predictability from strat-trop coupling in reforecasts and the...
TRANSCRIPT
Short-Term Climate Predictability
Associated With Stratospheric Influences
in Operational Forecast Systems
Thomas Reichler, Junsu Kim (Univ. Utah)
and
Arun Kumar (NOAA/NCEP/CPC)
Predictability from Strat-Trop Coupling in
Reforecasts and the GFDL Climate Model
Thomas Reichler1, Junsu Kim1, and Arun Kumar2
(1 Univ. of Utah, 2 NOAA)
Supported by the Center for High Performance Computing, Univ. of Utah
AGU Chapman Conference on The Role of the Stratosphere in Climate and Climate Change
Santorini, Greece, 24-28 September 2007
Questions & Method
• To what extent do current prediction systems
exploit stratospheric signals?
• How much practical skill is associated with it?
• Does an improved stratospheric component
increase skill?
2
I. Retrospective forecasts
II. Climate model simulations
GFDL AM2NCEP MRF
L28
p (
hP
a)
he
igh
t (k
m)
“high-top”“low-top”
Models: Vertical Structure
NCEP/GFS GFDL AM3
L28 LOW HIGH
“low-top”
7 3 18
3
2.5 hPa 3 hPa
0.02 hPa
Part I
CDC Reforecasts
• Fixed 1998 version of NCEP/GFS, T62 L28
• 1979 - present
• Daily forecasts out to day 15
• 15-member ensemble
• Hamill et al. (2006, BAMS)
Daily forecasts, capture each SSW event
Only out to day 15, miss some of the interesting
action at later times
No stratospheric data
NOAA/CDC Reforecasts
5
Annual mean RMS error in SLP over Northern
Extratropics between reforecasts and reanalysis
Reforecast Climatology Drift
Day 1
4
13
JFM AMJ JAS OND
SLP errors
6
• Ensemble means
• Focus: Northern Annular Mode (NAM)
• Sea level pressure (SLP), and Z850, 700,
500, 250, and 150
• Reference: NCEP/NCAR reanalysis, 1979-
2007
Validation Strategy
7
A. Downward
Propagating Features
Case Study: Winter 2004
JAN1 FEB1 MAR1
Day 8.5 NAMSLPDay 14.5 NAMSLP
Day 0.5 NAMSLP
9
NAMSLP
NCEP/NCAR reanalysis
NAM index
by level and time
Strong Events
• NCEP/NCAR reanalysis:
1979-2007
• NAM10<-3 → 16 events
• Discard 6 events which
exhibit no downward
propagation
• Remain with 10 strong
downward propagating
events
10
Day 15 Forecasts (EM)Reanalysis
Days after event
SSW Composite
10 strong events
NCEP/NCAR Reanalysis
11
Day 15
Day 0
Day 10
Day 5
SSW
Composite
12
CDC reforecasts
Corr
ela
tion
Lagged Correlation NAMSLP
NAM10/SLP
Evolution
NAM10 Reanalyses
Ind
ex
Days after SSW event
Reforecasts day 15
Composite NAMSLP Index
61 days centered
at day 19
Reforecast signal
reaches surface
delayed by ca. 10
days
Reanalyses
B. NAMSLP Predictability
Basic PredictabilityACCSLP vs. NAMSLP
ACCSLP
NAMSLPWinter:NAMSLP
15
NAMSLP Predictability
Da
y o
f ye
ar
Lead time (day)
r
16
• 1979-2007, all days
• Unsmoothed
(shading)
• 31 day running
means (contours)
• Sharp decline in
spring; maybe
related to breakdown
of polar vortex
• Gradual increase in
fall
Correlation: NCEP/NCAR reanalysis vs. CDC reforecast
r=0.6
Correlation AnomaliesCorrelations
Correlation between
reanalysis and
reforecasts
31 day unsmoothed
NAMSLP index
Composite of 10 SSW
events (NAM10 < -3)
Anomalies: February
climatology removed.
NAMSLP
Predictability
During SSWs
0
t [days]
AO index
-10
-20
-30
10
20
30
17
SSW
0
Conclusion I
Despite low vertical stratospheric resolution (7
levels), reforecasts exhibit clear downward
propagating signals.
