Preliminary Results of Global Climate Simulations With a High- Resolution Atmospheric Model P. B. Duffy, B. Govindasamy, J. Milovich, K. Taylor, S. Thompson,
Post on 16-Jan-2016
212 Views
Preview:
Transcript
Preliminary Results of Global Preliminary Results of Global Climate Simulations With a High-Climate Simulations With a High-Resolution Atmospheric ModelResolution Atmospheric Model
P. B. Duffy, B. Govindasamy, J. Milovich, P. B. Duffy, B. Govindasamy, J. Milovich, K. Taylor, S. Thompson, M. WehnerK. Taylor, S. Thompson, M. Wehner
Lawrence Livermore National LaboratoryLawrence Livermore National Laboratory
With help fromWith help fromJ. J. Hack, NCARJ. J. Hack, NCAR
Pushing the limits of climate model resolution:• Why do it?• What have we done?• What have we learned?
ContentsContents
Why Try to Increase Model Resolution?Why Try to Increase Model Resolution?
To allow global climate models to give meaningful results on regional scales.
• This will allow assessments of societal impacts of climate change
To improve simulations of terrestrial Carbon cycle• Results sensitive to precip. and surface temperature
Higher resolution• Includes fine-scale detail• Improves results on large scales
What Are Our Goals?What Are Our Goals?
Assess computational issues associated with running the model at high resolution;
Evaluate the realism of the simulated climate at high resolution vs. coarse resolution;
Examine resolution dependence of atmospheric response to increased greenhouse gases.
Pushing the limits of climate model resolution:• Why do it?• What have we done?• What have we learned?
ContentsContents
Six High-Resolution Simulations are Six High-Resolution Simulations are Complete or in ProgressComplete or in Progress All simulations use the CCM3 atmospheric model forced by prescribed
sea-surface temperatures (SSTs)
Analagous T42 simulations also have been performed
1. A present-climate simulation at T170 (512 x 256 cells)• Forced with climatological observed sea-surface temperatures• 12 simulated years completed
2. An increased GHG simulation at T170• Forced with predicted SSTs for 2100 based on SST CHANGES
from a coarse-resolution coupled model (CSM) simulation• 12 simulated years completed
Both of above use “untuned” (i.e. tuned for T42) model
• 3. A present climate simulation at T170 with “tuned” model• 6 simulated years completed
Simulations Complete or in Progress…Simulations Complete or in Progress…
• 4. An AMIP simulation at T239 (720 x 360 cells)• Forced with observed sea-surface temperatures for 1979-
1987• 6 years completed with untuned model
5. A present-climate simulation at T239• Uses model tuned for T170• ~1 simulated year completed so far
n 6. A present-climate simulation at T340 (!) (1024x512 grid cells)• This is short (1 simulated month)
Observations on Some Observations on Some Computational IssuesComputational Issues
We Have Run the High Resolution Model We Have Run the High Resolution Model on A Variety of Machineson A Variety of Machines
Lab Machine Vendor Model # nodes
CPUs/node
CPU type
Notes
LLNL White IBM RS/6000 SP
512 16 (NH-2)
Power 3 375 Mhz
World’s # 1
LLNL Frost IBM RS/6000 SP
68 16 (NH-2)
Power 3 375 Mhz
NERSC gseaborg IBM 4
NERSC seaborg IBM RS/6000 SP
208 16 Power 3 375 Mhz
World’s # 2
LLNL TC2000 Compaq 118 4 EV67 667 Mhz
LLNL PCR ??? 128? 2 Pentium4
Linux cluster
Computational Issues…Computational Issues…
• speedup curve on NERSC gseaborg machine
A Preliminary Look at Selected Results:A Preliminary Look at Selected Results:
Simulations of Present ClimateSimulations of Present Climate
€
S=
1N
fn −f( )− rn −r( )[ ]2
n=1
N
∑⎧ ⎨ ⎩
⎫ ⎬ ⎭
12
σrσ f
€
θ =
1N
fn− f( ) rn−r( )n=1
N
∑σfσr
€
σf =fn− f( )
2
n=1
N
∑N−1
€
R=σf
σr
Example“Taylor Diagram”
Result of ideal model would be plotted here
Two Points About Our Taylor DiagramsTwo Points About Our Taylor Diagrams
1. Taylor diagrams do not show errors in means (I.e. biases)
2. Comparison was performed on T42 grid
Thus, we are assessing how finer resolution affects large-scale results
Variables and primary validation data setsVariables and primary validation data sets
Variable Abbrv. Observation Comment
Surface air temperature
Outgoing longwave
Clear-sky outgoing LW
Outgoing shortwave
Clear-sky outgoing SW
Cloud fraction
Geopotential height
Precipitable water
Sea level pressure
Air temperature
Winds
Precipitation
Sensible heat flux
Latent heat flux
