1 / 49

Importance of Atmospheric and Oceanic Initial conditions in forecasting the MJO with the NCEP-CFS

Experiments conducted under NOAA’s Climate Test Bed. Importance of Atmospheric and Oceanic Initial conditions in forecasting the MJO with the NCEP-CFS. Augustin Vintzileos and David Behringer EMC/NCEP/NWS/NOAA - SAIC. 1. The seamless forecasting suite: from Weather to Climate.

Download Presentation

Importance of Atmospheric and Oceanic Initial conditions in forecasting the MJO with the NCEP-CFS

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Experiments conducted under NOAA’s Climate Test Bed Importance of Atmospheric and Oceanic Initial conditions in forecasting the MJO with the NCEP-CFS Augustin Vintzileos and David Behringer EMC/NCEP/NWS/NOAA - SAIC 1

  2. The seamless forecasting suite: from Weather to Climate Atmospheric initial conditions Forecast lead times 0-14 days 15-60 days 60 days and beyond Land initial conditions Synoptic Subseasonal Seasonal-to-Interannual Oceanic initial conditions Mainly affected by Atmospheric I.C. Affected by all I.C. Mainly affected by Oceanic I.C. but also by land I.C. (e.g., snow cover, soil moisture) MJO affecting weather statistics ENSO affecting weather statistics Weather 2

  3. Current global operational forecasting suite at NCEP GFS, GEFS (uncoupled) – Weather Prediction Atmospheric initial conditions CFS (coupled GFS) – Seasonal Prediction Forecast lead times 0-14 days 15-60 days 60 days and beyond Land initial conditions Synoptic Subseasonal Seasonal-to-Interannual Oceanic initial conditions Mainly affected by Atmospheric I.C. Affected by all I.C. Mainly affected by Oceanic I.C. but also by land I.C. (e.g., snow cover, soil moisture) MJO affecting weather statistics ENSO affecting weather statistics Weather 3

  4. Model Characteristics Weather Prediction: GFS = T382L64  T190L64 up to 15 days, initialized by operational NCEP analysis (GDAS), SSTs are dumped to mean seasonal values. GEFS = Ensemble forecast with GFS at T190L28 up to 15 days. Seasonal Prediction: CFS = coupled GFS at T62L64 up to month 10, initialized by Reanalysis-2, coupled to MOM3 and initialized by operational NCEP ocean analysis (GODAS) 4

  5. Open questions for subseasonal forecast: -- How important is resolution? -- How important is initialization? -- Atmospheric vs. Oceanic I.C. -- How to generate ensemble forecasts? 5

  6. BAMS September 2007 6

  7. What is the Madden - Julian Oscillation?Madden and Julian [1972]Phenomenological definition from the review by Zhang (2005): The Madden – Julian Oscillation is the dominant component of the intraseasonal (30-90 days) variability in the tropical atmosphere. It consists of large-scale coupled patterns in atmospheric circulation and deep convection, propagating eastward slowly (~5 ms-1) through the portion of the Indian and Pacific oceans where the sea surface is warm. No theoretical context explaining MJO Models have problems in representing the MJO 7

  8. Longitude – height Equatorial section 200 hPa 850 hPa Africa Indian Maritime Continent Pacific America Atlantic 8

  9. Longitude – height Equatorial section 200 hPa 850 hPa Africa Indian Maritime Continent Pacific America Atlantic 9

  10. Longitude – height Equatorial section 200 hPa 850 hPa Africa Indian Maritime Continent Pacific America Atlantic 10

  11. Why is successful forecast of the MJO so important? 11

  12. Predicting the MJO may help with prediction of precipitation in the western United States at subseasonal lead times e.g., Bond, N. and Vecchi, G.A., 2003: The Influence of the Madden-Julian Oscillation on Precipitation in Oregon and Washington. Weather and Forecasting. 18, 600Ð613. Predicting the MJO may help with prediction of Tropical Cyclone statistics at subseasonal lead times e.g.: Maloney and Hartmann, Science, 2000 Mo, Monthly Weather Review, 2000 Higgins and Shi, Journal of Climate, 2001 12

  13. MJO is also affecting ENSO ENSO forecast with the NASA NSIPP model: Vintzileos et al. (2005) Observed December 2003 A strong intraseasonal event in June 2003 changed the forecast of ENSO from strong cold to neutral conditions Forecast from March 2003 13

