1. Stratosphere-Troposhere Coupling in Dynamical Seasonal Predictions Bo Christiansen Danish Meteorological Institute

2. Opinion of some dynamical forecasters: Stratosphere-troposphere coupling is not important, the stratospheric signal is just an imprint of what is going on in the troposphere Stratosphere-troposphere coupling is already included in the models Let us forget about the stratosphere and increase the horizontal resolution Motivation

3. Layout of the talk: What kind of behaviour should we expect? Simple statistical forcasts based only on observations. Dynamical model has to do better than that. Why should we expect the stratosphere- troposphere coupling to be included in dynamical models? Some results from a dynamical seasonal prediction system

4. Downward propagation of zonal mean zonal wind in ERA40. Annual cycle and timescales faster than 30 days are removed. Watch the movie at www.dmi.dk/solar-terrestrial/staff/boc/homepage.shtml

5. Advantages compared to EOF based indices: simple physical meaning easy to calculate archived for most GCM experiments no risk of spurious modes due to noise and no mode mixing My choice of zonal index: the zonal mean zonal wind a 60 N At the surface it is strongly correlated with the AO/`NAO In the stratosphere it is strongly correlated with the strenghth of the vortex

6. 70 hPa surface 10 hPa 70 hPa surface Forecast of daily values Forecast of 14 days means Forecast skill as function of lead time T for different vertical levels of the predictor. Purple curve shows forecast when wind at surface and at 70 hPa are used as predictors simultaneously Only winter, DJF Predicting surface zonal wind

7. The forecast skill as function of lead time and the vertical level of the predictor. Daily values are predicted. Winter season. Shaded regions are where correlations are significantly different from zero at 99 and 95 % levels. Calculated by Monte-Carlo approach assuming normality and observed temporal structure. Predicting surface zonal wind

8. The forecast skill as function of lead time and the time over which the predictand is averaged. The level of the predictor is 70 hPa. The forecast skill as function of lead time and the strength of the predictor. 14 days means are forecasted.

9. Comparison with dynamical forecast. 51 events from the ECMWF ensemble seasonal prediction system 2 surface 70 hPa Model 70 hPa, 51 events Predictand is surface wind at 60 N, Daily values are forecasted. Model+70 hPa

10. Observations Full GCM Perp. Jan. GCM Holton Mass model Minimal model Downward propagation is robust and ubiquitous Zonal wind at 60 N

11. Quasi-Biennial Oscillation

12. Zonal mean wind at 60 N Vertical component of EP-flux at 60 N Vertical component of EP-flux at 100 hPa

13. Covariance between zonal mean wind at 10 hPa, 60 N and components in the balance equation for zonal monentum. Lag (days)

14. height wind Radiative equilibrium The basic mechanism

15. A minimal model Zonal wind trend Coriolis term Wave coupling 1-dimensional: Simple resistance: Nonlinear coupling (Charney-Drazin):

16. How much does the stratosphere control? ARPEGE GCM, perpetual Januarry 5 different transient perturbations in 10 different layers, 8 different initial conditions Christiansen, QJRMS., 129, 2003.

17. Experiments with constant troposphere show that vacillations can exist without growth of disturbances

18. Errors grow like a power-law, not exponential Not like deterministic chaos in low dimensional systems. However, systems with many degrees of freedom can show power-law growth of perturbations as shown by Lorenz (1969).

19. There are some reasons to believe that dynamical forecast models may already include the stratsphere-troposphere coupling: The coupling is present in models of different complexities At least part of the coupling can be explained by a simple mechanism (which unlike the QBO depends on large-scale waves) The coupling is well represented in the ARPEGE GCM Perhaps the stratosphere is only passively responding to tropospheric processes, perturbations may develop independent of the downward propagation

20. Hindcasts with 11 ensemble members1981-2005 Model has 62 vertical levels with top at 5 hPa But: Only archived at 10 levels.. 200, 50, 10 hPa ERA40 has.. 200, 150, 100, 70, 50, 30, 20, 10 hPa Initial conditions based on ERA40 for 1981-2001 and operational analysis for 2002-2005 Model started the first day of every months, giving 3x25 different DJF events ECMWFs dynamical ensemble seasonal prediction system

21. One example of the ensemble forecast Ensembe mean Target

22. Forecast Target Target reduced to 10 layers Forecast - Target One example of ensemble mean forecast

23. Lagged correlations between U at10 hPa and U at other levels Shaded regions are where correlations are significantly different from zero at 99 and 95 % levels. Calculated by a t-test assuming normality and independent predictions ERA40, all data

24. Observations Model

25. Forecast skill: Correlations between forecast and target

26. Forecast skill at the surface: Correlations between forecast and target Stat. model Dynamical Model

27. Conclusions Downward propagation is ubiquitous: found in observations and models of different complexity Downward propagation driven by waves from the troposphere and the two-way interaction between mean flow and waves is important Dynamical seasonal prediction model does include stratosphere- troposphere coupling But this coupling is too strong compared to observations Dynamical prediction model strongly overestimates the decorrelation time in the stratosphere. Also somewhat overestimated in the troposphere. Dynamical prediction model has more skill in the stratosphere compared to the statistical model for lead times up to 50 days.