Ufuk Utku Turuncoglu

İTÜ, Informatics Institute

ROMS Tutorial
Introduction and

1st Regional Earth System Modeling Summer School, 5-7 September 2016, ITU, Turkey
Air-sea Interaction
Ocean Modeling Challenges
Space-Time Scales
LOW-RES – FAST ( ~3º)
MID-RES – CLIMATE ( ~1º) O(+100 years/day)
O(10-100 years/day)

HIGH-RES ( ~0.1º)
O(1 year/day)

Processes with
various temporal and
spatial scales:

MID-RES can not capture

meso-scale eddies observed
in nature
Ocean Modeling:
Physics
• Forces
• Gravity (g)
• Pressure Gradient
Force (PGF) due to
horizontal height
and density
differences
• Coriolis force (f) due to rotating earth
• Friction – between wind and sea surface, between
current and sea floor, and between between water
masses having different velocities
𝑁𝑒𝑤𝑡𝑜𝑛' 𝑠 𝑆𝑒𝑐𝑜𝑛𝑑 𝐿𝑎𝑤 → 𝐹 = 𝑚𝑎
3 𝐹 = 𝑃𝐺𝐹 + 𝑔 + 𝑓 + 𝐹𝑟𝑖𝑐𝑡𝑖𝑜𝑛 → 𝑁𝑎𝑣𝑖𝑒𝑟 𝑆𝑡𝑜𝑘𝑒𝑠
Physics in ROMS
Glenn et al., 2016

momentum
balance

heat
balance

surface boundary cond. bottom boundary cond.

 First mode of Internal gravity wave speed

Ocean Modeling Challenges 𝑅=

𝒄𝟐𝒈
(𝒇𝟐 + 2𝜷𝑐E )
Rossby Radius of Deformation Coriolis meridional
parameter gradient of f
R. Hallberg / Ocean Modelling 72 (2013) 92–103 93

The horizontal resolution needed to resolve the first baroclinic deformation radius with two grid points
Solid black line represent the regions that can be solved by 1 ○ – 1/8○ horizontal resolution
Parameterizations
• It is used to include physical effects of unresolved sub-
grid scale processes such as turbulence
• Lateral mixing of momentum and tracers by mesoscale and
sub-mesoscale eddies
• Vertical mixing of momentum and tracers (surface and
bottom boundary layers, interior)
• Tidal mixing
• Convection
• Diurnal cycle for shortwave heat flux and penetration through
water column (Jerlow Water Type, …)
• ...
• Physically based
• As simple as possible (computational overhead!)
• As few parameters as possible
Parameterizations in ROMS
• Vertical mixing (momentum and tracers)
• K-profile (LMD; Large et al., 1994)
• Surface Boundary Layer
• Interior
• Mellor-Yamada 2.5 (Mellor and Yamada, 1982)
• Generic Length Scale (GLS; Umlauf and Burchard, 2001;
Warner et al., 2005)
• k-kl
• k-epsilon
• k-omega
• k-generic
• Horizontal mixing of momentum and tracers
• Geopotential (const. depth) / isopycnic (const. density)
Why do we need regional models?
• We need higher spatial resolution to solve science
problems such as
• Climate variability and prediction (i.e. sea level change,
extreme events)
• Coastal processes and ecosystems (i.e. upwelling)
• Estuaries, interconnected basins (i.e. TSS)
• Marginal seas (Mediterranean / Black / Marmara Seas)
• …
• Information for impacts and adaptation purposes
• Investigating processes that are better represented
in higher spatial and temporal scales
Regional Modelling Approaches
• High resolution global models (not practical due to
requirement of extensive computational resources
and data management issues – post-processing)
• Dynamic and static grid refinement in global
models (it might help to reduce biases)
• Imbedded high resolution model with in a coarser
model (two-way interaction among models can be
tricky)
• Regional ocean models with/without open
boundary conditions (coming from observations or
global models)
ROMS imbedded in POP In this case, regional
ocean model
(ROMS) is used
together with global
ocean model (POP)
to improve global
model’s SST

The coupling tools

such as ESMF and
MCT help to
integrate different
models
Ocean Model – Horizontal Grids
• Structured grid
• Generic representation of rectangular and curvilinear type of grids
• Grid positions can be computed (procedural)
• It can be uniform (dx=dy) and non-uniform (dx≠dy)

non-uniform dimension

origin spacing

uniform

Represented by the combination of origin, dimension

and spacing
 , ROMS)

Ocean Model –
Horizontal Grids
• Curvilinear grid
• When the solution domain is not
rectangular Cartesian grids cannot
be used.
• Structured curvilinear grids are
based on mapping of the flow
domain onto a computational
domain
http://www.emg.cmar.csiro.au

