1. The North Atlantic − The NAO, AO and the MJO
Hai Lin
Meteorological Research Division, Environment Canada
Workshop “Sub-seasonal to Seasonal Prediction”
Met Office, Exeter, Dec. 1-3, 2010

2. Outlines Challenge of prediction in the North Atlantic and Europe
Brief introduction of NAO / AO and its impact
NAO prediction on intraseasonal time scale
MJO contribution;
intraseasonal hindcast
NAO seasonal prediction
possible signal sources;
skill in four Canadian AGCMs

3. Challenge of subseasonal and seasonal prediction in the North Atlantic and European region
Strong variability due to atmospheric internal nonlinear interactions
Far from major source of interannual variability (e.g., ENSO)
Low forecast skill

4. What is the NAO?
The North Atlantic Oscillation is a large-scale seesaw in atmospheric mass between the subtropical high-pressure system over the Azores Islands and the subpolar low-pressure system over Iceland.
(From American Museum of Natural History website)

5. The NAO
The NAO is one of the most important modes of atmospheric variability in the northern hemisphere
The NAO has a larger amplitude in winter than in summer
The NAO accounts for 31% of the variance in winter surface air temperature north of 20°N (Hurrell, 1995)

6. The AO
The Arctic Oscillation has a global scale, more zonally symmetric, also called the Northern Annular Mode (NAM)
Connection to stratosphere (e.g., Baldwin and Dunkerton 2001)
The NAO can be regarded as a local representation of the AO in the North Atlantic

7. Impact of the NAO Subtropical high pressure and Icelandic low
Westerly winds and storm activity across the Atlantic Ocean
Temperature and precipitation in Europe, northeastern Canada and Greenland
Impact on forecast skill

8. How is the NAO variability generated?
Causes within the atmosphere: interactions among different scales and frequencies in the atmosphere lack of forecast skill beyond 2 weeks
Causes external to the atmosphere
(on seasonal and interannual time scales):
Sea surface temperature (SST) anomaly in the North Atlantic
Changes in ice and snow cover
SST anomaly in the tropics

9. NAO forecasts
Intraseasonal time scale
impact of the MJO

10. The Madden-Julian Oscillation (MJO)
Discovered by Madden and Julian (1971). Spectrum analysis of 10 year record of SLP at Canton, and upper level zonal wind at Singapore. Peak at days.
Dominant tropical wave on intraseasonal time scale
30-60 day period, wavenumber 1~3
propagates eastward along the equator (~5 m/s in eastern Hemisphere, and ~10 m/s in western Hemisphere)
Organizes convection and precipitation

11. Composites of tropical
Precipitation rate for 8 MJO phases, according to Wheeler and Hendon index.
Xie and Arkin pentad data,
The red arrows mark Phases 2-3 and Phases 6-7. They are characterized by a dipole convection anomaly distribution in the tropics. As will be shown later, such a dipole diabatic heating has a great impact on the Northern Hemisphere extratropics and the NAO.

12. Connection between the MJO and NAO
NAO index: pentad average
MJO RMMs: pentad average
Period:
Extended winter, November to April (36 pentads each winter)

13. Lagged probability of the NAO index Positive: upper tercile; Negative: low tercile
Phase 1 2 3 4 5 6 7 8
Lag −5
−35%
−40%
+49%
Lag −4
+52%
+46%
Lag −3
Lag −2
+50%
Lag −1
Lag 0
+45%
−42%
Lag +1
+47%
−46%
Lag +2
+42%
−41%
Lag +3
+48%
−48%
Lag +4
−39%
Lag +5
From left to right: 8 phases of the MJO. From top to bottom: lags in pentads. Negative lags mean that the NAO leads the MJO, while positive lags indicate that the NAO lags the MJO.
The numbers are probability of occurrence of strong positive NAO (red) and strong negative NAO (blue). Those passing 5% significance level (bootstrap method) are shown.
Here we look at lags > 0 for impact of the MJO on the NAO. 1-3 pentads after MJO Phases 2-4, there is high probability of strong positive NAO. 1-4 pentads after MJO Phases 6-8, high chance of strong negative NAO.
For lags < 0, it is the impact of NAO on MJO, we are not focus on this aspect in this talk.
(Lin et al. 2009)

