1. Evaluation of Tropical Pacific Observing Systems Using NCEP and GFDL Ocean Data Assimilation Systems Y. Xue 1, C. Wen 1, X. Yang 2, D. Behringer 1, A. Kumar 1, G. Vecchi 2, A. Rosati 2, R. Gudgel 2 1 NCEP/NOAA, USA 2 GFDL/NOAA, USA CTB Seminar, November 13, 2015, NCWCP, College Park, MD Xue et al. 2015: Evaluation of Tropical Pacific Observing Systems Using NCEP and GFDL Ocean Data Assimilation Systems. Climate Dynamics, Online

2. What is Operational Ocean Reanalysis? Ocean Model Reanalysis Surface Fluxes 19792015 Data Assimilation Scheme (optimally combine model solutions and ocean observations) Ocean Observations MooringsAltimeter SST Argo XBT - Initialization for dynamical seasonal predictions - Providing historical context for ENSO variability - Real-time operational ENSO monitoring

3. NCEP Global Ocean Data Assimilation System (Ocean-alone, Implemented in 2003) Global Ocean Data Assimilation System (GODAS) Climate Forecast System (CFSv1) SST XBT Moorings Altimeter Argo R2 Surface Fluxes SST Anomaly Forecast Forecasters Official ENSO Forecast Official Probabilistic Surface Temperature & Rainfall Forecasts Seasonal Forecasts for North America CCA, CA Markov CCA, OCN MR, ENSO Ocean Initial Conditions IRI

4. Ocean Monitoring Products Based on GODAS (A partnership between CPC and COD/CPO to deliver climate relevant products to the society) Synthesis of global ocean observations by NCEP’s Global Ocean Data Assimilation System (GODAS) Monthly Ocean Briefing since May 2007 Products used widely by operational climate prediction centers, researchers, fishery managers, news media, program managers, teachers and students Contact: Yan Xue, NOAA/CPC Synthesis of Ocean Observations http://www.cpc.ncep.noaa.gov/products/GODAS

5. 5 Is the Quality of Ocean Reanalysis Sensitive to Data Distribution? Xue et al. 2012 1980-1993 1994-2003 2004-2009

6. 6 Black: All data Red: TAO/TRITON Blue: XBT Green: Argo Recent Challenge in TAO/TRITON Mooring Array 8S-8N Jun 2012 Nov 2014 TAO Argo What are influences of missing TAO data on ENSO monitoring and ENSO prediction?

8. Courtesy of Neville Smith & Billy Kessler Challenges in ENSO Monitoring and Prediction

9.  Extend CLIVAR-GSOP/GODAE OceanView Ocean Reanalyses Intercomparison Project (ORA-IP) into real-time  Deliver ensemble ocean monitoring products in real time  Quantify uncertainties in the ocean state estimation for the purpose of ENSO monitoring and prediction  Understand how variations in observing systems influence uncertainties in ocean reanalyses  Provide support for the TPOS 2020 project on the design of the future tropical Pacific observing system  Asses how NCEP ocean reanalyses compare with other state-of-art ocean reanalyses Yan Xue Climate Prediction Center9 A Real-time Ocean Reanalyses Intercomparison Project for Quantifying Uncertainties in ENSO State (Motived by TPOS Workshop in Jan. 2014)

