Regional Earth System Model (RegESM) using NUOPC/ESMF Ufuk Turuncoglu (1,2) (1) Informatics Institute, Computational Science and Engineering, ITU, Turkey (2) ESP Section, ICTP , Italy The Third Workshop on Coupling Technologies for Earth System Models, 20-22 Apr. 2015
30
Embed
Regional Earth System Model (RegESM) using … · Outline • Evolution of the RegESM modeling system • Basic Design • Common Problems and Solutions – Unaligned land-sea masks
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Regional Earth System Model (RegESM) using NUOPC/ESMF���
���Ufuk Turuncoglu (1,2)���
���������
(1) Informatics Institute, ���Computational Science and Engineering, ���
ITU, Turkey���(2) ESP Section, ���
ICTP, Italy
The Third Workshop on Coupling Technologies for Earth System Models, 20-22 Apr. 2015
Outline• Evolution of the RegESM modeling system• Basic Design• Common Problems and Solutions
Why do we need new model?• To gain experience about design and use of coupled modeling
systems– The applications of coupled modeling systems getting increased in
last decade along with the rapid development in HPC– Need to consider different components of the climate system
• To design easy to use and extend modeling system– ENEA’s (Italian Energy Agency) PROTHEUS system was based on
MITgcm+RegCM3+OASIS– It is not easy to use, extend and upgrade– It uses multiple executable, no online interpolation support
• To support Med-CORDEX and other applications …– Set of coordinated experiments to better understand
Mediterranean climate: standalone and coupled RCMs– It would be good to have another regional coupled modeling
system that uses different model components (ensemble)
Evolution of Regional Earth System Model (RegESM)• The first prototype version is created in 2012 @ICTP
– No driver– RegCM was hosting other components of the coupled model– Single ocean component was supported (ROMS-Ice)– Poor energy conservation (for exchange fields)– Applied to Caspian Sea (Turuncoglu et al., 2013, GMD)– Hard to extend (i.e. adding new components as wave, river routing)
• Then, more generic version is designed and released in 2013-… @ITU– Centralized driver using ESMF’s NUOPC layer (via connectors)– All the components plugged into driver (less dependence to the model
component code)– Support for two different ocean model (ROMS and MITgcm)– Energy conservation is improved (custom global conservation using bilinear)– Support for unaligned land-sea masks– Applied to Mediterranean, Black Sea and Caspian Sea (ITU)– More applications are on the way - Caribbean (ICTP), High Res.
Mediterranean (ICTP and ENEA), Indian Ocean, Antarctica (ITU)
RTM:Max Planck’s HD (mod. 1.0.2) special thanks to ���Prof. Stefan Hagemann
WAV:ECMWF’s WAM (4.5.3 MPI)
DRV:RegESM (7.0.0b38)
atmosphere(RegCM)
land(BATS)
ocean(ROMS / MITgcm)
sea iceunnamed
driver
land(CLM)
river routing
(HD)
initial and boundaryconditions
anthropogenicemissions
naturalemissions
full gaschemistry
aerosol
wave(WAM)
initial conditions
initial conditions
ESMF+NUOPC
in progress ready and tested
atmospheric���chemistry
Two differentland surface ���
model
Two differentoceanmodel
representation is barrowed from ���Alexander and Easterbrook
ESMF/NUOPC in RegESM• Model components are coupled using NUOPC “connectors”
– AAA• The modeling system has single executable
• Easy to use and debug• Library for all components (i.e. libatm.a)
• The connector uses all PETs of the ���active model components
• NUOPC layer help to manage:• Execution order of the components• Synchronization (i.e. slow vs. fast coupling���
time steps) • Exchange fields (via field table)• …
Run Sequence in ESMF/NUOPC• The NUOPC Layer is capable of implementing many different
coupling schemes– Explicit:
• Exchange data in same time. Two ���connector (i.e. backward and ���forward)
– Explicit with slow and fast ���clock support:• Different interaction time ���
step among the components– Semi-implicit (leap-frog style ���
interaction)– Fully-implicit
• Complex interaction among the���components.
ATM
OCNt=0h t=1h ...
ATM-OCNOCN-ATM
coupling time
Ex: 1 hour
EXPLICIT RUN SEQUENCE
Initializationof models
ATM
OCNt=0h t=1h ...