Sannino et al., 2002, GRL

http://www.mohid.com/wiki/index.php?title=Grid
Ocean Model –
Horizontal Grids
• Unstructured grid
• It is more suitable to use in complex
coastlines such as interconnected basins
• Formulation in unstructured
meshes more difficult and expensive (2-4 times slower)
Ocean Model – Vertical Grids
• The choice of a vertical coordinate system is one of the most
important aspects of a model design and strongly depend
on application
• Z-coordinate (i.e. MITgcm, MOM):
• It is the simplest system and used by

Equally spaced
various ocean models
• Z-coordinate model has problem with
the regions of sloping topography such as
coastal regions. In here, the levels
intersect the bathymetry and
unrealistic vertical velocities near

Hres in surface
the bottom can result.
• Increasing the number of vertical levels
will improve the representation of the
near-bottom flow, but at a high
computational cost.
Ocean Model – Vertical Grids
• Sigma or S-coordinate (i.e. ROMS, POM):
• Help to keep same number of vertical layer in everywhere
independent from the depth of the water column. The layer
thickness vary with the depth
• The layers are more closely spaced

Equally spaced
near the surface and/or bottom than
in the interior, thus allowing the
boundary layers to be better resolved.
• This type of coordinate is most appropriate
for continental shelf and coastal regions,
where the bottom and surface boundary
layers may merge.

Hres in surface
• They have difficulty handling sharp
topographic changes from one grid
point to another. Pressure-gradient errors
can give rise to unrealistic flows.
• Smoothing might help !
Ocean Model – Vertical Grids
• Isopycnal coordinates:
• These models use the potential density
referenced to a given pressure as the
vertical coordinate.
• This system basically divides the water
column into distinct homogeneous layers,
whose thicknesses can vary from place
to place and from one time step to
the next.
• This choice of coordinate works well for modeling tracer (T, S)
transport, which tends to be along surfaces of constant density.
• The main advantage of these models is the computational cost.
They could run in higher horizontal resolution rather than vertical
one.
• Their applications are limited with the costal regions and boundary
layers (surface and bottom).
Black Sea Example
• A sensitivity study of NEMO ocean model with different
vertical grid configurations
Spurious zonal velocities
along 31º
E after 3.5 months of
simulation: (a) z-coordinate;
(b) s-coordinate; (c) s-
coordinate with enveloping
topography; (d) s-on-top-of-
z; (e) s-enveloped-top-of-z.

Shapiro et al., 2013

Black Sea Example
Observation, MyOcean

ROMS ocean model

Model, ROMS configured for
Black Sea

Due to sharp
bathymetry
gradient, the model
produces unrealistic
vertical mixing of
tracers
Sensitivity to horizontal resolution
1/12º
http://noc.ac.uk/science-
technology/climate-sea-
level/changing-circulation/high-
ORCA05

resolution-global-modelling

http://www.geomar.de/en/research/fb1/fb1-tm/ocean-models/
1/4º
ORCA025

1º
ORCA12
Ocean Motion –
Ekman Transport
• The Ekman spiral, named
after Swedish scientist Vagn
Walfrid Ekman (1874-1954)
who first theorized it in 1902,
is a consequence of the Coriolis
effect.
• As a result, each successively deeper layer of water moves
more slowly to the right or left, creating a spiral effect.
• Because the deeper layers of water move more slowly than
the shallower layers, they tend to “twist around” and flow
opposite to the surface current.
Ocean Motion –
Upwelling
• Where Ekman transport moves surface waters away from
the coast, surface waters are replaced by water that wells
up from below in the process known as upwelling. This
example is from the Northern Hemisphere.
• Where Ekman transport moves surface waters toward the
coast, the water piles up and sinks in the process known as
downwelling. This example is from the Northern
Hemisphere.
Ocean Motion –
Upwelling …
Ocean Motion –
Upwelling …
Ocean Motion –
Upwelling …
Regional examples: Caspian Sea, Black Sea
Input for Ocean Models
Input for Ocean Models …
• Definition of horizontal and vertical grid
• Interpolation of bathymetry (ETOPO, GEBCO etc.)
• Initial conditions (WOA, SODA, global ocean models).
MEDAR/MEDATLAS can be used for Mediterranean and
Black Seas.
• Atmospheric forcing (Reanalysis datasets such as ERA-
Interim, NCEP/NCAR, COARE, global and/or regional
atmosphere model)
• Open boundary conditions (T, S, SSH, U, V …; SODA, Global
ocean circulation model such as MOM)
• River input (freshwater) from GRDC, RivDis etc. It could be
also calculated using river routing or hydrological model.
Input for Ocean Models …