14. Tropical influence Z500 anomaly (Lin et al. JCLIM, 2009)
Composite of Z500 anomaly after MJO Phase 3 (upper panels), and Phase 7 (lower panels). Phase 3 has a dipole tropical convection anomaly (above normal convection in Indian Ocean and below normal convection in western Pacific). So does Phase 7, but with an opposite sign.
At the simultaneous composite of Phase 3 (a), a negative PNA-like pattern is generated. One pentad later (b), Rossby wave dispersion can be seen with intensification of downstream centers. After another pentad (c), a positive NAO is formed. There is some contribution from transient eddies in the last stage.
For Phase 7, the Rossby wave propagation is less clear. A negative NAO is generated at lag=2 pentads.
(Lin et al. JCLIM, 2009)

15. Impact on Canadian surface air temperature
Lagged winter SAT anomaly in Canada
This is an example of the impact of the MJO on surface temperature. Such signal would be useful for intraseasonal prediction in North America.
(Lin et al. MWR, 2009)

16. Why the response to a dipole heating is the strongest ?
Barotropic instability of 2-D basic flow: similar mechanism as Simmons et al. (1983)
Rossby wave generation determined by relative position of tropical forcing wrt jet stream (Lin 2010)
To demonstrate this:
Primitive equation GCM (T31, L10)
Linear integration, winter basic state
with a single center heating source
Heating at different longitudes along the equator from 60E to 150W at a 10 degree interval, 16 experiments
Z500 response at day 10
These two points are possible mechanisms.

17. Similar pattern for heating 60-100E a) 80E
Day 10 Z500 linear response
Similar pattern for heating E
a) 80E
b) 110E
When there is a dipole as MJO Phase 3 (heating at 80E and cooling at 150E), (note for cooling the response of c reverse sign), a-c makes a very strong extratropical response.
Similar pattern for heating W
c) 150E
Lin et al. (2010)

18. ISO hindscast with GEM
GEM clim of Canadian Meteorological Centre (CMC)--
GEMCLIM 3.2.2, 50 vertical levels and 2o of horizontal resolution
3 times a month (1st, 11th and 21st)
10-member ensemble (balanced perturbation to NCEP reanalysis)
NCEP SST, SMIP and CMC Sea ice, Snow cover: Dewey-Heim (Steve Lambert) and CMC
45-day integrations
GEM refers to the Global Environmental Multi-scale model, which is an operational model of Canadian Meteorological Centre.

19. NAO forecast skill extended winter – Nov – March tropical influence
A simple measure of skill:
temporal correlation of NAO index btw forecast and observations

20. The dynamical model behaves clearly better than the persistence forecast. Skill above 0.5 up to the 3rd pentad.
(Lin et al. GRL, 2010)

21. The solid lines are skill of ensemble forecast
The solid lines are skill of ensemble forecast. Red line is for those forecasts whose initial condition has a strong MJO (amp >1), whereas the blue line for those with a weak MJO (amp <1) in the initial condition. The dashed lines are average of individual members. The vertical bars are spread among members. The amp of MJO is defined as square root of (RMM12+RMM22).
It is clear that an NAO forecast starting from a strong MJO has better skill.
(Lin et al. GRL, 2010)

22. NAO forecasts grouped by the MJO phase at the initial condition
NAO forecasts grouped by the MJO phase at the initial condition. Red line corresponds to those with a dipole convection anomaly.
(Lin et al. GRL, 2010)

23. Correlation skill: averaged for pentads 3 and 4
When the initial condition has a strong MJO, there is a better forecast skill (15-20 days). The pattern reflects the NAO.
Correlation skill: averaged for pentads 3 and 4

24. Correlation skill: averaged for pentads 3 and 4
This also leads to improved skill in surface temperature over North Atlantic and European region.
(Lin et al. GRL, 2010)

25. NAO seasonal forecasts
Possible signal sources:
Sea surface temperature (SST) anomaly in the North Atlantic (e.g., Rodwell et al. 1999)
Changes in ice and snow cover (e.g., Cohen and Entekhabi 1999)
SST anomaly in the tropics (e.g., Jia et al. 2008)
Now we switch to seasonal forecast of the NAO.