  14. Subseasonal forecasting with the CFS Outline: -- A metric for MJO -- Some initial forecast experiments with the CFS: The Maritime Continent Prediction Barrier -- Multi-resolution and multi-I.C. re-forecast experiments -- Conclusions and work to follow 14

  15. Verifying fields are from Reanalysis-2 • Use the zonal wind at 200 hPa from 2002 to 2006 averaged between 20°S-20°N • Compute and remove the mean annual cycle and the zonal mean • Perform and EOF analysis of the resulting field (no time filtering) Defining a metric for the MJOWe use a simplified version of the Wheeler and Hendon Index: 15

  16. First and second EOFs of the zonal wind at 200 hPa averaged between 20°S – 20°N Indian Atlantic Pacific 10 days EOF2 -EOF1 -EOF2 EOF1 r=0.6 A full oscillation in 40 days 16

  17. Reconstructed U200 vs. GPCP Precipitation, May – July, 2002 Upper level divergence 20S-20N averaged, filtered U200 anomaly field 5S-5N averaged, total unfiltered precipitation field 17

  18. Forecasting the MJO with the CFS In a first set of experiments the operational CFS at T126 was initialized 4 times per day by Reanalysis-2 and GODAS from 2000 to 2005 (Saha, Vintzileos, Thiaw, Johansen) • MJO forecast skill is obtained by: • Projecting forecast and observed fields on the two MJO EOFs • Computing the pattern correlation between forecast and observed MJO 18

  19. Forecast Skill as a function of initialization day and lead time for: May – June 2002 June 6th-9th June 6th-9th June 6th-9th 19

  20. Reconstructed U200 vs. GPCP Precipitation, May – July, 2002 Upper level divergence 20S-20N averaged, filtered U200 anomaly field 5S-5N averaged, total unfiltered precipitation field 20

  21. Reconstructed U200 vs. GPCP Precipitation, May – July, 2002 June 9th Upper level divergence 20S-20N averaged, filtered U200 anomaly field 5S-5N averaged, total unfiltered precipitation field 21

  22. Longitude – height Equatorial section 200 hPa 850 hPa Maritime Continent Africa Indian Pacific America Atlantic Observations 22

  23. Longitude – height Equatorial section 200 hPa 850 hPa Maritime Continent Africa Indian Pacific America Atlantic Observations 23

  24. Longitude – height Equatorial section 200 hPa 850 hPa Maritime Continent Africa Indian Pacific America Atlantic Observations 24

  25. Longitude – height Equatorial section 200 hPa 850 hPa Maritime Continent Africa Indian Pacific America Atlantic Model 25

  26. Longitude – height Equatorial section 200 hPa 850 hPa Maritime Continent Africa Indian Pacific America Atlantic Model 26

  27. A real time GEFS forecast example of the barrier (graphs courtesy Jon Gottschalck CPC) Observed MJO event of March 2008 is crossing the Maritime Continent Based on the Wheeler and Hendon (2004) index Forecast MJO ‘collapses’ immediately after initialization before crossing the Maritime Continent 27

  28. Horizontal resolution and atmospheric I.C.: Reforecasts: May 23rd to August 11th from 2002 to 2006, 1 forecast every 5 days Forecast lead: 60 days Model resolution: Atmosphere: T62 = 200Km x 200Km T126 = 100Km x 100Km T254 = 50Km x 50Km Ocean: the standard CFS resolution Initial conditions: Atmosphere, Land: from Reanalysis 2 (CDAS2) and from GDAS Ocean: from GODAS 28

  29. Operational GDAS versus Reanalysis-2 initial conditions: June 2002 GDAS Precipitable Water GPCP Precipitation Reanalysis 2 Precipitable Water drift Time evolution of mean energy at wave numbers 10-40 when CFS is initialized by R-2 (red) or by GDAS (blue). 29

  30. Skill for the MJO mode (verification CDAS2) GDAS Persistence forecast Skill up to 14 – 18 days Reanalysis-2 Persistence forecast GDAS T62 T126 T254 30

  31. Pattern correlation as a function of initialization day and lead time The CFS has better skill than persistence during the propagation of the dry phase of the MJO through the Maritime Continent. However, during the transition of the wet phase of the MJO through the Maritime Continent the CFS is not better than persistence June 8th June 8th 31

  32. …and the Ocean? • There is consensus that the ocean plays an important role for the evolution of the MJO • CFS is initialized by GODAS which is optimized for Seasonal-to-Interannual forecast • Its SST is damped to the weekly Reynolds SST • Contains information from 2 weeks before and two weeks after • Therefore we do not expect GODAS to represent well the intraseasonal modes. 32

