1st Regional

Earth System Modeling Summer School, 5-7 September 2016, ITU, Turkey
Introduction to Model Coupling in
Earth System Science and Recent
Developments
Ufuk Utku Turuncoglu
ITU, Informatics Institute
ICTP, ESP Section
Earth System
• It is represented by complex and non-linear interaction
between different elements (atmosphere, hydrosphere,
geosphere and biosphere).
• All the processes have different spatial and temporal scales

http://wearebrainstorm.com/products/earth-science-set-of-4-prints
Earth System …
• All the processes have different spatial and temporal scales
• Response time under forcing also differs

IPCC Report
NWF by
Climate Modeling
Richardson
• Evolution:

http://prezi.com/pakaaiek3nol/timeline-of-climate-modeling/
Coordinated Experiments
CMIPs

Automatized NWF
by Von Neumann
High Performance Computing
• Evolution:

Close relationship
between development
in HPC and climate
modelling

The development
curve (or slope) is
more steep after
80s!
Complexity of Climate Models

• Model complexity is increasing in terms of physics and

model components
• Along with the development in HPC, the horizontal and
vertical resolutions of models are increasing
Evolution of Models

• More processes are

represented (i.e. chemistry,
sea-ice, glaciers, dynamic
vegetation)
• Higher temporal and
spatial scales
Climate Assessments
• Evolution of IPCC Reports

http://www.scientificamerican.com/article/climate-assessments-25-years-of-the-ipcc/
No decrease in uncertainty!
but models are more accurate in 20C
What is Model Coupling?
• It is a sophisticated way of simulate climate system and
complex interactions between its components.
• The individual components are not perfect !
• Our theoretical understanding of climate is still incomplete, and
certain simplifying assumptions are unavoidable when building
these models (Reichler and Kim, 2008) !
• Coupled model does not always gives best results but help to
understand interactions and processes between components !
Basic effects of ocean-atmosphere coupling
• Increases variance of both media
• Decrease energy fluxes
between them
• Prescribing mid-latitude
SSTs does not lead to a
correct simulation of
low-frequency thermal
variance in the atmosphere.
• We need coupled atmosphere-ocean model !!!
• How does it work?
• Strong wind speed -> lower SST via mixing and turbulent flux.
Negative correlation = atmosphere drives ocean
• Enhanced (reduced) wind speed over warm (cold) SSTs. Positive
correlation = ocean drives atmosphere

Farnetti, ICTP Regional Climate Model WS

Difficulties in Model Coupling?
• Example from Atmosphere + Ocean Coupling
1. Initial state of the ocean is not precisely known

http://www.ipcc.ch/publications_and_data/ar4/wg1/en/ch1s1-5-3.html
Lack of continuous and high resolution observations.
Remote sensing might help but covers only ocean surface and
recent decades. The quality also depends on many issues
(atmospheric correction and used algorithm, coastal regions etc.)
2. An imbalance in the surface flux (heat, momentum and
freshwater) much smaller than the observational accuracy is
enough to cause a drifting of coupled GCM simulations into
unrealistic states
• It is not easy tightly couple models that are written by different
groups
• Monolithic model development approach does not work anymore!
• High accuracy of conservation (heat and mass) is needed!
Types of Model Coupling
There are two types of model coupling:
1. Offline
Models run sequentially and the interaction between them is in
only one way. The simplest example is dynamical downscaling
(GCM to RCM) or one-way nesting (RCM to RCM)
2. Online
In this case, models interact in both way and feedback
mechanisms among components represented (two way nesting,
fully coupled atmosphere-ocean models or ESMs.)
 Techniques for Model Coupling

http://www.cerfacs.fr/~coquart/pagecerfacs/seminars/201405_OASIS3MCT.pdf
The ESM components might be coupled using different
programming approaches
1. Merge individual model codes
host / parent J easy to code,
program atm
… ATM program ocn
… OCN portable
! read initial and boundary data ! read initial and boundary data
call read() call read() L not flexible and
! run model
call run()
! run model generic
call run()
! write results to file ! write results to file
… …
end program atm end program ocn

program atm subroutine ocn (exchange vars)