26. Historical forecast (HFP2)
4 global models GEM: 2°x2°, 50 levels
AGCM2: 625 km (T32), 10 levels
AGCM3: 315 km (T63), 32 levels
SEF: 210 km (T95), 27 levels
Once a month (beginning of each month)
4-month integrations
10 members each model
Persistent SST anomaly
Sea ice and snow cover anomalies relaxed to climatology

27. Skill of seasonal mean NAO index as a function of starting month
Skill of seasonal mean NAO index as a function of starting month. The horizontal line represents 5% significance level.
It is seen that for lead=0 forecast, late winter and spring forecasts are skillful. Consistent among 4 models.

28. The skill drops to non-significant for lead=1 month forecasts
The skill drops to non-significant for lead=1 month forecasts. There is no skill for NAO seasonal forecast if skipping month 1.

29. NAO seasonal forecast skill
Lead=0: skill in late winter to spring
Four models have similar performance
Lead=1 month: no skill
Possible explanation:
skill comes from initial condition
models do not have a correct response pattern in the NAO (this will be explored in the next couple of slides)

30. Identify dominant forced patterns
For the DJFM run:
SVD analysis between November tropical Pacific SST and
DJF or JFM ensemble mean Z500
The expansion coefficient of SVD2 (Z500) is significantly correlated with the observed NAO index
As for DJFM run, November SST anomaly is used, so we use November SST for the SVD analysis. Ensemble mean of Z500 forecast represents the forced component. The SVD will identify the source and response in the model.

31. Leading pairs of SVD in observations
November SST vs JFM z500
Z500
SVD1 (left panels) represents a typical ENSO signal.
SVD2 (right panels): positive NAO in Z500, and negative SST anomaly in central equatorial Pacific
This is from observations.
SST

32. Leading pairs of SVD in GEM ensemble mean
November SST vs JFM z500
For GEM model (ensemble mean)
SVD1 very well reproduce the observed ENSO signal.
SVD2: Z500 is very different from the NAO. Therefore the model response to the second SVD SST has a wrong spatial pattern.

33. Leading pairs of SVD in GCM3 ensemble mean
November SST vs JFM z500
The same feature as in GEM can be seen for GCM3.

34. NAO skill of ensemble forecast
Temporal correlation with DJF observed NAO index
Forecast NAO index
Forced SVD2
GCM2
-0.13
0.30
GCM3
0.26
0.57
SEF
0.33
0.47
GEM
0.25
0.39
Although the model SVD2 has a biased spatial pattern (from NAO), its time evolution is significantly correlated with the observed NAO.
This is for lead=0.
Lead = 0

35. NAO skill of ensemble forecast
Temporal correlation with JFM observed NAO index
Forecast NAO index
Forced SVD2
GCM2
-0.31
0.35
GCM3
0.27
0.43
SEF
0.12
0.42
GEM
0.20
0.31
This is also true for lead=1 month.
Lead = 1 month

36. NAO skill of ensemble forecast
Model has a biased NAO pattern
The forced SVD2 pattern has a time evolution that matches well the observed NAO index  can be used as a skillful forecast of the NAO index
The biased pattern is likely caused by bias in model climatology, as the Rossby wave propagation and Atlantic response depends on basic state. However, the time evolution of the signal source can provide some useful information for seasonal forecast of the NAO index.

37. Summary Significant impact of the MJO on the NAO
NAO intraseasonal forecast skill influenced by the MJO
Some skillful NAO seasonal forecast possible in late winter and spring
Seasonal forecast of NAO has biased spatial pattern, some statistical post-processing procedure can improve the skill