SEMI-IMPLICIT RUN SEQUENCE
ATM
OCNt=0h t=1h ...
IMPLICIT RUN SEQUENCEt=1/2 h
2
34 5
1 2
1
2
34
2
1
2
34
2
34 5
4
3
1
3
4
data exchangemodel integrate
x2
16
16
7 7
Run Sequence in RegESM• The RegESM uses explicit coupling schemes along with the
support of fast and slow time steps.– Fast interaction among ATM, OCN and WAV (i.e. 1 or 3 hours)– Slow interaction between ATM and RTM, RTM-OCN (i.e. 1 day)– Semi-implicit type coupling is also implemented.
i.e. three component case (ATM+OCN+RTM) + explicit coupling
Special thanks to ESMF Group (especially to Gerhard Theurich) for their support and help
Sequential vs. Concurrent Execution• RegESM also supports two different approach to run the model
components.– Sequential: model components are run in order– Concurrent: all models are active at same time (it does not allow
overlapping of the used PETs – cores / CPUs)• User able to assign last PET or individual PET to RTM (seq.)
i.e. three component case (ATM+OCN+RTM)
Managing Exchange Fields• Exchange fields between model components are defined by
extra configuration file (exfield.tbl)
• The user can chose the exchange fields from the field pool
extrapolation is activated conservation ��� is activated
add_offset and���scale_factor used for���
unit conversion
interpolation ���type stencil for source���
and destination grids���used in interpolation
exchange fields(set of pre-defined���fields can be used)
Unaligned Land-Sea Masks• The unaligned land-sea mask might produce unrealistic heat
fluxes around coastlines• It is more apparent when horizontal resolution differences are
high among model components (i.e. 50 km ATM, 8 km OCN)• The interpolation must be performed only over sea for ATM-
OCN components or only over land for other interactions• Possible solutions:
– Manual editing of land sea masks• It is not generic solution and must be repeated when horizontal ���
grids and application are changed L– Two-step interpolation
• It is generic and independent from application J• It might produce high interpolation error around coastlines and
1. Interpolate from ATM to OCN���using bilinear type interpolation. Use only sea grid points
2. Use result of previous step (step 1) and interpolate data from OCN to OCN from mapped grid points ���to unmapped ones (over sea) using���nearest-neighbor type regridding.
3. Merge results of step 1 and 2 to���create result field
Interpolation Error• Pseudo spherical harmonics (L=32, M=16) from SCRIP
• The error increases in the regions that has complex land-sea mask and when the resolution difference is high
• The error is also depend on the field itself and its behavior. ���More tests with realistic fields are needed !!!
f = 2+ sin16 (2θ )cos(16φ) ⇒ θ = lat, φ = lon
50 km ATM 1/12° (~8km) OCN
Interpolation Error …• Two step interpolation (ATM to OCN)
Min: -0.207���Max: 0.071
Min: -0.025���Max: 0.014
LRES
ATM
- 50
km
HRE
S ATM
- 10
km
Relative Error = (Mod/Obs)-1
• High error around coastline���especially Azov Sea
• The error pattern might be���different with different field
• Errors are reduced when it is compared with 50 km ATM case
• Keep the resolution difference minimal !!!
RTM Grid
OCN Grid
River
R
unaffectedgrid points
effectedgrid points
1
2
3
1 Find river location (i,j) in RTM grid
2 Find closest OCN model grid to RTMgrid
3 Find OCN grids in effective radius (R)
Algorithm
Interaction between OCN and RTM• How we define rivers in ocean models?
– As source / sink points– As surface boundary conditions (BC) such as rain or E-P
• The RegESM support both of them– Fully automatized algorithm���
to distribute river discharge���as surface BC
– Also, possible to change���effected grids as a function of ���river discharge !!!
river position as grid coordinates (i,j) or geographic coordinates (lat,lon)
represent as constant monthly values or RTMeffective radius
Interaction between OCN and RTM• How we define rivers in ocean models?