… … single executable!
! read initial and boundary data ! read initial and boundary data
call read() call read()
! run model ! run model
call run() call run()
! run ocn model ! write results to file
call ocn(exchange vars) …
! write results to file end subroutine ocn
…
end program atm
more information in Valcke, 2009 EGU
 Techniques for Model Coupling
The ESM components might be coupled using different
programming approaches
2. Use existing communication protocols: MPI, InterComm …
host / parent J existing code
program atm
… ATM program ocn
… OCN and concurrent
! read initial and boundary data ! read initial and boundary data components
call read() call read()
! run model
call run()
! run model L not easy to
call run()
! write results to file ! write results to file implement, not
… …
end program atm end program ocn flexible and
generic
program atm program ocn
… … multiple executable!
! receive exchange field ! receive exchange field
call recv() call recv()
! run model ! run model
call run() call run() ATM OCN
! send exchange field ! send exchange field
call send() call send()
! write results to file ! write results to file
… …
end program atm end program ocn more information in Valcke, 2009 EGU
 Techniques for Model Coupling
The ESM components might be coupled using different
programming approaches
3. Use coupling framework & libraries: FMS, ESMF, MCT …

• Code divided into units and calling interface redesigned

(i.e. initialization, run and finalize)
• Hierarchical model structures are supported

J flexible, efficient, portable, support to

use generic utilities for interpolation, time
management and unit conversions),
support both sequential and concurrent
components
L Existing codes must be restructured

more information in Valcke, 2009 EGU

 Techniques for Model Coupling
The ESM components might be coupled using different
programming approaches
4. Use a coupler: OASIS, OASIS-MCT, MPCCI, PALM …

• Coupler link the model components and handle coupling processes

such as interpolation, synchronization etc.
• Components models might be separate or single executable

J flexible, efficient, portable, support to

use generic utilities for interpolation, time
management and unit conversions),
support concurrent components
L multi-executable:
more difficult to debug; harder to
manage for the OS and possible waste
of resource if seq. execution enforced
more information in Valcke, 2009 EGU
 What is driver / coupler ?
• The driver can be defined as a “glue”
• It merges the different components of the modeling system
• In general, the modeling system might be designed to have
• Single executable
• Multiple executable (each component has an executable)
• In the multiple executable case, the usage of the modeling
system might be difficult
• The driver is generally responsible for
• Interaction among the components:
exchange boundary data between
the components, regridding etc.
• Synchronization:
to coordinate the time evolution
of the physical models (coupling time step)
 http://www.climateurope.eu/earth-system-modeling-a-definition/

Earth System Models

• Defines interaction between components to simulate the
state of the climate system in regional and global scale
• ESMs include processes, impacts, and complete feedback
cycles; for example, they can simulate droughts as well as
the resulting change in plant cover due to the drought,
which may lead to more or less drought (Heavens et al,
2013).
• Climate Model vs. Earth System Model
blue boxes
represent the
processes
included in a
climate model;

green boxes
represent the
additional
components that
may be included in
an Earth System
Model
 Coupled Model Design
• Global Models (GCMs and ESMs)
• There is no single design and ultimate solution!

http://climatesight.files.wordpress.com/2011/08/poster.pdf
Kaitlin Alexander and Steve Easterbrook
Regional Earth System Modeling (RESM)
• Higher resolution representation of physical processes
• Includes more sophisticated physical parameterizations and
additional processes along with their non-linear interactions
• It might also include human behavior (pollution, irrigation
etc.)
• Apart from the global ESMs, they require boundary
condition (global ESMs, reanalysis datasets etc.), which adds
extra complexity to the system

