1. 1 A review of CFS forecast skill for 2005-2007 Wanqiu Wang, Arun Kumar and Yan Xue CPC/NCEP/NOAA

2. 2 Model - T62/L64 GFS (2003) / MOM3 Initial conditions - R2, GODAS Forecast - Implemented Aug 2004 - 9 target months - 2 forecast members per day - 40-member ensemble from latest 20 days - 1981-2003 retrospective forecasts for calibration The NCEP CFS (Climate Forecast System)

3. 3 Skill assessment - SST and precipitation in the tropics - T2m, precipitation, and z200 in the extratropics Analysis of Nino34 SST forecast - Comparison with retrospective forecast - Impacts of atmospheric wind feedback - Impacts of ocean initial condition - Impacts of intraseasonal variability Outline

4. 4 Skill assessment - SST and precipitation in the tropics - T2m, precipitation, and z200 in the extratropics

5. 5 Nino34 SST (K) 3-mon lead 1-mon lead 6-mon lead  Larger amplitude of forecast anomalies  Forecast tends to prolong and enhance existing anomalies  Forecast transitions in 2006/2007 delayed. Obs

6. 6 SST (K) Precipitation (mm/month)  Forecast SST anomalies in the Pacific persist longer  Forecast precipitation anomalies in the Pacific extend too far east Observation Forecast (1-mo lead) 5S-5N average

7. 7  Modest skill  Precipitation skill generally follows SST skill Tropical (20S-20N) anomaly correlation

8. 8 2006 Nino34 SST (K) Pattern correlation skill Low precipitation skill Large variations in T2m and Z200 skill Highest T2m skill in 2005/2006 spring when observed Nino34 SSTs are near zero.

9. 9 Land T2m (K) Z200 (m) Forecast failed to produce observed positive anomalies over most of the period possibly due to lack of increasing CO2 and associated warming Forecast positive T2m anomalies in northern spring Observation Forecast

10. 10 - Comparison with retrospective forecast - Impacts of atmospheric wind feedback - Impacts of ocean initial condition - Impacts of intraseasonal variability Analysis of Nino34 SST forecast

11. 11 Nino34 SST (K)  CFS captured transition to cold phase in 2005  CFS tended to prolong and enhance existing anomalies in late 2005

12. 12 Nino34 SST (K)  CFS tends to prolong and enhance existing anomalies  Reasonable forecast from Dec2006

13. 13 Nino34 SST (K)  Early onset of the cold event  Overall reasonable forecast from 2007 initial conditions Nino34 SST (K)

14. 14  Lower forecast correlation and Heidke skill  Comparable RMS errors Anomaly correlation Heidke skill RMS error (K) Hcst: 1981-2004 Fcst: 2005-2007 Nino34 SST (K)

15. 15 Nino34 SST (K)  Forecast performance after 2005 appears similar to 1981-2004  CFS tends to produce larger amplitude and delay the transitions Nino34 SST (K)

16. 16  Large interannual variability  Lowest predictability in early 1990s  Recent years are in a low predictability regime  Lower skill corresponds to smaller standard deviation Anomaly correlation Heidke skill Observed standard deviation (K) Statistics of three-year running segments (x-axis: beginning of three-year segments) Nino34 SST (K)

17. 17  Large interannual variability  Lowest predictability in early 1990s  Recent years are in a low predictability regime  Lower skill corresponds to smaller standard deviation  RMS errors do not clearly vary with standard deviation Statistics of three-year running segments (x-axis: beginning of three-year segments) Month 6 forecast Nino34 SST (K)

18. 18 The CFS forecasts captured overall observed tropical anomalies during 2005-2007. But forecast anomalies in the Pacific have larger amplitude, persist longer, and extend too far east. For North America, forecast skill of precipitation is low. Forecast skill of T2m varies greatly. Best T2m forecast occurred in the spring of 2005 and 2006 when tropical SST anomalies are weak. (impacts of land-surface initial conditions?) The CFS captured interannual variability in zonal mean T2m and Z200 to some extent. But failed to reproduce the observed overall warmth. (need of the inclusion of external forcing?) In terms of Nino34 SST, 2005-2007 is in a low predictability period because of weak anomaly amplitude. Errors in ocean initial condition, atmospheric wind feedback, and high- frequency variability have contributed to SST forecast errors. Summary

19. 19 Thank you !

20. 20 Backup slides Analysis of Nino34 SST forecast

21. 21  Skill increases with standard deviation  Smaller variation for larger standard deviation Nino34 SST Statistics of three-year running segments Forecast correlation versus observed standard deviation

22. 22 Nino34 SST Possible impacts of ocean initial conditions

23. 23 Forecast from JFM 2006 Initial negative temperature errors in the EPAC may be responsible for forecast errors

24. 24 Forecast from JFM 2006 Erroneous temperatures likely contributed to the forecast SST errors.

25. 25 Forecast from ASO 2006 Forecast errors are not clearly associated with initial ocean temperature errors.

26. 26 Nino34 SST Possible impacts of atmospheric wind feedback

27. 27 Forecast from ASO 2006 Erroneous response in the tropical WPAC likely contributed to the forecast SST errors.

28. 28 Forecast from ASO 2006 Erroneous Taux response in the tropical WPAC likely contributed to the forecast SST errors.

29. 29 Forecast from JFM 2006 Erroneous response in the tropical WPAC likely contributed to the forecast SST errors.

30. 30 Forecast from JFM 2006 Erroneous Taux in the tropical WPAC likely contributed to the forecast SST errors.

31. 31 Nino34 SST Possible impacts of high-frequency variability

32. 32 Evolution of Equatorial Pacific SST (ºC), 850-mb Zonal Wind (m/s), 0- 300m Heat Content (ºC) and MJO Activity - Large SST variations associated with intraseasonal variability -Demise of 2006/07 warm event is associated with the MJO event originated in Indian Ocean in Dec 2006 - Recent westward extension of negative SST anomalies is due to enhanced westward zonal current

33. 33 Backup slides Spatial fields

34. 34 SST (K) Precipitation (mm/month) Observation 1-mon lead Fcst Observation 1-mon lead Fcst  Better forecast in the Pacific than other regions  Larger amplitude of forecast SST anomalies in central-eastern Pacific  Forecast precipitation anomalies central-eastern Pacific extend further east

35. 35 DJF 2005/2006 DJF 2006/2007  Larger SST amplitude in the EPAC  Observed Z200 and T2m represent mixture of different forcings  Forecast Z200 and T2m anomalies represent response to tropical SSTs

36. 36 MAM T2m (K) Reasonable forecast for 2005/2006 Forecasts for 2007 similar to 2006. COR = 0.87 COR = -0.58 COR = 0.52