SSW signals arrive too late at surface; delay
increases with lead time.
Nevertheless, SSW events improve AO
predictability 0-15 days after the events and at
leads of 8-15 days.
Beside the 10 SSW events, additional skill from
stratospheric influences may be realized during
other warm events or during cold events.
Part II
AGCM Experiments
Basic simulation features and sensitivities to forcings
• Very similar to IPCC-AR4 version
• Non-orographic gravity wave drag (Alexander and
Dunkerton 1999)
• RAS deep convection
• Resolution• Horizontal: N45 (144x90)
• Vertical: L24 and L48 (LOW and HIGH)
• LOW and HIGH have identical physics! The only
difference is number and location of vertical levels
20
GFDL AM2/3 Nalanda
A. Mean Climate
and Variability
AMIP-type runs forced with observed boundary
conditions and atmospheric composition (1983-2003)
1983-2001L24NO / L48NO
Mean Temperatures
LOW HIGH
Strato -0.7 / 7.8 2.3 / 3.2
Tropo -0.9 / 2.1 -0.5 / 2.3
L24 L48
Strato -0.7 / 9.8 2.8 / 4.5
Tropo -0.6/ 1.9 -0.2 / 1.8
(mean bias / rms error) [K]
JJA
DJF
ERA LOW-ERA HIGH-ERA
22
1983-2001L24NO / L48NO
Mean U-Winds
(mean bias / rms error) [m/s]
LOW HIGH
Strato -1.2 / 13 -1.3 / 5.7
Tropo -0.2 / 1.5 0.0 / 1.7
LOW HIGH
Strato -4.7/ 20 -4.5 / 14
Tropo -0.0/ 1.7 0.3 / 2.1
ERA LOW–ERA HIGH–ERA
JJA
DJF
23
1983-2001L24NO / L48NO
Interannual Variability U-Wind
JJA
DJF
ERA LOW HIGH
(mean bias / rms error) [m/s]
LOW HIGH
Strato -1.8 / 2.7 -1.1 / 2.3
Tropo 0.0 / 0.3 0.1 / 0.5
LOW HIGH
Strato -2.2/ 4.4 -0.9 / 4.3
Tropo 0.0/ 0.2 0.2 / 0.3
24
LOW HIGH
• [u] and
EOF [u]
• 1983-
1999
• L24/48
NO
ERA
MAM
DJF
JJA
SON
25
SAMu
NAMu
MAM
DJF
JJA
SON
LOW HIGHERA40
26
NAM/SAM Timescales
Reanalysis
LOW
HIGH
NAM SAM
MonthP
ressu
re
NAM10 Lagged Correlations
• DJFM, 1983-2003
• 61 day time series
• 21 day averages
• L24/48NO
HIGH
LOW
Reanalysis
28
Baldwin & Dunkerton CompositesReanalysis LOW
HIGH• 1983-2003
• Reference EOFs from
reanalysis
• Composite on NAM10 < -3
• Reanalysis: 12 events
• Low: 9 events
• High: 12 events
B. Equilibrium Climate
Simulations
Experimental Setup
31
• 20-40 year long AMIP type simulations, const. forcing
• Change only one forcing at a time:
• Varying SSTs: Primarily test stratosphere’s response
to tropospheric climate change
• Frozen SSTs: Primarily test troposphere’s response
to stratospheric climate change
O3 CO2 SST Aerosol Sun Levels
1950 280 1950s 1950 1950 LOW
2000 380 2000s HIGH
720 A1B
1120 4xCO2
=CTRL
Sensitivty to O3 (SON)
Δu
90S 60S 30S Eq 30N 60N 90N 60S 30S Eq 30N 60N 90N
60S 30S Eq 30N 60N 90N
LOW HIGH
[m/s]
[K]
32
ΔT
O3
2000 minus1950
90S 60S 30S Eq 30N 60N 90N
-2.8 ppm
90S 60S 30S Eq 30N 60N 90N