dynamical
downscalling

Prein et al., 2015 @ Reviews of Geophysics

MORCE, IPSL
Coupled Model Design
• Examples for Regional Models

http://www.smhi.se/polopoly_fs/1.28917!/RO_47.pdf
RCA4-NEMO,
SMHI

Data exchange and interpolation is

REMO-RCM, one of the most important
Max Planck component of the model design!
http://www.remo-rcm.de/Atmosphere-Ocean-coupled.1269.0.html
Regional Earth System Modeling / ITU
Year Description Domains
2012 • No driver Caspian Sea
RegCM is hosting also ocean component (Turuncoglu et
• Single ocean model is supported (ROMS) al., 2013;
• Poor mass and energy conservation for exchange fields Geophysical
• No automatized extrapolation (unaligned land-sea masks !!!) Model Dev.)
• Hard to include additional components such as river, wave etc.
2013 • Centralized driver using ESMF’s NUOPC layer (via connectors) Med. Sea
• All components are plugged into the driver
• Added support for two different ocean model component
(ROMS and MITgcm)
• Mass and energy conservation is improved via customized bilinear
interpolation along with global conservation support
• Support for extrapolation (unaligned land-sea masks)
• River Routing (Max Planck’s HD) component is included
2014 • The wave component (ECMWF’s WAM) is included (Surenkok & Med. Sea
2015 Turuncoglu, 2015; EGU) Black Sea
• Extensive benchmarking (PRACE – 2010PA2442)
2016 • ESMF library is updated to 7.0.0 Med. Sea
• Validation for Mediterranean domain Caribbean
(Turuncoglu & Sannino, 2016; Climate Dynamics) Indian Ocean
• Extensive validation with different configuration (2/3/4 component, South Atlantic
different coupling intervals etc.) – paper is on-the-way
RegESM Design
• Model components merged with ESMF/NUOPC
natural
ATM:
emissions Two different land initial and boundary
conditions ICTP’s RegCM 4.4 / 4.5
surface model
atmosphere
(RegCM) OCN:
full gas
chemistry Rutgers Univ.
ROMS (r737)
aerosol
land
(CLM)
land
anthropogenic
emissions MITgcm (63s / 64s)
(BATS)

initial
conditions WAV:
ECMWF’s WAM
wave
(WAM)

ocean
(ROMS / MITgcm) driver
4.5.3 MPI
initial
conditions
sea ice
unnamed
ESMF+NUOPC
RTM:
river
routing
(HD)
Max Planck’s HD
Two different (1.0.2 modified)
ocean model Special thanks to
Prof. Stefan Hagemann
# Following combination of model components can be used: 2 component: ATM-OCN, ATM-WAV,
3 component: ATM-OCN-RTM, 4 component: ATM-OCN-WAV-RTM
Performance Benchmark @ PRACE
• Test with Mediterranean domain (Standard + Extended)
ATM: 12 km - 24 layer
ATM-Extended OCN: 1/12 deg. - 32 layer (ROMS)
ATM

OCN # extended domain is configured

to feed the computational resources

• Tests:
• Different coupling interval (30 min., 1 hour, 3 hours)
• Different execution type (sequential vs. concurrent)
• Different number of component (ATM-OCN, ATM-OCN-RTM)
• Test Environment:
• CURIE @ France (PRACE – 2010PA2442)
Performance Benchmark …
• Individual model components
Better scaling
results for
extended
domain

To find best 2d
decomposition
parameters for
ROMS
Performance Benchmark … # core % diff
30m/1hr
% diff
3hr/1hr

• Coupling interval (only two component) 64 -0.38 1.29

128 1.06 2.15
192 1.45 2.89
256 1.86 1.58
288 9.20 0.95
320 -1.27 5.55
336 17.50 3.84
384 29.66 0.18
416 13.04 -0.92
528 13.50 -4.83
576 9.57 -0.75
640 1.93 8.68
AVG 8.09 1.72

• The effect of coupling interval is very limited J

• 30 min case has more fluctuations (it might be related with
the overload of the cluster)
• It is better to repeat tests couple of time to take more
reliable measurements L
Performance Benchmark …
• Number of model components (const. coupling time step)

Sequential
coupling

• Last processor is shared between OCN and RTM

• RTM component reduces the performance ~ 30% in higher
processor counts L
• Solutions:
• Integrated RTM component (with RegCM4, i.e. Chym)
• Using higher resolution and parallelized (MPI) RTM component
such as RAPID etc. It could also help to improve river rep.
 Model for Mediterranean Basin
The Scientific and
Technological Research
Council of Turkey
(TUBITAK)
founded 2 year project
(under grant 113Y108),
ended in Dec. 2015