Zonal wind stress
Meridional wind stress
TAS
OLR
LWCS
SW
SWCS
CLT
Z
PRW
PSL
T
U, V
P
SH
LH
TAUU
TAUV
Jones/PCMDI
ERBE
ERBE
ERBE
ERBE
ISCCP
ECMWF
ECMWF
ECMWF
ECMWF
ECMWF
Xie & Arkin(NCEP)
COADS
COADS
COADS
COADS
merger of available data
Nearly global coverage bysatellites (1985 -1988)
best coverage available(1984-1988)
reanalysis
merged land and satellitedata
empirically calculated,based on available data
Effects of Increasing Resolution From T42 -> T170
Model Improvement: AMIP 1 vs. AMIP 2Model Improvement: AMIP 1 vs. AMIP 2
CLT cloud fraction
P precipitation
PSL sea level pressure
PRW precipitable water
SH sensible heat
T200 200 hPa temperature
τu zonal wind stress
τv meridional wind stress
U200 200 hPa zonal wind
V200 200 hPa meridional wind
Z500 500 hPa geopotential height
T42 -> T170 AMIP 1 -> AMIP 2
Increasing Resolution vs. Actual Thinking
Effects of Increasing Resolution From T170->T239
old
Effects of Tuning on Spatial Patterns of Results at T170
old
Tuning Reduces BiasesTuning Reduces Biases
Precipitation Over USAPrecipitation Over USA
DJF Precipitation over USA gets Better as DJF Precipitation over USA gets Better as Resolution IncreasesResolution Increases
JJA Precipitation Over USAJJA Precipitation Over USA
As Resolution Increases, Convective As Resolution Increases, Convective Precipitation Decreases…Precipitation Decreases…
… … and Large-Scale Precipitation Increasesand Large-Scale Precipitation Increases
Arctic Climate and Sea IceArctic Climate and Sea Ice
Many Climate Models Simulate Arctic Many Climate Models Simulate Arctic SLP Distribution Poorly…SLP Distribution Poorly…
JJA sea level pressure distribution is more realistic at T170 than at T42
Effects of Increased Effects of Increased Greenhouse GasesGreenhouse Gases
We Calculate SST change from Coarse-We Calculate SST change from Coarse-Resolution Coupled Model (CSM) SimulationResolution Coupled Model (CSM) Simulation
Simulated Time
Sea
Sur
face
Tem
pera
ture
1990s 2090s
Regional Details of Predicted Climate Regional Details of Predicted Climate Changes Can be Very Different at T170Changes Can be Very Different at T170
T42
QuickTime™ and aCinepak decompressor
are needed to see this picture.
T170
QuickTime™ and aCinepak decompressor
are needed to see this picture.
T239
QuickTime™ and aCinepak decompressor
are needed to see this picture.
What Have We Learned?What Have We Learned? It is possible to run short global climate simulations at 50 - 75 km
resolution.
CCM3 is reliable but not efficient at these resolutions;• Eulerian spectral dynamics seems to run inefficiently at high
resolution.
Physics parameterizations seem robust to an increase in resolution• Most likely, retuning, but not reformulation, is needed.
In most aspects, large-scale results are more realistic at T170 than at T42; T239 is even better. I.e. using high resolution improves the large-scale results.
Partial re-tuning of cloud and hydrological parameters removes biases but has little effect on spatial patterns.
Climate changes due to increased greenhouse gases at T170 vs. T42 are very similar globally, but quite different regionally.
What’s Next? With More Funding, We CouldWhat’s Next? With More Funding, We Could
Work with the NCAR Environmental and Societal Impacts Group (ESIG) to design simulations of maximum benefit to the impacts community
• Use appropriate CO2 scenarios
• Save needed climate statistics• Distribute output to the community;
“Downscale” to fine resolution results of climate forecasts to be performed by NOAA;
Drive a 20 km nested model with output from our global T170 model;
Perform Initial Tendency Error Analysis (ITEA) at high resolution;
Compare results of high-resolution global model to those of a nested model at same resolution;
What’s Next?…What’s Next?…
Run the new NCAR model (CAM1) at high resolution• Investigate computational behavior (scaling, etc.)• Evaluate simulated climate;
Perform a short coupled-model simulation with the atmosphere at high resolution;
Can high-resolution atmosphere improve simulation of Arctic sea ice?
• Force sea ice model with winds, etc. from high-resolution atmospheric model;
Does higher resolution improve simulation of terrestrial C-cycle?
Go to even higher resolution
Etc.
We will make model output available for analysisWe will make model output available for analysis
If interested, contact me:
pduffy@llnl.gov
top related