– The Max Planck’s HD model has 0.5x0.5 fixed global grid– The temporal resolution is also fixed (1-day)– Only major rivers are represented – It calculates river discharge using surface+sub-surface runoff ���
if the grid point has data– The surface+sub-surface runoff came from ATM
salt from ATM+OCN+RTM
Domain specific ���routing algorithms ���might be ���implemented to represent rivers more accurately?
active monthly values
Do we really need wave coupling?• The RegCM model generally overestimates the wind speed over
the sea due to poor representation of air-sea interface
• In general, the standalone atmosphere model tends to overestimate the wind speed but it is more dominant in winter season
• It also effects the circulations in the ocean components
Observational ���data is satellite ���derived 0.25° ���resolutionBlended Sea ���Winds ���(Zhang et al., 2006)
Parameterization of Ocean Fluxes in RegCM• The model supports three different scheme to represent ���
ocean fluxes– BATS1e
• It uses Monin-Obukhov Similarity Theory (MOST). – Zeng et al., 1998
• Widely used parameterization implemented in RegCM 4.X that allows more sophisticated representation of air-sea interface
– COARE bulk flux algorithm (experimental)• It is implemented for ocean-atmosphere coupling. It also includes
representation of fluxes for sea-ice condition• The latest version of the RegCM model (4.4.X) also includes
slab ocean model– Based on GFDL FMS Slab ocean model (or mixed layer model)
• islab_ocean =1
Parameterization of Ocean Fluxes in RegCM• Representation of surface roughness length of momentum (z0)
in Zeng Ocean Flux Algorithm (Zeng et al, 1998)
iocnrough = 1
iocnrough = 2
• Roughness and drag coefficient increases with wind speed
zo = βu*2
g0.0065 ���
(Charnock’s relation, 1955)
Friction velocity
Acceleration of gravity
zo = βu*2
g+α
νa
u*0.013 0.011
Kinematic viscosity of ���dry air (Andreas, 1989)
Smooth flow contribution ���at low wind speeds
(Smith, 1988)
CD =u*U(z)!
"#
$
%&
2
U(z) = u*κ log(z z0 )
Horizontal ���wind speed
u*2 =
τ aρa
Wave Coupling• ECMWF’s wave model (WAM) is selected• WAM is a 3rd generation model
– The model is slightly modified for ESMF coupling• Different coupling types?
– Depends on used model components and application– It is better to support as much as possible combination– Need to add threshold for roughness length in ATM for stability
ATM
WAV
Wind or stress
Charnock
ATM
WAV
Wind Roughness Length
ATM
WAV
Wind Phase Speed + Mean Wave
Dir.
ATM
WAV
Wind Sig. Wave Hgt. + Wave Length
Domain Decomposition in WAM• 2D domain is mapped onto 1D array of following increasing
latitude lines
• It seems that it is a common convention in wave modeling (i.e. WAVEWATCH-III) but ESMF does not support yet …
• Effect of number of component: ATM-OCN / ATM-OCN-RTM
• Also need to test concurrent case.• RTM component could be tightly coupled with ATM model• The current coupled model uses connector. • It is hard to find optimum PET distribution in concurrent case.
More tests are needed …
Benchmark Results …
Last PET is shared ���between OCN and ���RTM
Having additional ���component for RTM ���reduces performance
Repository and Project Home• RegESM project is hosted by GitHub���
https://github.com/uturuncoglu/RegESM
• The user documentation is ready– Basic model design– Installation of libraries and model ���
components– Definition of configuration files– Known bugs and limitations– Bugs:
• Handled by GitHub• Model is tested in different���
architectures (IBM Blue Gene)
Future Plans• The modeling system needs more accurate (+ parallel ?) river
rooting component– RAPID (The Routing Application for Parallel computation of
Discharge), https://github.com/c-h-david/rapid/• Ocean + wave interaction is still missing. It must be defined ���
for both ROMS and MITgcm cases• Support for ESMF’s high order conservative regridding• Making modeling system self-describing (via ESMF Attributes)
and integrating with scientific workflows (SWf)• We need more application to test the modeling system• Build a user community
– ENEA, ICTP @ Italy + ITU @ Turkey– Training events related with coupled model design and use …
This study has been supported by a research grant (113Y108) provided by The ScienPfic and Technological Research Council of Turkey (TUBITAK) and partly by The Abdus Salam InternaPonal Center for TheorePcal
Physics (ICTP) Associateship Scheme. The compuPng resources used in this work were provided by the NaPonal Center for High Performance CompuPng of Turkey (UHEM) under grant number 5003082013. We also
acknowledge PRACE for awarding us access to resource CURIE based in France at Très Grand Centre de Calcul (TGCC) under project number 2010PA2442.