10. NameMethod & Forcings In Situ Data Altimetry Data ResolutionPeriodVintageReference NCEP (GODAS) 3D-VAR T,/SSTNO (Yes since 2007) 1°x 1° (1/3° near Eq)1979- present 2003Behringer and Xue (2004 ECMWF (S4) OI T/S/SSTYes1°x1° (1/3° near Eq)1959- present 2011Balmaseda et al. (2012) JMA3D-VAR T/S/SSTYes1°x1° (1/3° near Eq)1979- present 2015Usui et al. (2006) GFDL (ECDA) EnKF coupled T/S/SSTNo 1°x 1° (1/3° near Eq)1970- present 2010Zhang et al. (2009) NASAEnOI Partially coupled T/S/SSTYes1/2°x 1/2° (1/4° near Eq) 1980- present 2011Rienecker at al. (2011) BOM (PEODAS) EnKFT/S/SSTNo2°x 1.5 ° (1/2° near Eq.)1980- preesnt 2009Yin et al. (2010) NCEP (CFSR) 3D-VAR Coupled T/SSTNo1/2°x 1/2° (1/4° near Eq) 1979- present 2010Xue et al. (2011) UK Met (GLOSEA5) 3DVART/S/SSTYes1/4°x 1/4°1993- present ??Waters et al. (2014) MERCATOR (GLORYS2) KF-SEEKT/S/SSTYes1/4°x 1/4°1993- present ?? Operational Ocean Reanalyses 7 products from 1979-present 9 products from 1993-present

11. http://www.cpc.ncep.noaa.gov/products/GODAS/multiora_body.html (based on 1981-2010 climatology)

12. http://www.cpc.ncep.noaa.gov/products/GODAS/multiora93_body.html (based on 1993-2013 Climatology) In development

14. 14 Red: TAO/TRITON Blue: XBT Green: Argo Black: All Data Tropical Pacific Observing Systems 2004-2011 8S-8N

15. 15 CTLALLnoMoornoArgo In situ data included no profiles all profiles all except moorings all except Argo XBT ×√√√ TAO/TRITON ×√×√ Argo ×√√× Coordinated Observing System Experiments at NCEP and GFDL (2004-2011)  What are the mean biases and RMSE (monthly anomalies) in the control simulations when no in situ observations are assimilated?  How well are mean biases and RMSE constrained by assimilation of all in situ observations?  What are influences of withholding mooring or Argo data on mean biases and RMSE?

16. NCEP’s Global Ocean Data Assimilation System MOM4p1, 0.5 o resolution, 1/4 o in 10 o S-10 o N, 40 levels Daily fluxes from NCEP Reanalysis 2 3D-VAR, Univariate in temperature and salinity (Behringer 2007) Horizonal covariance with scale elongated in zonal direction, vertical covariance as function of local vertical temperature gradient Temperature profiles and OISST assimilated Synthetic salinity derived with climatology T/S relationship and temperature profiles from XBT and moorings, observed salinity from Argo assimilated Altimetry SSH not assimilated Temperature (salinity) at 5m is relaxed to daily OISST (Levitus salinity climatology) with 10 (30) day relaxation time scales.

17. GFDL’s Ensemble Coupled Data Assimilation System MOM4, 1 o resolution, 1/3 o in 10 o S-10 o N, 50 levels Coupled model CM2.1 Ensemble Kalman Filter with 12 ensembles (Zhang et al 2007) Flow-dependent ensemble covariance of temperature and salinity 3-D air temperature and wind from NCEP renalyses assimilated Observed temperature and salinity profiles and observed SST assimilated Altimetry SSH not assimilated

18. Evaluation of OSEs Validation data – TAO temperature and current – Altimetry SSH from AVISO – Surface current from OSCAR – EN4 temperature and salinity analysis (objective analysis based on in situ data only, used as reference but not the “truth”) Evaluation Methods – Mean and annual cycle – Standard deviation (STD) – Root-mean-square error (RMSE) and anomaly correlation coefficient (ACC) – Integrated RMSE in upper 300m

19. All in situ obs. included NCEP GFDL Mean Upper 300m Heat Content (HC300) Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

20. All in situ obs. included NCEP GFDL Mean Temperature at Eq Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

21. All in situ obs. included NCEP GFDL STD of Temp. Anom. at Eq Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

22. 22 Temp. Anom. at 130 o W-100 o W, 2 o S-2 o N at depth of 75m Black dotted line: Obs Black line: ALL Red line: noMoor Green line: noArgo Blue line: CTL GODAS is very sensitive to withholding TAO data in the far eastern eq. Pac., while ECDA is not.

