1. Alexey Karpechko & Elisa Manzini
Diagnosing stratospheric contribution to climate change in the CMIP5 models
Alexey Karpechko &
Elisa Manzini

2. Contributions from many people, most importantly:
James Anstey,
Seok-Woo Son,
Paolo Davini,
Giuseppe Zappa,
Natalia Calvo,
Steven Hardiman,
and others…

3. Stratosphere and climate change
The atmosphere will change in response to GHG emissions
The stratosphere will change too
How will the stratosphere change?
And what are the implications of stratospheric changes for tropospheric climate change?

4. Arctic polar vortex changes -> ?
In the Northern Hemisphere future changes in the Arctic polar vortex remain poorly understood, although a convergence of model results is appearing.
Equatorward shift of the polar night jet
Seems to be the most common response to doubling CO2 in stratosphere-resolving models
but
…not reproduced by all models
Large decadal variability may mask the forced signal (Butchart et al. 2000)
Sigmond et al (2004)

5. Arctic polar vortex changes
There seems to be a systematic difference between high-top (stratosphere-resolving) and low-top models
Low-top models typically simulate strengthening of zonal winds throughout the polar stratosphere but…
Not necessarily → a high-top is not needed to simulate weakening of the polar winds:
ECHAM5, U, 2xCO2, JFM
Scaife et al (2011)

6. Arctic polar vortex changes: Impacts on the troposphere
Positive vs negative response of the Northern Annular Mode
Spread of the results (Shindell et al. 1999; Fyfe et al 1999; Gillett et al. 2003; Sigmond et al. 2008, 2010; Scaife et al. 2011, Karpechko and Manzini 2012)
Likely related to changes in polar stratospheric winds (weaken or strengthen)
Future weakening of the polar stratospheric winds drives the NAM towards negative phase
(But there are other factor influencing future NAM changes)
Karpechko and Manzini (2012)

7. Future Arctic polar vortex changes
Equatorward shift of the Arctic polar night jet seems to be the most common response to GHG increases in stratosphere-resolving models
Large interannual and decadal variability
Lack of understanding of the mechanisms
Up-down or down-up influence?
Impacts on future surface climate remain unclear

8. CMIP5 models What do CMIP5 climate simulations say about
future changes in the Arctic wintertime polar vortex
and
its implications for surface climate change?

9. CMIP3/IPCC AR4 models High top models:
the lid is above the stratopause
Low top models:
the lid is below the stratopause
Low tops dominate
based on Cordero and Forster (2007)

10. CMIP5/IPCC AR5 models High tops dominate
from Charlton-Perez et al (2013)
High tops dominate

11. Data 24 CMIP5 models 10 Low-tops, 13 High-tops, 1 intermediate (1 hPa)
Historical and rcp 8.5 simulations
Monthly mean sea level pressure (SLP), zonal mean U and T
42 simulations for SLP and U
7 models with more than 1 simulation
24 simulations for T
One simulation per model only
Focus on DJF mean (unless otherwise mentioned)
Difference between mean and mean 11

12. CMIP5 zonal wind changes
Multi-model mean change
~70% models simulate weakening polar stratospheric winds
Large spread among individual models
Largest spread is around 70°N and 10hPa
Why spread?
What are its implications for surface climate?
Intermodel standard deviation (σ)
Define stratospheric winds (SUA) index:
(-1∙ΔU)

13. CMIP5 vs CMIP3 models the period 101-140 minus the period 1-40 DJF
1%CO2 experiment in CMIP3&CMIP5
the period minus the period 1-40
DJF
CMIP5: Zonal winds weaken north 70N
CMIP3: Zonal winds strengthen through the stratosphere 13 13

14. CMIP5 zonal temeparture changes
Multi-model mean change
Largest intermodel spreads in:
Polar stratosphere (SUA index)
Upper tropical troposphere (tropical warming)
Lower high latitude troposphere (Arctic amplification)
Is the stratospheric spread related to those in the troposphere?
Or is it not?
Intermodel standard deviation (σ)

15. 1. Impact of tropical warming on tropospheric dynamics: 30oS EQ 30oN

16. 1. Impact of tropical warming on tropospheric dynamics: 30oS EQ 30oN
STRONGER SUBTROPICAL WESTERLY JET
30oS EQ
30oN
60oN
90oN