• Atmosphere: RegCM4 revision 4283 (~50 km)

• Ocean: ROMS revision783 (1/12 deg. ~ 9 km)
• Closed boundary in Atlantic – used as a buffer zone
• The coupling time step is 3 hour
• ATM-OCN: wind stress, net heat and freshwater flux (E-P), shortwave rad.,
surface pressure and OCN-ATM: sea surface temperature
• Prescribed river discharge (generated by Max Planck HD model)

# It is the first attempt for the validation of ROMS (Regional Ocean Modeling System) ocean model for Med.
 Turuncoglu and Sannino, 2016 @ CD

Validation
• Sea Surface Temperature

SST anomaly
over Med. Sea

Seasonal
SST
Climatology

ERSST used in
standalone
simulation
 Turuncoglu and Sannino, 2016 @ CD

Validation …
• Surface Wind and
Circulation

Wind speed
underestimated
over Gulf of Lion

# coupled model tends to decrease wind speed over the sea when it is compared with standalone simulation

Surface
and 300 m
circulation
is well
represented
Validation …
• Heat flux components
over Med.
• Coupled and
Standalone model
simulations are very
similar except LHF
• The net heat flux is
in the range for both
CPL and STD runs
SWF LWF SWF+LWF SH LH NET

CORE.2 180.41 -81.24 99.17 -20.18 -99.80 -20.81

The coupled model
NOCS 200.02 -62.21 137.81 -8.79 -91.93 37.10 reduces LH over
EINT 218.26 -100.14 114.12 -17.03 -112.12 -15.03 Mediterranean Sea
R50E 200.75 -82.34 118.41 -11.38 -121.72 -14.70
C50E 200.70 -81.31 119.39 -9.85 -110.52 -0.99
Turuncoglu and Sannino, 2016 @ CD
Validation …
• E, P and E-P
over Med.
• Coupled model
tends to reduce
evaporation
• The monthly
distribution of
E, P and E-P are
very similar for
STD and CPL
• The accepted
E-P is
1000 mm/yr

Turuncoglu and Sannino, 2016 @ CD

Validation …
• Spatial distribution
of E, P and E-P
• The effect of coupled
model is more
apparent in EMED
• The CPL model
has more P in
south of EMED
• The E-P estimates
are consistent with
available obs. for
CPL model

Turuncoglu and Sannino, 2016 @ CD

Data Access
• Currently, ITU is a data provider for MedCORDEX project
• Both coupled (RegCM4+ROMS) and standalone model
simulations (RegCM4) are uploaded
• Spatial resolution is MED44
• Daily and monthly averages of surface variables
• Monthly average of atmospheric variables (ua, va, ta, z) at 850 and
500 mb
• Documentation of simulations
• ITU-RegCM4 - http://mistrals.sedoo.fr/?editDatsId=1434
• ITU-RegESM1 - http://mistrals.sedoo.fr/?editDatsId=1433
• Turuncoglu and Sannino, 2016 @ Climate Dynamics
• Data Access
• It is distributed via MedCORDEX database
• https://www.medcordex.eu/medcordex_help_get.php
Extensive Testing of Modeling System

Run Run Id Coupling Coupling Active Description

Short Name Type Time Step Models
AO1 A1 Explicit 30 mins ATM+OCN
AO2 A2 Explicit 1 hour ATM+OCN
AO3 A3 Explicit 3 hours ATM+OCN base run
AO4 A5 Semi-implicit 3 hours ATM+OCN
AW1 B1 Explicit 3 hours ATM+WAV u and v <-> rough.
AW2 B2 Explicit 3 hours ATM+WAV wdir, friction vel.<-> rough.
AOR C1 Explicit 3 hours ATM+OCN+RTM river discharge as SBC
AORW D1 Explicit 3 hours ATM+OCN+RTM+WAV
Standalone E, F, G, H - - ATM/OCN/RTM/WAV
Comparison based on # used models
• Inter-annual variability and monthly climatology of
E, P and E-P
• The effect of
WAV
component
is minimal in
E, P and E-P
• The OCN
mainly affects
the E-P balance
by reducing E
and P
• The monthly
climatology
is also modified

