1. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 1 Strategy for seasonal prediction developments at ECMWF Roberto Buizza, Franco Molteni, Magdalena Balmaseda, Thomas Jung, Linus Magnusson, Kritian Mogensen, Tim Stockdale, Yuhei Takaya* and Frederic Vitart European Centre for Medium-Range Weather Forecasts (* Visiting scientist at ECMWF in 2009, now back at CPD of JMA)

2. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 2 Outline 1.ECMWF current and future seasonal systems S3 and S4 2.Recent results from 46-day EPS coupled integrations 3.Model error simulation schemes 4.Model improvements (e.g. convection in the tropics, stratosphere), inclusion of a mixed-layer and a dynamical sea-ice model 5.Conclusions: potential future strategy for seasonal prediction

3. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 3 Delayed Ocean Analysis ~12 days Real Time Ocean Analysis ~8 hours HRES T L 1279L91 (d0-10) HRES T L 1279L91 (d0-10) SF T L 159L62 (m0-7/12) SF T L 159L62 (m0-7/12) EPS T L 639L62 (d0-10) T L 319L62 (d10-15/32) EPS T L 639L62 (d0-10) T L 319L62 (d10-15/32) Atmospheric model Wave model Ocean model Atmospheric model Wave model 1. Baseline operational systems (2010)

4. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 4 1. The ECMWF systems # fcsHor resolutionVert levFc lengthWaveOcean HRES/DA1T127916 km91 (<0.01 hPa, 75km)0-10d WAM (0.25°, 28km) No EPS51 T63932 km 62 (<5 hPa, 35km) 0-10d WAM (0.5°, 56km) No T31964 km10-15/32dHOPE0.3-1.4 degL29 S341T159125km62 (<5 hPa, 35km)0-13mNoHOPE0.3-1.4 degL29

5. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 5 1. ECMWF operational system S3 OASIS-2 HOPE ~ 1.4 deg. lon 1.4/0.3 d. lat. TESSEL IFS 31R1 1.1 deg. 62 levels 4-D variational d.a. Multivar. O.I. Gen. of Perturb. System-3 CGCM Initial Con. Ens. Forecasts

6. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 6 1. The new seasonal forecasting system S4 OASIS-3 NEMO ~ 1.4 deg. lon 1.4/0.3 d. lat. H-TESSEL IFS 36R4 0.7/1.1 deg. 91 levels 4-D variational d.a. 3-D v.d.a. (NEMOVAR) Gen. of Perturb. System-4 CGCM Initial Con. Ens. Forecasts

7. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 7 1. System 4: main features It is expected that S4 will improve the quality of seasonal forecasts thanks to the combined contribution of:  A new ocean model: NEMO v3.0 + 3.1 coupling interface:  ORCA-1 configuration (~1-deg. resol., ~0.3 lat. near the equator)  42 vertical levels, 20 levels with z < 300 m  Better ocean ICs with variational ocean data assimilation (NEMOVAR):  3-D var with inner and outer loop  Collaboration with CERFACS, UK Met Office, INRIA  First re-analysis (1957-2009), no assim. of sea-level anomalies  Second re-analysis and real-time system including SLA  Better atmospheric model cycle (IFS cycle 36r4):  New physics package, including HTESSEL land-surface scheme, snow model (with EC- Earth), new land surface initialization  Prescribed sea-ice conc. with sampling from recent years

8. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 8 Outline 1.ECMWF current and future seasonal systems S3 and S4 2.Recent results from 46-day EPS coupled integrations 3.Model error simulation schemes 4.Model improvements (e.g. convection in the tropics, stratosphere), inclusion of a mixed-layer and a dynamical sea-ice model 5.Conclusions: potential future strategy for seasonal prediction

9. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 9  The Madden-Julian Oscillation (MJO) is a 40-50-day oscillation with peak activity during NH winter and spring. It is a near-global scale, quasi-periodic eastward moving disturbance in the surface pressure, tropospheric temperature and zonal winds over the equatorial belt.  The MJO is the dominant mode of variability in the tropics in time scales in excess of 1 week but less than 1 season.  The MJO influences the Indian and Australian summer monsoons (Yasunari 1979, Hendon & Liebman 1990). It has an impact on ENSO. Westerly wind bursts produce equatorial trapped Kelvin waves, which have a significant impact on the onset and development of an El-Niňo event (Kessler & McPhaden 1995). It has an impact on tropical storms (Maloney et al 2000, Mo 2000) and on NH weather.  Recent results have indicated that the 32d EPS has an improved description of the MJO and is capable to capture at least part of the tropical/extra- tropical interaction due to the MJO. These improvements supports the potential further extension of the EPS to 46-60 days. 2. The MJO and the skill over the NH extra-tropics

10. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 10 These plots show OLR anomalies in ERA-40 (left panel) and in 15d forecasts run (from 29 Dec to 12 Feb) with different model cycles from 28r3 (ope in Oct ‘04) to 35r3 (ope from Sep ‘09 to Jan ‘10). More recent cycles (from 32r2) are better capable to simulate MJO activities. 29/12 05/01 12/01 20/01 28/01 04/04 12/02 32R3 ERA40 days 28R329R131R132R2 10/04 04/0509/0606/07 11/07 33R135R135R3 06/08 09/08 09/09 2. The MJO in subsequent IFS cycles

11. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 11 The skill of probabilistic forecasts of 2mT in the upper tercile for cases with MJO-in-ICs and NO-MJO-in-ICs have been compared (1989-2008). Results based on 45 cases show that the skill of weekly average d19-25 is higher if these is an active MJO in the ICs. MJO in ICs BSS(EU)=0.03 NO MJO in ICs BSS(EU)=-0.09 2. The MJO and the skill over NH in NDJFMA MJO in ICs BSS(NH)=0.04 NO MJO in ICs BSS(NH)=-0.06

12. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 12 2. The MJO and the skill over NH in NDJFMA DAY 5-11 Z500T850Precip DAY 12-18 Z500T850Precip DAY 19-25DAY 26-32 Z500T850Precip Z500T850Precip MJO in ICs - NO MJO in ICs

13. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 13 2mT 2. 32d EPS scores over NH (ROCA) MSLPTP D5-11 D12-18 D19-25 D26-32 D5-11 D12-18 D19-25 D26-32 D5-11 D12-18 D19-25 D26-32 Results based on 45 cases (ICs 4 times a year from ‘91 to ’02, 5m) indicate that 32d EPS weekly average probabilistic forecasts of 2mT and MSLP over NH land points have some level of skill up to forecast day 32. This is shown hereafter for the probabilistic prediction of occurrence of upper tercile values. Motivated also by these results, work has continued to investigate whether a 46-day extension could provide skilful forecasts up to 6 weeks ahead.

14. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 14 2. Further extension of the EPS to 46 days? 15-member ensembles starting on the 15 th of each month from 1991 to 2007 (1979-2008 for the 15 th July starting date) have been integrated for 46 days using the same configuration as the operational monthly forecasts. Results indicate that those forecasts are significantly more skilful than the seasonal forecasts of month 2 issued the same day (15 th of the month) Average ROC area for PR(2MT>upper 1/3) computed for all NH land point for DJF. Blue is the SF d30-61 forecast (available on the 15 th of the month). Red is the EPS d15-46 forecast, which would also be available on the 15 th of the month.

15. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 15 Outline 1.ECMWF current and future seasonal systems S3 and S4 2.Recent results from 46-day EPS coupled integrations 3.Model error simulation schemes 4.Model improvements (e.g. convection in the tropics, stratosphere), inclusion of a mixed-layer and a dynamical sea-ice model 5.Conclusions: potential future strategy for seasonal prediction

16. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 16 Since 1999, ECMWF has been running its ensemble prediction systems with stochastic schemes designed to simulate model error. The first version of the stochastically perturbed parameterized tendency (SPPT) scheme implemented in the EPSS in 1999 was designed to simulate the effect of random model errors due to parameterized physical processes. S3 has been running with this scheme. Since Nov 2010, the EPS has been running with a revised version of SPPT and a stochastic back-scatter scheme (SPBS). The SPPT revisions (implemented in 2009 and 2010) includes:  Perturbations with 3 different spatio-temporal scales and amplitudes  A revised (vertically consistent) treatment of super-saturation  A tighter clipping of the Gaussian distribution (2σ instead of 3σ to address potential numerical instabilities). It is expected that the inclusion of the revised-SPPT and the BS in S4 should improve its reliability. 3. Stochastic schemes: SPPT

17. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 17 500 km 6 h 1000 km 3 d 2000 km 30 d 3. The multi-scale SPPT (in ope EPS since N10)

18. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 18 Work is continuing to develop a stochastic back-scatter scheme:  Rationale: a fraction of the dissipated energy is backscattered upscale and acts as streamfunction forcing for the resolved-scale flow (Shutts & Palmer 2004, Shutts 2005, Berner et al 2009)  Streamfunction forcing is given by: Streamfunction forcing Backscatter ratio Total dissipation rate Pattern generator 3. SPBS (in ope EPS since N10)

19. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 19 Outline 1.ECMWF current and future seasonal systems S3 and S4 2.Recent results from 46-day EPS coupled integrations 3.Model error simulation schemes 4.Model improvements (e.g. convection in the tropics, stratosphere), inclusion of a mixed-layer and a dynamical sea-ice model 5.Conclusions: potential future strategy for seasonal prediction

20. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 20 ECMWF Model Formulation: Relaxation Formulation: Relaxation coefficient, λ, depends on latitude, longitude and height. x ref is based on (linearly interpolated) analysis fields. 4. Tropics and stratosphere (relaxation exps)

21. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 21 CON TR-CON STRA-CON Day6-Day10Day16-Day20Day26+Day30 (λ=1.0) Relaxing the tropics (middle) leads to a reduction of fc error both over NA and EU. A better representation of the stratosphere (bottom) has also a positive impact over the high-latitudes in the longer forecast range. 4. Tropics, stratosphere and NH-extra-tropics

22. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 22  Y Takaya started implemented the KPP mixed-layer model in the IFS and started the work to assess its impact on the EPS monthly time scale.  Results obtained by Y Takaya (now at JMA) using the KPP model in the tropical band (40°S-40°N) coupled to the atmospheric model grid indicated that in some cases the KPP model improved the simulation of the MJO propagation. 40N 30N 30S 40S 4. Mixed-layer model

23. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 23 IFS NEMO LIM2 Oasis 3 Work has started to test the impact of using the Louvain-la- Neuve sea Ice Model (LIM2, http://www.astr.ucl.ac.be/) on long-range integrations. LIM2 is a 2-layer ice model with parameterised ice thickness distribution. Preliminary results indicate that the 8-year average sea-ice extension predicted using LIM2 propagates too far south. 4. Sea-ice modelling LIM2 OBS

24. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 24 Outline 1.ECMWF current and future seasonal systems S3 and S4 2.Recent results from 46-day EPS coupled integrations 3.Model error simulation schemes 4.Model improvements (e.g. convection in the tropics, stratosphere), inclusion of a mixed-layer and a dynamical sea-ice model 5.Conclusions: potential future strategy for seasonal prediction

25. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 25 Conclusions  Recent EPS developments have indicated that the latest model cycles (say since 2008) are more capable to simulate the MJO and to capture the link between the MJO and the extra-tropics. EPS weekly average forecasts of variables such as 2mT and MSLP are skilful up to 4 weeks in advance. EPS d15- 46 average fcs issued on the 15 th of each month are better than m2 S3 fcs. This suggests that one possible way to generate better monthly outlooks could be to extend the EPS up to 46-60 days.  Work is progressing to assess the impact of using the revised, multi-scale SPPT and the SPBS model error schemes in the seasonal system. This should also lead to seasonal forecast improvements.  Recent relaxation experiments have indicated that improving the tropics (better physics) and the stratosphere simulation can lead to improvements in the NH extra-tropics. This is one of reasons why S4 will use 91 instead of 62 vertical levels and the TOA will be moved from ~5Pa to ~1Pa.

26. JMA WS (9 Dec 2010) - Roberto Buizza et al : Strategy for seasonal prediction developments at ECMWF 26 Conclusions Work will continue to further improve the ECMWF seasonal forecasts by: 1.Implementing the coupled seasonal system 4 (S4) – The new system will use NEMO and NEMOVAR 3D-Var ocean data-assimilation, atmospheric model cycle 36r4 with a better physics and 91 vertical levels with the TOA at 1Pa. 2.Improving the model error simulation schemes in the EPS and SF – Work is progressing to assess the impact of the revised-SPPT and the SPBS schemes on seasonal forecasts. 3.Including a dynamical sea-ice model and a mixed-layer model – Work has started to assess whether using the LIM2 dynamical sea-ice model and a mixed-layer model can lead to further improvements. 4.Merging the EPS and the SF in a Seamless Probabilistic System (SPS) – Once NEMO has been implemented also in the EPS, assess the possibility to merge the EPS and the SF (e.g. by using EPS d15-46 forecast to predict the first next calendar month and then further truncate the resolution). 5.Increasing the resolution of the ocean model - Assess the impact of using a higher-resolution ocean model (1/4 degree NEMO) in the probabilistic forecasting systems.