23. All in situ obs. included NCEP GFDL Mean Upper 300m Heat Content (HC300) Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

24. All in situ obs. included NCEP GFDL Mean Temperature at 130 o W-100 o W Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

25. All in situ obs. included NCEP GFDL STD of Temp. Anom. at 130 o W-100 o W Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

26. 26 11 o N-12 o N, 130 o W-100 o W

27. 27 L x =12 o L y =8 o Variance of Temp. Increment 11 o N GODAS relies on local data to correct model bias. When there is no local data at 11N when Argo data are withheld, GODAS draws influences from nearby data such as TAO at 8N, which causes very large analysis increment because of large model drift.

28. 28 10 Day Window 2 Day Window ECDA GODAS Reducing data assimilation window from 10 to 2 days reduces the strong fitting to observations, and therefore too strong sensitivity to withholding Argo data.

29. All in situ obs. included NCEP GFDL RMSE of HC300 Anomaly Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

30. All in situ obs. included NCEP GFDL Mean Salinity at Eq Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT

31. All in situ obs. included NCEP GFDL STD of Salt Anom. at Eq withholding TAO withholding Argo withholding TAO, Argo and XBT

32. All in situ obs. included NCEP GFDL RMSE of Salt Anom. at Equator Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT TRITON

33. All in situ obs. included NCEP GFDL RMSE of Salt Anom. in Upper 300m Impacts of withholding Moorings Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT TRITON

34. 34 Mean Zonal Current at Eq 165E 170W 140W 110W For both GODAS and ECDA, assimilation of temperature and salinity observations do not improve the equatorial current analysis. More advanced data assimilation schemes are needed to improve current analysis near the equator where density constraints on current are weak.

35. All in situ obs. included NCEP GFDL Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT Mean Surface Zonal Current For both GODAS and ECDA, assimilation of temperature and salinity observations improves little mean surface zonal current analysis.

36. All in situ obs. included NCEP GFDL Anomaly Correlation of Surface Zonal Current with OSCAR Impacts of withholding TAO Impacts of withholding Argo Impacts of withholding TAO, Argo and XBT For both GODAS and ECDA, inclusion of Argo data significantly improves surface zonal current variability.

37. Summary of NOAA OSE Project  Without assimilation of any in situ data, both GODAS and ECDA had large mean biases, STD biases and RMSE.  Assimilation of in situ data significantly reduced mean biases, STD biases and RMSE in all variables except zonal current at equator.  For constraining temperature analysis, the TAO/TRITON and Argo is complimentary in the equatorial Pacific, and the Argo is critical in off- equatorial regions.  For constraining salinity, sea surface height and surface current analysis, the influence of Argo data is more important. The influence of the TRITON data on salinity is critical.  GODAS (ECDA) is sensitive to withholding TAO data in the far eastern (western) eq. Pacific. GODAS is more sensitive to withholding Argo data in off-equatorial regions than ECDA. The results suggest that multiple ocean data assimilation systems should be used to assess sensitivity of ocean analyses to changes in the distribution of ocean observations.

38. 38 Plan for future OSE Simulations (2004-2015) XBTMoorArgo CTL xxx ALL √√√ noMoor √x√ noArgo √√x XBT √xx Moor x√x Argo xx√ noXBT x√√ EqMoor x √ (2S, Eq, 2N √

39. Questions to be Addressed  Differences of XBT/Moor/Argo from CTL assesses the value of individual observing system on reducing biases and RMSE in CTL  Differences between noXBT and ALL assesses the impacts of removing XBT when the mooring and Argo data are present  Differences between Argo and noMoor or between Moor and noArgo or between ALL and noXBT assesses the value of XBT  Differences between EqMoor and noXBT assesses the value of off-equatorial moorings when the Argo and equatorial mooring data are present.  Sensitivity to the new irium Argo floats deployed near the equator since Jan 2014

40. 40 Single Profile OSE Experiments OSE assimilating one profile (OneProf) -Start from CTL and assimilate one profile only -Compare OneProf with CTL -See how long the model memorizes the impacts of single profile  help design observation sampling in space and time

41. 41 Thanks!