A – ATM-OCN / B – ATM-WAV / C – ATM-OCN-RTM / D – ATM-OCN-RTM-WAV

 Applications Mediterranean
ATM OCN RTM

Models RegCM MITgcm HD

4.4.5.7 c64s
Res. 20 km, 23L 1/12º, 75L 0.5º
353x253 276x408 global
ICBC ERA-Int 12 km online
ERSST ALADIN ATM
Details BATS GGL90 17 major
Grell+MIT rivers
• Coupled with RegESM driver
• 3 hours coupling interval between ATM and OCN
and 1 day in interaction with RTM
• 1979-2013 (35 years)

Central America South Atlantic Central Asia

ATM OCN RTM ATM OCN RTM ATM OCN RTM

Models RegCM MITgcm HD Models RegCM MITgcm HD Models RegCM MITgcm HD

4.4.5.9 c63s 4.4.5.9 c63s 4.4.5.9 c63s
Res. 50 km, 23L 1/8º, 40L 0.5º Res. 50 km, 23L 1/8º, 40L 0.5º Res. 50 km, 18L 1/6º, 45L 0.5º
308x170 1050x540 global 330x206 800x510 global 170x216 276x408 global
ICBC ERA-Int 0.25º online ICBC ERA-Int 0.25º online ICBC ERA-Int 0.25º online
ERSST MOM ATM ERSST MOM ATM ERSST MOM ATM
Details CLM, Grell, KPP 10 major Details CLM, UW- KPP 7 major Details CLM, UW- KPP 30 major
KF rivers PBL, MIT rivers PBL, MIT rivers
• Coupled with RegESM driver • Coupled with RegESM driver • Coupled with RegESM driver
• 3 hours coupling interval between ATM and OCN • 3 hours coupling interval between ATM and OCN • 3 hours coupling interval between ATM and OCN
and 1 day in interaction with RTM and 1 day in interaction with RTM and 1 day in interaction with RTM
• ? • ? • 1979-2007 (25 years)
 RegESM Modeling System

1/ easy to use and extend 1/ only global conservation 1/ sharp gradient between
flexible modeling system is supported and might interactive and prescribed
2/ model components can have a problem for SST (issue #12)
be upgraded easily large domains 2/ no wind rotation
3/ state-of-art driver 2/ the bottleneck due to algorithm for Polar
design that follows common sequential RTM component Stereographic (POLSTR)
conventions / standards 3/ WAM uses 1d projection (issue #14)
4/ ready to use with new decomposition and limits
non-hydrostatic core higher number of processor
5/ supports both CLM and BATS for seq. type coupling
Plans: Short - Mid - Long
# Description Domains
1 • Using modeling system for different applications and domains Med. Sea.
• Future climate scenarios using CMIP5 models Black Sea
Caspian Sea
2 • New applications using hydrostatic core at higher spatial (3-12 km) Med. Sea.
and temporal resolution for Med-CORDEX-2, extreme events and
fast-moving processes
3 • Wave effect on current (WEC): 1) gradient of radiation stress tensor
or 2) vortex force (VF)
• Additional wave component such as WW3 to support curvilinear
grids in the wave component. It will allow to cover whole
atmospheric model domain
• Higher resolution river routing component for better representation
of rivers (i.e. Chym, CaMa-Flood etc.)
4 • Continuous Integration (CI)
• Standardization of model installation (integrate with Travis-ci
to test the build)
• Usage of virtualization technologies such as Docker containers
to run and test modeling system in the cloud (Google, Azure,
Amazon etc.)
5 • New approaches to analyze fast-moving processes in high res.
 https://developer.nvidia.com/index

Questions !!!
[email protected]
http://faculty.itu.edu.tr/turuncogl1/