Rivers contributes to the net freshwater

budget of the ocean / inland seas / lakes

It also affects the dynamics by changing the

salinity and stratification around coastlines

The river plumes might have big impact in

ocean.
Input for Ocean Models …
• Definition of horizontal and vertical grid
• Interpolation of bathymetry (ETOPO, GEBCO etc.)
• Initial conditions (WOA, SODA, global ocean models).
MEDAR/MEDATLAS can be used for Mediterranean and
Black Seas.
• Atmospheric forcing (Reanalysis datasets such as ERA-
Interim, NCEP/NCAR, COARE, global and/or regional
atmosphere model)
• Open boundary conditions (T, S, SSH, U, V …; SODA, Global
ocean circulation model such as MOM)
• River input (freshwater) from GRDC, RivDis etc. It could be
also calculated using river routing or hydrological model.
 Grid Generation
• Realistic and/or analytic
• ROMS uses s-coordinate in
vertical and it follows bathymetry
• The distribution of vertical layers
can be controlled using set of
parameters
• In horizontal, curvilinear grid is
supported
• Grid Generation Tools:
• MATLAB: SeaGrid, Gridgen, EasyGrid …
• Python: octant, pyroms, …
• Analytic grid can be also defined using Fortran
• The grid generation plays crucial role in the stability
 Grid Generation
Hsmooth-Hraw
• Bathymetry smoothing:

• Grid stiffness ratios:

" Δh % " hi − hi−1 % Beckman & Haidvogel Number (1993)
rxo = max $ ' = max $ '
# 2h & h + h
# i i−1 & ) Reduced by smoothing < 0.25

"z −z % Haney Number (1991)

i, j,k i−1, j,k + zi, j,k−1 − zi−1, j,k−1
rx1 = max $$ '' Reduced by smoothing and vert.
# zi, j,k + zi−1, j,k − zi, j,k−1 − zi−1, j,k−1 & coordinate modification < 6.0

• Tool: Matlab smoothing toolbox (LP_Bathymetry)

http://drobilica.irb.hr/~mathieu/Bathymetry/
 Initial and Lateral Boundary Conditions
• Realistic and/or analytic
• Requires (3d for IC and 2d for Lateral BC)
• salinity,
• temperature
• current (u and v)
• sea surface height fields.
• The input files are defined in netCDF format
• The input files should be recreated when the horizontal and
vertical grid changed!
• There is no need to change files for atmospheric forcing if
horizontal grid is same
• Atmospheric forcing
• Momentum, heat and freshwater fluxes or atmospheric
conditions
 Installation
• The code is distributed using SVN repository
• Users need to register to access the
codehttps://www.myroms.org/index.php?page=RomsCode
• Get code
svn checkout https://www.myroms.org/svn/src/trunk MyDir
• Directories

OS/Architecture/compiler specific files

ROMS source code, files related with test cases

 Installation
• Directories under ROMS/

ROMS build script, build.sh

Input parameter files for tests, *.in and varinfo.dat

Files to create analytic initial and boundary conditions
Option files for test cases, CPP options

Source files for non-linear model

Source files for utility modules

 Installation
• Main steps:
• Modify build environment (i.e. Compilers/Linux-ifort.mk)
• External libraries (netCDF, HDF etc.)
• MPI compiler (OpenMPI, IntelMPI, …)
• OpenMP ?
• Copy build script from ROMS/Bin/build.sh to installation
directory
• Create or copy application specific header file
• CPP options for model customization
• Edit build.sh
• Case identifier (i.e. UPWELLING)
• Specify source directory
• MPI, compiler definitions
• Run build script to install model, ./build.sh
 Installation
• Application specific options are defined by CPP definitions
• CPP definitions act as a filter to create customized model
source code
• CPP options include:
• Momentum equations (mixing, advection, bottom friction, …)
• Tracer equations (mixing, advection, relaxation, rivers …)
• Pressure gradient algorithm
• Atmospheric boundary conditions (i.e. bulk flux alg.)
• Analytical field definitions (grid, initial conditions etc.)
• Vertical mixing parameterizations (GLS, MY, LMD, …)
• Open boundary conditions
• Biological (i.e. NPZD, EcoSim) + Sediment Transport models
• Sea-ice …
Referances
• http://www.oc.nps.edu/nom/modeling/vertical_grids.html
• http://imos.org.au/fileadmin/user_upload/shared/IMOS%2
0General/ACOMO/ACOMO_2014/presentations/wednesday
/Robin_Robertson__Recovered_.pdf