Transcript Slide 1

The Met Office climate model HadSM3
and climateprediction.net
Michael Saunby
© Crown copyright 2007
25 January 2007
Page 1
The Hadley Centre for Climate Change
 Branch of the Met Office - UK national meteorological
service
 Opened by Prime Minister Mrs Thatcher in 1990
 Funded by DEFRA and MOD to research climate
change
 Moved from Bracknell to Exeter in 2004
 120 staff
© Crown copyright 2007
Page 2
climateprediction.net
University of Oxford, Rutherford Appleton
Laboratory, The Open University
Funded by NERC, DTI, EU and BBC
Goals
 Improve public understanding of the nature of
uncertainty in climate prediction.
 Harness the power of idle home and business PCs
to help forecast the climate of the 21st century.
© Crown copyright 2007
Page 3
The Unified Model - origins
 By the end of the 1980s the Met Office had developed
separate numerical prediction models to sustain
climate research and operational forecast capabilities.
 Each model had its own control, file and output
structure, as well as separate scientific formulations,
with separate teams responsible for development.
 UM reduces the effort required to implement models
on new hardware.
 CDC Cyber 205, Cray YMP-8, C90, T3E, NEC SX6,
SX8
© Crown copyright 2007
Page 4
Unified Model - today
 Met Office using NEC SX6 and SX8
 MPI introduced at 4.1 (T3E)
 Ported Unified Model (PUM). Cray, IBM, Clusters,
even single CPU
 Current release UM 6.3. HadGEM1
 UM4.7 HadCM3 etc. widely used, esp. ensembles
 C for I/O. Tcl/Tk for UI
© Crown copyright 2007
Page 5
Met Office supercomputing
SX-6
SX-8
CPU/node
8
8
Memory
32GByte
64GByte
Peak/node
64GFlops
128GFlops
3 clusters –
•15 x SX-6
•19 x SX-6
•21 x SX-8
© Crown copyright 2007
Page 6
Development environment
 Linux workstations (32 bit)
 Linux front-end and fileservers. (64 bit Intel)
 NEC SX6/SX8. Super UX. Vector processors
 Code management with FCM (Open Source) – uses
Subversion and lots of Perl
 Automated change from fixed to free format
 Last release ~750k lines Fortran
 ~250k lines changed!
© Crown copyright 2007
Page 7
The climate system
ATMOSPHERE
Terrestrial
radiation
Greenhouse gases and aerosol
Icesheets
snow
Clouds
Solar
radiation
Precipitation
Sea-ice
OCEAN
Biomass
LAND
© Crown copyright 2007
Page 8
Climate model development
HadGEM2
FLUME
UM structure
ancillary files
STASH/ diagnostics
HadGEM1
New
new
deep
convection
OPA
cloud
scheme
Ocean
N96 L50
atmosphere, ocean,
sulphur
Running on NEC
Regional model
N144
HI-GEM
Reading CGAM
N96 earth system
atmosphere, ocean,
sulphur, chem.,
carbon cycle, blackcarb
N48 earth system
atmosphere, ocean,
sulphur, chem., black carb
N48 L38 baseline
atmosphere, ocean,
sulphur, chemistry (some runs)
control - present/pre-industrial
Regional model
carbon cycle (TRIFFID, OCC),
FAMOUS GEM
low resolution
N96 L19
HadAM3H
Decadal
HadCEM
atmosphere, ocean
Seasonal (Glosea)
Climate variability
HadCM3
ENSEMBLES
QUMP
HadCM3
© Crown copyright 2007
HadCM3
atmosphere,
ocean, sulphur
HadCM3 & FAMOUS
various resolution
carbon cycle
chemistry
Page 9
CPDN volunteer computing challenges...
 Model is about 1 million lines of Fortran (40MB src)
 Proprietary, licenced by UK MetOffice
 distribute executable/binary form only
 Resolution used: 2.75x3.75 degrees (73 lat x 96 long)
 Typically run on a supercomputer (i.e. Cray T3E) or 8node Linux cluster (minimum)
 Ported to a single-processor, 32-bit Linux box
 Original: Windows only, now also Mac OS X, Linux
 Intel Fortran Win & Linux, IBM XLF for Mac, soon Intel
Mac
 Many validation runs made on single-proc/32-bit to
compare to supercomputer 64-bit
 Current coupled model takes ~6 months to run on a
P4/2GHz PC 24/7!
© Crown copyright 2007
Page 10
Volunteer Computing
 A specialized form of “distributed computing”
 Uses BOINC - Berkeley Open Infrastructure for Network Computing
 Was around before '99 but took off with SETI@home
 SETI@home peak cap with 500K users about 1 PF = 1000 TF
 for comparison Earth Sim in Kyoto = 35TF max
 climateprediction.net (CPDN) running at about 60 TF (60K concurrent
users each 1GF machine average, i.e. PIV 2GHz conservatively rated)
 Offers high CPU power at low cost
© Crown copyright 2007
Page 11
The experiments

Expt 1: Unified Model with simple, thermodynamic ocean. (HadSM3)

Aim: To identify parameter combinations which have little effect on the
mean climate but a large effect on climate sensitivity.
Double CO2
15 yr, 2 x CO2
Calibration
15 yr spin-up

Derived fluxes
15 yr, base case CO2
Diagnostics from final
8 yrs.
Control
Expt 2: Fully coupled model. (HadCM3)

Distribute pre-packaged simulations of 1950-2050.
Downweight or eliminate runs which compare badly with observations.
Re-distribute the surviving versions to simulate 2000-2050.

