br />
flux<br />
diffusion<br />
should occur but cannot because we<br />
eddies 1994)*<br />
Eddy*diffusivity*<br />
make the hydrostatic approximation<br />
Nondimensional*<br />
& d # & d # & d #<br />
κ $ ! = hw $ ! G$<br />
!<br />
(vertical acceleration is excluded!)<br />
& ∂ τ # τ<br />
1↓2 τis*nonOzero*only*<strong>for</strong>*<br />
w ' τ ' ( d)<br />
= −κ<br />
$ ! shape*func)on*<br />
% h " % h " % h "<br />
τ<br />
−γ<br />
τ<br />
tracers*in*convec)ve*<strong>for</strong>cing*<br />
• so vertical mixing must be accomplished % ∂z<br />
Mixed*layer*<br />
"<br />
condi)ons,*to*account*<strong>for</strong>*<br />
mixed Mixed*layer*<br />
vertical<br />
Ver)cal*<br />
via a very large coefficient <strong>of</strong><br />
ver)cal*fluxes*<br />
counterOgradient*flux$<br />
vertical<br />
non-dimensional<br />
DownOgradient*<br />
& d #<br />
depth*<br />
& d #<br />
velocity<br />
& dvelocity*scale*<br />
#<br />
diffusion*<br />
diffusion<br />
κ $ ! =<br />
Nonlocal*flux*<br />
shape function<br />
hw $ ! Gscale<br />
τ<br />
$ !<br />
τ<br />
• many models use the K-Pr<strong>of</strong>ile % h "<br />
Allows*<strong>for</strong>*gradients*in*mixed*layer,*and*<br />
% h " % h<br />
Nondimensional*<br />
Convec)ve*buoyancy*flux*<br />
"<br />
shape*func)on* penetra)ve*convec)on**<br />
Parameterization (Large et al., 1994)<br />
Mixed*layer*<br />
= large mixing in the upper ocean due & dto # many & d # & d # Ver)cal*<br />
κ τ $ ! = hwτ<br />
$ ! G<br />
processes (but dominated by wind) and very depth*<br />
$ !<br />
% h " % h " % h " z* velocity*scale* h*<br />
much weaker mixing in the deep ocean Mixed*layer* due to Ver)cal*<br />
Convec)vely*<br />
internal wave breaking and tides<br />
depth* velocity*scale*<br />
Allows*<strong>for</strong>*gradients*in*mixed*layer,*and*<br />
modified*<br />
buoyancy*<br />
penetra)ve*convec)on** b*<br />
Allows*<strong>for</strong>*gradients*in*mixed*layer,*and*<br />
penetra)ve*convec)on**<br />
*<br />
1994)*<br />
Nonlocal*flux*<br />
Nonlocal*flux*<br />
Conve<br />
z*<br />
z*<br />
Con<br />
Sources <strong>of</strong> Error<br />
Errors in ocean model arise from several<br />
sources:<br />
• mathematical error from the truncation<br />
and rounding <strong>of</strong> numbers from one time<br />
step to the next<br />
• propagating errors from errors in the<br />
<strong>for</strong>cing fields<br />
• inaccurate parameterizations<br />
• missing processes or missing interactions<br />
between processes<br />
• spurious (numerical) diapycnal mixing<br />
associated with the numerical advection<br />
scheme
Validation<br />
• validation is a challenge because ocean observations are sparse<br />
• things are tough but getting better. Now:<br />
- satellite measurements <strong>of</strong> sea surface height (SSH), sea surface temperature (SST), salinity<br />
and winds<br />
- in situ measurements from coastal stations, buoys, fixed current meters, ships, drifters, floats<br />
and gliders<br />
• sustained observations are critical to ocean model development !!!
That’s nice, but what is needed from the ocean<br />
component model to get climate change correct?<br />
• to get heat and CO 2 uptake correct<br />
• a good representation <strong>of</strong> the circulation including the meridional<br />
overturning circulation (MOC)<br />
• a good vertical mixing scheme to get correct mixed layer depths and<br />
deep water <strong>for</strong>mation in small, local regions as observed<br />
• a good parameterization <strong>of</strong> the effects <strong>of</strong> mesoscale eddies - they<br />
aren’t resolved in (most) climate models<br />
• ...<br />
[courtesy <strong>of</strong> Peter Gent]
your challenges to solve:<br />
<strong>Ocean</strong> Models: The Future<br />
“The ideal ocean climate model has high enough resolution to resolve eddies and<br />
topography, zero numerical diffusion and is efficient enough to integrate <strong>for</strong> 1000s<br />
<strong>of</strong> years.”<br />
• model biases and model drift<br />
• projections <strong>of</strong> sea-level rise: most current GCMs are Boussinesq and must calculate<br />
the steric contribution to sea-level rise a-posteriori<br />
• the realistic representation <strong>of</strong> mixing: how much, where? why?<br />
• spurious mixing: models are too diffusive; in z-coordinate resolution models numerical<br />
mixing depends on resolution<br />
• overflows<br />
• getting the energy out <strong>of</strong> the mesoscale eddy field<br />
• sub-mesoscale effects/parameterization<br />
• mixed layer depth and dynamics<br />
• effects <strong>of</strong> the internal wave field<br />
• parameterizations <strong>for</strong> partially resolved eddy fields; resolution-dependent<br />
parameterizations<br />
• ICE: dynamic ice-sheets and ice shelves, iceberg transport (lack <strong>of</strong> leads to cold-fresh<br />
bias around Antarctica!)<br />
• ...
Want to know more?<br />
• MIT Open Courseware: “12.984 Atmospheric and <strong>Ocean</strong>ic Modeling”<br />
http://ocw.mit.edu/courses/earth-atmospheric-and-planetary-sciences/12-950-<br />
atmospheric-and-oceanic-modeling-spring-2004/index.htm<br />
• Griffies, S M, 2004: Fundamentals <strong>of</strong> <strong>Ocean</strong> <strong>Climate</strong> Models, Princeton, NJ:<br />
Princeton University Press, 518 pp.<br />
• Griffies, S M, C Böning, F O Bryan, E P Chassignet, R Gerdes, H Hasumi, A C<br />
Hirst, A M Treguier, and D Webb, 2000: Developments in ocean climate<br />
<strong>modelling</strong>. <strong>Ocean</strong> Modelling, 2, 123-192.<br />
• Griffies, S M, and A Adcr<strong>of</strong>t, 2008: “Formulating the equations <strong>of</strong> ocean<br />
models” In <strong>Ocean</strong> Modeling in an Eddying Regime, Geophysical Monograph 177,<br />
M. W. Hecht, and H. Hasumi, eds., Washington, DC, American Geophysical Union,<br />
281-318.<br />
• Griffies, S M, 2009: “<strong>Science</strong> <strong>of</strong> ocean climate models” In Encyclopedia <strong>of</strong> <strong>Ocean</strong><br />
<strong>Science</strong>s, 2nd edition, Elsevier, DOI: 10.1016/B978-012374473-9.00714-1.<br />
• Griffies, S M, A Adcr<strong>of</strong>t, A Gnanadesikan, R W Hallberg, M J Harrison, S Legg, C<br />
M Little, M Nikurashin, A Pirani, B L Samuels, J R Toggweiler, and G K Vallis, et<br />
al., 2010: Problems and prospects in large-scale ocean circulation models In<br />
<strong>Ocean</strong> Obs '09, 21-25 September, Venice, Italy, ESA Special Publication, DOI:<br />
10.5270/<strong>Ocean</strong>Obs09.cwp.38.
[courtesy <strong>of</strong> Andreas Klocker, CoE]