17. 2. Impact of polar amplification on tropospheric dynamics: 30oS EQ

18. 2. Impact of polar amplification on tropospheric dynamics: 30oS EQ
WEAKER WINDS AT HIGH LATS
30oS EQ
30oN
60oN
90oN

19. 3.
A stratospheric pathway?
30oS EQ
30oN
60oN
90oN

20. 3.
A stratospheric pathway?
30oS EQ
30oN
60oN
90oN

21. 3.
A stratospheric pathway?
30oS EQ
30oN
60oN
90oN

22. 3.
A stratospheric pathway?
30oS EQ
30oN
60oN
90oN

23. 1. Separating a stratosphere-congruent signal

24. Approach
Yj (φ,p) – T, U, or SLP change in j-th model between and
X1j – tropical warming index ( change between and )
X2j – polar amplification index ( change between and )
X3j – stratospheric change index ( change between and )
All indices are normalized and have zero mean and unity standard deviation.
Index correlation matrix:
Tropics
Arctic
SUA
Indices correlate
Tropics
Arctic
SUA

25. Approach
A multiple regression is not used because of correlation between indices. A step-by- step regression is applied instead: 1. 2. 3.
X2j – is calculated on residuals after the 1st regression
X3j– is calculated on residuals after the 2nd regression
The stratosphere-congruent part (a3 regression coefficient) does not change between the approaches because the SUA index does not correlate with the other indices.

26. Multi-model mean change (Y) Polar amplification (X2)
Zonal temperatures
Multi-model mean change (Y)
Tropical warming (X1)
Polar amplification (X2)
SUA index
(X3)
Regressions
σ(Y)
σ2(ε1)/σ 2(Y)
σ2(ε2)/σ 2(Y)
σ2(ε)/σ 2(Y)

27. Multi-model mean change (Y) Polar amplification (X2)
Zonal winds
Multi-model mean change (Y)
Tropical warming (X1)
Polar amplification (X2)
SUA index
(X3)
Regressions
σ(Y)
σ2(ε1)/σ 2(Y)
σ2(ε2)/σ 2(Y)
σ2(ε)/σ 2(Y)

28. Multi-model mean change (Y) Polar amplification (X2)
Sea level pressure
Multi-model mean change (Y)
Tropical warming (X1)
Polar amplification (X2)
SUA index
(X3)
Regressions
σ(Y)
σ2(ε1)/σ 2(Y)
σ2(ε2)/σ 2(Y)
σ2(ε)/σ 2(Y)

29. High tops vs low tops
Different climate sensitivity in high- tops and low tops
=> different Arctic SLP response
The reason is not clear
No evidences that different climate sensitivity is related to different stratospheric parts
There are evidences that the difference is NOT related to startospheric parts:
Similar climate sensitivity in high/low versions of the same model
figure by S.-W. Son

30. 2. Internal variability?

31. Intra- vs intermodel spread
ANOVA test for SUA index:
Intermodel variance: 18.2 m2/s2
Mean intramodel variance: 3.6 m2/s2
F-statistics: 5.1
Intermodel spread dominates
(-1)∙
The intermodel spread in SUA index is unlikely explainable by internal variability

32. 3. Up-down or down-up influence?

33. Lagged correlation between changes at different altitudes
NAM changes at the end of winter (February)
Correlated with zonal wind changes at different altitudes and in different months
Remove the tropical warming signal

34. Lagged correlations Downward influence
NAM in February correlates better with stratospheric wind changes in previous months, rather than with tropospheric ones
Downward influence

35. Summary (1)
~70% of CMIP5 models simulate weakening of the Arctic stratospheric winds in winter (equatorward shift of the polar night jet)
(Almost all CMIP3 models simulate strengthening of the stratospheric winds)
Spread of the polar stratospheric wind changes among individual models is large
Multiple realizations available for some models show much smaller spread around their respective means
This suggests that internal variability unlikely explains the intermodel spread in the Arctic stratospheric wind changes

36. Summary (2)
The intermodel spread in stratospheric wind changes is not related to model climate sensitivity (i.e. tropical warming or polar amplification)
There is a strong coupling between zonal wind changes and sea level pressure changes:
A considerable inter-model spread in sea level pressure change is associated with the inter-model spread in the Arctic winter stratospheric change
Weakening of the Arctic stratospheric winds corresponds to a shift towards more negative NAM
(But overall future NAM change depends also on climate sensitivity)
Lagged correlations suggest that changes in stratospheric winds influence tropospheric changes (downward influence)