Estimate uncertainty from collated results and map the response manifold.


© Crown copyright 2007
Page 12
climateprediction.net users worldwide
>300,000 users total : ~60,000 active (currently running)
>19 million model-years simulated (as of January 2007)
~200,000 completed simulations
The world's largest climate modelling supercomputer!
(NB: a black dot is one or more computers running climateprediction.net)
© Crown copyright 2007
Page 13
climateprediction.net Screensavers
© Crown copyright 2007
Page 14
Example BBC experiment run
© Crown copyright 2007
Page 15
Climate sensitivities from climateprediction.net
The frequency distribution
of simulated climate
sensitivity using all (2,578)
model versions (black), all
model versions except
those with perturbations to
the cloud-to-rain conversion
threshold (red), and all
model versions except
those with perturbations to
the entrainment coefficient
(blue).
Sensitivity is the equilibrium
response of the global
mean temperature of
doubling atmospheric levels
of carbon dioxide.
Stainforth et al, Nature, 27 Jan ‘05
Frequency Distribution of Simulations
From Stainforth et al, Nature, 27 Jan ‘05
© Crown copyright 2007
Page 17
climateprediction.net recent publications










D. A. Stainforth, T. Aina, C. Christensen, M. Collins, N. Faull, D. J. Frame, J. A. Kettleborough, S. Knight, A. Martin, J.
M. Murphy, C. Piani, D. Sexton, L. A. Smith, R. A. Spicer, A. J. Thorpe & M. R. Allen, Uncertainty in predictions of the
climate response to rising levels of greenhouse gases, Nature, 433, pp.403-406, 27/01/2005
D. J. Frame, B. B. B. Booth, J. A. Kettleborough, D. A. Stainforth, J. M. Gregory, M. Collins, and M. R. Allen,
Constraining climate forecasts: The role of prior assumptions, Geophysical Review Letters, 32, L09702, May 2005.
C. Piani, D. J. Frame, D. A. Stainforth, and M. R. Allen, Constraints on climate change from a multi-thousand member
ensemble of simulations, Geophysical Review Letters, 32, L23825, December 2005.
G.C. Hegerl, T.J. Crowley, W.T. Hyde and D. J. Frame, Climate sensitivity constrained by temperature reconstructions
over the past seven centuries, Nature, 440, p1029-1032, April 2006.
Allen, M., N. Andronova, B. Booth, S. Dessai, D. Frame, C. Forest, J. Gregory, G. Hegerl, R. Knutti, C. Piani, D.
Sexton, D. Stainforth, 2006, Observational constraints on climate sensitivity, in Avoiding Dangerous Climate
Change, (Eds.) J.S. Schellnhuber, W. Cramer, N. Nakicenovic, T.M.L. Wigley, G. Yohe., Cambridge Univ. Press.PDF
of complete book (17 MB), see chapter 29
Knutti, R., G.A. Meehl, M.R. Allen and D. A. Stainforth, Constraining climate sensitivity from the seasonal cycle in
surface temperature, Journal of Climate, in press
Designing a Runtime System for Volunteer Computing, David P. Anderson, Carl Christensen and Bruce Allen, To
appear in Supercomputing ’06 (the International Conference for High Performance Computing, Networking, Storage
and Analysis), Tampa, Florida, USA, November 2006.
N. Massey, T. Aina, M. Allen, C. Christensen, D. Frame, D. Goodman, J. Kettleborough, A. Martin, S. Pascoe and D.
Stainforth, Data access and analysis with distributed federated data servers in climateprediction.net , Advances in
Geosciences, 8, p49-56, 2006.
Carl Christensen, Tolu Aina, David Stainforth, The Challenge of Volunteer Computing With Lengthy Climate Modelling
Simulations, Proceedings of the 1st IEEE Conference on e-Science and Grid Computing, Melbourne, Australia,
5-8 Dec 2005
David Stainforth, Andrew Martin, Andrew Simpson, Carl Christensen, Jamie Kettleborough, Tolu Aina, and Myles
Allen, Security Principles for Public-Resource Modeling Research, Proceedings of the 13th IEEE Conference on
Enabling Grid Technologies (ENTGRID), Modena, Italy, June 2004
© Crown copyright 2007
Page 18
More information
 http://www.bbc.co.uk/sn/climateexperiment/
 http://www.climateprediction.net/
 http://www.metoffice.gov.uk/research/hadleycentre/
© Crown copyright 2007
Page 19
Thanks to
Carl Christensen
Matt Collins
Gareth Jones
Jamie Kettleborough
Paul Selwood
© Crown copyright 2007
Page 20