Transcript ppt - Zettaflops.org

Why Climate Modelers Think
We Need a Really, Really Big
Computer
Phil Jones
Climate, Ocean and Sea Ice Modeling (COSIM)
Climate Change Prediction Program
Co-PI SciDAC CCSM Collaboration
Climate System
Climate Modeling Goals
•
•
•
Understanding processes and how they
interact (only one on-going experiment)
Attribution of causes of observed climate
change
Prediction
– Natural variability (ENSO, PDO, NAO)
– Anthropogenic climate change (alarmist
fearmongering) – IPCC assessments
– Rapid climate change
•
Input on energy policy
Climate Change
IPCC TAR 2001
Greenhouse Gases
• Energy production
• Bovine flatulence
• Presidential campaigning
Rapid Climate Change
Polar and THC
State of the Art
•
•
•
•
•
•
T85 Atmosphere (150km)
Land on same
1 degree ocean (100km)
Sea ice on same
Physical models only – no biogeochemistry
5-20 simulated years per CPU day
– Limited number of scenarios
Community Climate System
Model
Land
Atmosphere
CAM
7 States
10 Fluxes
150km
NSF/DOE
Physical Models
(No biogeochem)
LSM/CLM
6 States
6 Fluxes
Once
per
per
Flux Coupler
hour
6 States
6 Fluxes
4 States
3 Fluxes
Once
7 States
9 Fluxes
6 States
13 Fluxes
day
per
Once
Once
6 Fluxes
Ocean
POP
100km
hour
11 States
10 Fluxes
per
hour
Ice
CICE/CSIM
Performance
Performance Portability
•
Vectorization
– POP easy (forefront of retro fashion)
– CAM, CICE, CLM
•
Blocked/chunked decomposition
–
–
–
–
Sized for vector/cache
Load balanced distribution of blocks/chunks
Hybrid MPI/OpenMP
Land elimination
Performance Limitations
•
Atmosphere
– Dynamics (spectral or FV), comms
– Physics, flops
•
Ocean
– Baroclinic, 3d explicit, flops/comms
– Barotropic, 2d implicit, comms
•
All
– timestep
Prediction and Assessment
Many century-scale
simulations (>2500yrs)
@~5yrs/day
Cycle vampires:
Many dedicated cycles
at computer centers
Attribution
“Simulations of the response to natural forcings alone … do not
explain the warming in the second half of the century”
“..model estimates that take into account both greenhouse
gases and sulphate aerosols are consistent with observations
over this*period” - IPCC 2001
Stott et al, Science 2000
The annual mean change
of temperature (map) and
the regional seasonal
change (upper box: DJF;
lower box: JJA) for the
scenarios A2 and B2
The annual mean change
of precipitation (map)
and
the regional seasonal
change (upper box: DJF;
lower box: JJA) for the
scenarios A2 and B2
If elected, we plan…
•
High resolution
– Cloud resolving atmosphere (10km)
– Eddy-resolving ocean (<10km)
– Regional prediction
•
Fully coupled biogeochemistry
– Source-based scenarios
•
More scenarios, more ensembles
– Uncertainty quantification
Towards Regional Prediction
Resolution and Precipitation
(DJF) precipitation in the California region in 5 simulations, plus
observations. The 5 simulations are: CCM3 at T42 (300 km), CCM3
at T85 (150 km) , CCM3 at T170 (75 km), CCM3 at T239 (50 km),
and CAM2 with FV dycore at 0.4 x 0.5 deg.
CCM3 extreme precipitation events depend on model resolution.
Here we are using as a measure of extreme precipitation events the
99th percentile daily precipitation amount. Increasing resolution
helps the CCM3 reproduce this measure of extreme daily precipitation
events.
Eddy-Resolving Ocean
Obs
2 deg
0.28 deg
0.1 deg
Only decades…
Chemistry, Biogeochemistry
•
Atmospheric chemistry
– Aerosols, ozone, GHG
•
Ocean biogeochemistry
– Phytoplankton, zooplankton, bacteria, elemental cycling,
trace gases, yada, yada…
•
Land Model
– Carbon, nitrogen cycling, dynamic vegetation
•
Source-based scenarios
– Specify emissions rather than concentrations
•
Sequestration strategies (land and ocean)
Aerosol Uncertainty
Atmospheric Chemistry
•
•
•
Gas-phase chemistry with emissions, deposition, transport and photochemical reactions for 89 species.
Experiments performed with 4x5 degree Fvcore – ozone concentration at
800hPa for selected stations (ppmv)
Mechanism development with IMPACT
–
–
–
A)
Small mechanism (TS4), using the ozone field it generates for photolysis
rates.
B) Small mechanism (TS4), using an ozone climatology for photolysis rates.
C) Full mechanism (TS2), using the ozone field it generates for photolysis rates.
Zonal mean
Ozone, Ratio A/C
Zonal mean
Ozone, Ratio B/C
Ocean Biogeochemistry
•Iron Enrichment in the Parallel Ocean
Program
•Surface chlorophyll distributions in POP
for 1996 La Niña and 1997 El Niño
Global DMS Flux from the
Ocean using POP
The global flux of DMS from the ocean to the atmosphere is shown as an annual mean.
The globally integrated flux of DMS from the ocean to the atmosphere is 23.8 Tg S yr-1
.
Increasing the deficit (1010-1012)
•
Resolution (103-105)
– x100 horiz, x10 timestep, x5-10 vert
•
Completeness (102)
– Biogeochem (30-100 tracers)
•
Fidelity (102)
– Better cloud processes, dynamic land, others
•
Increase length/number of runs(103)
– Run length (x100)
– Number of scenarios/ensembles (x10)
Storage
•
Atmosphere
– T85 29 GB/sim-yr, 0.08 GB/tracer
– T170 110 GB/sim-yr, 0.3 GB/tracer
•
Ocean
– 1 1.7 GB/sim-yr, 0.2 GB/tracer
– 0.1 120 GB/sim-yr, 17 GB/tracer
Beyond Moore’s Law
•
Algorithms
–
–
–
–
50% of past improvements
Tracer-friendly algorithms (inc remap advect)
Subgrid schemes
Implicit or other methods
Remapping Advection
• monotone
• multiple tracers free
• 2nd order
Subgrid Orography Scheme
•
•
•
Reproduces
orographic
signature without
increasing dynamic
resolution
Realisitic
precipitation,
snowcover, runoff
Month of March
simulated with
CCSM
Comparison of sea ice shear (%/day) from CICE (a,c)
and ‘old’ (b,d) models
(a)
(b)
Feb 20,
1987
(c)
(d)
Feb 26,
1987
Beyond Moore’s Law
•
New architectures
– Improved single-processor performance
– Scaling vs. throughput