Sensitivity to CO2 (JJA)
LOW HIGH
1120 ppm minus CTRL720 ppm minus CTRL380 ppm minus CTRL
ΔT
Δu
-90 -60 -30 Eq 30 60 90
33
[m/s]
[K]
-90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90
LOW HIGH LOW HIGH
Tropical Expansion
• Change in width of Hadley Cell, derived from
mean meridional mass streamfunction
Degre
es latitu
de
LOW
HIGH
O3 CO2 CO2+O3 SST SST+CO2 SST+
O3+CO2
SST+O3
-0.6
-0.4
-0.2
-1E-15
0.2
0.4
0.6
0.8
1
1.2SS
T50
O3_
SST5
0
CO2_
SST5
0
CO2_
O3_
SST5
0
2xCO
2_SS
T50
2xCO
2_O
3_SS
T50
4xCO
2_SS
T50
4xCO
2_O
3_SS
T50
SST2
k
SST2
k_a
O3_
SST2
k
O3_
SST2
k_a
CO2_
SST2
k
CO2_
O3_
SST2
k
L24 SAM DJF
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2SS
T50
O3_
SST5
0
CO2_
SST5
0
CO2_
O3_
SST5
0
2xCO
2_SS
T50
2xCO
2_O
3_SS
T50
4xCO
2_SS
T50
4xCO
2_O
3_SS
T50
SST2
k
SST2
k_a
O3_
SST2
k
O3_
SST2
k_a
CO2_
SST2
k
CO2_
O3_
SST2
k
L48 SAM DJF
SAM Response
SAM DJF
LOW
HIGH
35
Conclusion II
• HIGH shows improved - but not yet satisfying
stratospheric performance; troposphere
generally degrades (tuning problem?)
• Having a well-resolved stratosphere does not
necessarily guarantee a better troposphere
• 20 year simulation length is often not enough to
cleanly separate signal from noise
• This is preliminary analysis, more work is
underway …
• Everybody is invited to look at these and more
simulations
Thank you.
Response to SSTs: JJA
L24 L48 L24 L48 L24 L48
x4CO2A1B2000s (obs.)
ΔT
Δu
-90 -60 -30 Eq 30 60 90
38
[m/s]
[K]
-90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90 -90 -60 -30 Eq 30 60 90
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2SS
T50
O3
_SS
T50
CO
2_
SST5
0
CO
2_
O3
_SST
50
2xC
O2
_SS
T50
2xC
O2
_O
3_SS
T50
4xC
O2
_SS
T50
4xC
O2
_O
3_SS
T50
SST2
k
SST2
k_a
O3
_SS
T2k
O3
_SS
T2k_
a
CO
2_
SST2
k
CO
2_
O3
_SST
2k
L24 NAM JJA
-0.6
-0.4
-0.2
-1E-15
0.2
0.4
0.6
0.8
1
1.2
SST5
0
O3
_SS
T50
CO
2_
SST5
0
CO
2_
O3
_SST
50
2xC
O2
_SS
T50
2xC
O2
_O
3_SS
T50
4xC
O2
_SS
T50
4xC
O2
_O
3_SS
T50
SST2
k
SST2
k_a
O3
_SS
T2k
O3
_SS
T2k_
a
CO
2_
SST2
k
CO
2_
O3
_SST
2k
L48 NAM JJA
Annular Mode Response
NAM JJA
L24
L48
39
Trend in HC from mmc (400-600)SST50, CO2, SST2kPS_EXP/plt_exp_hc.ps (plt_exp_hc.pro)
40
Trend in HC from mmc (400-600)SST50, CO2, 2xCO2, 4xCO2PS_EXP/plt_exp_hc.ps (plt_exp_hc.pro)
41
Timescales NAM u1000-100
Winter Summer
NAMu
SAMu
NAM10 Lagged Correlations
• Lagged correlation of NAM10 with NAM at any other level
• NCEP/NCAR reanalysis
• 61 day time series, 21 day averages
• Similar to: Baldwin & Dunkerton, (1999)
DJFM 1979-2007
43
Composite of 10 SSW events
NAM10 Lagged Correlations
CDC reforecasts (15 members)
Day 0
• 10 SSW
composites
44
Day 15
Day 8