Upload
sutton
View
22
Download
0
Embed Size (px)
DESCRIPTION
Uncertainties in the Climate Mean of Reanalyses, Observations, and the GFDL Climate Model. Thomas Reichler and Junsu Kim Univ. of Utah, Salt Lake City, USA. Supported by the Center for High Performance Computing, Univ. of Utah. - PowerPoint PPT Presentation
Citation preview
Thomas Reichler and Junsu Kim
Univ. of Utah, Salt Lake City, USA
Supported by the Center for High Performance Computing, Univ. of Utah
3rd WCRP International Conference on Reanalysis, Tokyo, Japan, 28 th January – 1st February 2008
Motivation
bb
Multi-variate model performance index
wor
sebe
tter
avg.
Some models outperform NCEP/NCAR reanalysis
Error
Models
Questions1. Why are NCEP/NCAR reanalyses not better
than freely evolving coupled model?
2. How do the other reanalyses do?
3. Can the reanalyses be improved?
4. How large are the observational uncertainties?
3
DataObservationsMany global datasets; often multiple
data for same quantity
Reanalyses NCEP/NCAR NNR
NCEP/DOE NDR
ERA40 ERA
JRA25 JRA
Model GFDL CM2.1 GFD
Base period ’79-’99 (in most cases)4
Climate Quantities
5
“Physics” (18)
“Dynamics” (13)
ERA-40 as reference
(2000-2005) (1985-1989) (1984-1999)
RSUT: ObservationsAnnual mean outgoing shortwave radiation TOA
2000-
<10 Wm-2 RMS error amongst different observations:= Observational uncertainty
Mean of different observations:= Best observational estimate
RSUT: Reanalysis vs. ObservationsRMS
24
22
19
15
12
Wm-2
Observational uncertainty (<10 Wm-2) is smaller than reanalysis error (20 Wm-2) Observational uncertainty is acceptable
Reanalyses and model show similar error patterns, independent of observations Errors are real Common biases due to similar physics?
What about other quantities?
RSUT Summary
Common Reanalyses Biases
Break-down by
• product NNR, NDR, ERA, JRA, GFD
• quantity physics - dynamics
• region NH, TR, SH
• season DJF, MAM, JJA, SON
• observation 1-5
Normalized RMS error:
Error Analysis
22
n nn
n n
r oNRMS w
“Physics”
UPPER
PH
YS
ICS
DY
NA
MIC
S
SURFACE CLOUDS / RADIATION
E
3 2 2 3 5 5 2 2 1 2 3 4 4 4 3 3 4 4
small largeNRMS-error
Validated against multi-observational
mean
OBS
MOD
REA
• Large uncertainties for surface fluxes• Largest errors for “clouds and radiation”• Model sometimes as realistic as reanalyses
“Dynamics”UPPER
PH
YS
ICS
DY
NA
MIC
SSURFACE CLOUDS / RADIATION
E
GLANN
small largeNRMS-error
Validated against ERA-40
MOD
REA
Lack of global observations Smaller differences than “physics” Model clearly not as close Except: meridional wind (MMC, VA) and
specific humidity (HUS)
Cumulative ErrorsPhysics•Largest errors over SH and during spring and summer•Model does quite well
Dynamics•JRA closest to ERA, NNR most different•Large model errors, in particular Tropics
Physics Dynamics
NNR
NDR
JRA
ERA
GFD
Conclusion1. Why do NNR not better
than models?
2. What about other reanalyses?
3. Room for improvement?
4. Observational uncertainties?
• Mostly “physics” quantities• Tuning, model physics,
forcings, data assimilation
• Common biases• Overall, ERA closest to
observations
• Yes, see 1.
• Large uncertainties in surface fluxes
Thanks
Global RMS errorsOBS ERA GFD
PR 0.8 1.6 1.3 mm/day
CLT 10 10 14 %
RLUT 6 12 11 Wm-2
RSUT 10 20 17 Wm-2
RLDS 13 10 12 Wm-2
RLUS 14 10 13 Wm-2
RSDS 17 28 20 Wm-2
RSUS 14 9 9 Wm-2
PSL 0.8 0.6 2.0 hPa
TAS 0.1 1.1 2.0 K
TAUU/V 1.6 1-2 1-2 10-2 Nm-2
HFLS 20 13 20 Wm-2
HFSS 7 3.5 7 Wm-2
Observational Uncertainties
uncertainity
error
Uncertainty Ratio:
How tolerable are errors?
<<1 tolerable