  33. Standard Deviation of the 20-90 day filtered SST 2002 - 2006 As expected GODAS generally presents weaker intra-seasonal variability than observations 2002 - 2006 Intraseasonal variability increases in free runs with the coupled CFS 33

  34. Is there any relation between oceanic intra-seasonal variability and the MJO? 34

  35. First two eigenvectors of the daily observed SST correlation matrix (10% and 7% of total intraseasonal variance) 35

  36. The MJO EOFs Compare to the correlation between Principal Component 1 and Principal Component 2 of the daily OI SST and the anomalies of Zonal Wind at 200 hPa at each grid point 36

  37. The MJO EOFs There is remarkable resemblance between the U200 EOFs and the correlation of U200 anomalies and the SST Principal components 37

  38. There is an empirical relationship between the SST and the MJO suggesting that initial states for the ocean and the atmosphere should be coherent 38

  39. Ocean Initial Conditions: Re-forecasts with SET22-MOM3 instead of OP3-MOM3 May 23rd to August 11th from 2002 to 2006, 1 forecast every 5 days Forecast lead: 45 days Model resolution: Atmosphere: T126 = 100Km x 100Km Ocean: the standard CFS resolution Initial conditions: Atmosphere, Land: from GDAS Ocean: (a) from operational GODAS and (b) Experimental Ocean Analysis 39

  40. Operational Ocean Analysis Temperature Depth Weekly Reynolds SST 40

  41. Experimental Ocean Analysis Temperature Depth Daily Reynolds SST Analysis from January 2002 to December 2006 41

  42. Comparison of operational GODAS (blue) with experimental GODAS (red) The experimental GODAS clearly contains higher frequencies 42

  43. Bringing GFS from version OP3 to version SET22 After day 10 there is very slight though consistent forecast skill increase when using the improved model Up to day 10 there is no impact on forecast skill from the improved model 43

  44. We managed to inject a more realistic intra-seasonal variance to the ocean initial condition without any visible spurious effects. But… Can the coupled model retain these intra-seasonal modes or is this new information quickly lost? 44

  45. Drift of average standard deviation of intraseasonal SST as a function of lead time 45

  46. Impact of Oceanic Initial Conditions on Forecast Skill Up to day 6 the impact of atmospheric initial conditions is dominant. Even if oceanic I.C. are better there is no improvement in skill. After day 6, the improved oceanic initial conditions lead to consistently, albeit marginally, better forecast. However, during this period the intraseasonal SST modes weaken. 46

  47. Findings: Atmospheric Initial Conditions are shown to be quite important for forecasting the MJO. A better set of initial conditions improved the MJO forecast by 3-5 days. Perhaps the new CFS-Reanalysis will provide even higher quality initial conditions. However the breakthrough for MJO forecasting will come when the Maritime Barrier Prediction Barrier issue will be resolved. There is an empirical relation between intraseasonal SST modes and the MJO. This finding adds to the argument of most forecasting centers that coupling to the ocean improves forecast of the MJO and suggests that oceanic initial conditions are important for MJO forecast. Oceanic Initial Conditions were shown to improve only marginally forecast skill and that for lead times beyond day 7. However there is a relatively fast systematic decrease in amplitude of the SST intraseasonal modes which most probably affects forecast skill. Neither AIC nor OIC were capable to break through the Maritime Continent Prediction Barrier. 47

  48. Issues concerning subseasonal forecasting that we are addressing: The CFS is a useful tool for forecasting the MJO; its skill can be significantly improved by resolving the Maritime Continent Prediction Barrier. We have prioritized a number of issues that will help to further improve the skill of the model: • Advanced diagnostic studies of atmospheric processes in the hindcast experiments presented here will allow to determine reasons for the Maritime Continent Prediction Barrier. Refinement of existing or addition of missing atmospheric parameterizations will allow to break through this barrier • However the Maritime Continent is sometimes a Barrier even for the observed MJO. Determining reasons for which MJO re-organizes or not as it crosses the Indonesia region is important. Large scale intraseasonal oceanic modes could be the solution to this problem. • Currently, in the coupled model there is a decrease of the amplitude of intraseasonal modes. We believe that the ocean model is the reason for that. Therefore, improving the ability of the ocean model to simulate intraseasonal modes is high on the list of tasks. 48

  49. Questions?

More Related