1. CFSv2 Research at COLA: Understanding the effect of air-sea coupling in Asian-Pacific monsoon prediction and improving sea ice simulation for climate study Bohua Huang 1,2 and Jieshun Zhu 2 1 Department of Atmospheric, Oceanic, and Earth Sciences, George Mason University 2 Center for Ocean-Land-Atmosphere Studies (COLA) 1 Acknowledgments: J. Shukla, J. Kinter, L. Marx, C.-S. Shin, Z.-Z. Hu, X. Wu, A. Kumar The Second Taiwan West Pacific Global Forecast System Planning Workshop CWB, Taipei, Taiwan, May 7-8, 2014

2. An Overview Multiple Analyses Ensemble Ocean Initialization ENSO prediction skill and reliability (Zhu et al. 2012, 2013a) US summer monsoon precipitation (Zhu et al. 2013b) Indian Ocean SST and Asian monsoon prediction (in progress) Effects of air-sea coupling in Asian-Pacific monsoon prediction Decadal prediction: reducing climate drift by improving sea-ice simulation Diagnostic analyses Simulation/hindcast of South Pacific dipole mode (Guan et al. 2014a,b) Fast annual cycle in monsoon (Shin and Huang, in preparation) 2

3. Air-sea coupling in monsoon prediction A comparison of One- and Two-Tier Approaches One-Tier: CGCM for seasonal prediction Relatively new: operational since 2000s (Palmer et al. 2004; Saha et al. 2006;…) CGCM bias affects quality of prediction (e.g., tele-connection) Computationally expensive (trade-off between resolution and physics) Two-Tier: Tier-1: SST prediction Tier-2: AGCM with predicted SST forcing (Bengtsson et al 1993) Yielding boundary-forced predictability Avoid CGCM bias Achieving computational efficiency (or higher resolution) A useful research tool (e.g., climate downscaling, signal/noise separation, etc) 3 An issue with Tier-2: Lack of air-sea interaction In particular, atmospheric feedback to SST is vital over monsoon regions. Without feedback, SST forcing is excessive in Tier-2 (e.g., Fu et al. 2002; Wu and Kirtman 2004, 2005, 2007; Wu et al. 2006; Wang et al. 2005; Cherchi and Navarra 2007; Chen et al. 2012; Hendon et al. 2012; Hu et al. 2012… ) Few studies have compared the one tier and two tier systems in a prediction mode ( Kug et al. 2008; Kumar et al. 2008 ): Less controlled experiments. Zhu, J. and J. Shukla, 2013: The Role of Air–Sea Coupling in Seasonal Prediction of Asia–Pacific Summer Monsoon Rainfall. J. Climate, 25, 5689-5697, doi:10.1175/JCLI-D-13-00190.1.

4. Forecast Model: NCEP CFS version 2 1)Atmosphere: T126, L64 2)Ocean (MOM4): 0.5°x0.5° (0.25° lat, 10°S-10°N), L40 3) Coupling: every half hour Experiment design (Coupled vs. Uncoupled) 1) An identical AGCM is used in Tier-1 and Tier-2 predictions 2) Tier-1: Coupled run based on CFSv2 Tier-2: Daily mean SSTs from Tier-1 are prescribed as boundary conditions Initial Conditions (1982-2009) 1)Tier-1: Ocean IC: Instantaneous states from ECMWF ORA-S4 2)Tier-1/2: Atmosphere/Land IC: CFSR (4-member, Apr 1-4) Model and Experiments 4 One-Tier vs. Two-Tier: 6-month hindcasts starting from April (1982-2009)

5. 5 (1982-2009) mm/day

6. Averaging first, Then Calculating Sdev. 6 Standard Deviation (1) (1982-2009) mm/day

7. 7 Standard Deviation (2) (1982-2009) Calculating Sdev first, Then averaging. mm/day

8. 8 (1982-2009) mm/day

9. 9

10. Large differences occur in monsoon regions

11. because of large mean rainfall? Then why not in Atlantic?

12. Is warm pool temperature a factor? (a warm water phenomenon?)

13. Leading Modes of Precipitation CMAP CGCM AGCM PC 25.0% 44.7% 27.7% O-C: 0.80 O-A: 0.77 10.8% 13.6% 16.7% O-C: 0.53 O-A: 0.51 A-C: 0.88 A-C: 0.60 8.8% 5.6% 10.5% O-C: 0.25O-A: 0.41 A-C: 0.38

14. Leading Modes of Precipitation CMAP CGCM AGCM PC 25.0% 44.7% 27.7% O-C: 0.80 O-A: 0.77 10.8% 13.6% 16.7% O-C: 0.53 O-A: 0.51 A-C: 0.88 A-C: 0.60 8.8% 5.6% 10.5% O-C: 0.25O-A: 0.41 A-C: 0.38

15. 15 CGCM AGCM A-C Surface Heat Balance

16. 16 CGCM AGCM A-C Surface Heat Balance Lack of SST feedback

17. 17 Atmosphere forces ocean CGCM AGCM A-C Lack of SST feedback

18. 18 An Example

19. 19 (1982-2009) mm/day

20. 20

21. 21

22. 22 (1982-2009)

23. Summary I 1) In the absence of air–sea coupling, Tier-2 predictions produce higher rainfall biases and unrealistically high rainfall interannual variations. 2) The prediction skill of precipitation, as measured by anomaly correlation, does not show significant differences between the two types of predictions. Observed leading modes can be reproduced by both to a certain degree. 3) RMSEs are significantly larger for the AGCM (Tier-2) predictions compared to the CGCM (Tier-1) predictions. 4) The reduced RMSE skills in the Tier-2 predictions are due to a lack of coupled surface feedback to the prescribed SST anomalies, which, particularly, results in an unrealistic SST-rainfall relationship over the TWNP region. 23 Incorporating regional ocean–atmosphere feedback is important for rainfall prediction over the Asia–Pacific region.

24. Improving sea ice simulation for climate study In the framework of seamless prediction, there is merit in extending operational seasonal prediction systems, such as CFSv2, to predicting decadal variability and climate change (The extension is not as straightforward as it seems) Long-term integration/prediction tests CFSv2 behavior on wider ranges with new targets. The resulted model improvements may benefit seasonal prediction Achieving both goals simultaneously is not an easy task. Research institutes, such as COLA, can play a useful role in bridging the two 24 Huang, B., J. Zhu, L. Marx, X. Wu, A. Kumar, Z.-Z. Hu, M.A. Balmaseda, S. Zhang, J. Lu, E.K. Schneider and J.L. Kinter, 2014: Climate Drift of AMOC, North Atlantic Salinity and Arctic Sea Ice in CFSv2 Decadal Predictions. Clim. Dyn. submitted.

27. Red contours: zonal stress provided by AGCM Blue contours: zonal stress “seen” by OGCM Black curves: true sea ice range The discrepancy in some ice-free regions is due to a misidentification of sea-ice index in coupler 30-year simulations

28. Corrected sea ice cover reduces warm SST bias in summer

29. AMOC is enhanced but still weaker  than observations

30. Sea Ice Concentration

34. Sensitivity Experiment Adjusting (1) Sea-ice albedo (2) Dry-wet ice transition temperature

35. 35

36. Summary II 1) A major climate drift occurs in the CFSv2 decadal hindcasts, with the weakening of AMOC, reduction of North Atlantic salinity and thinning of the Arctic sea ice. 2) Among other factors, the melting sea ice generates excessive freshwater in the Arctic Ocean that can be transported to the North Atlantic. 3) Adjusting sea ice albedo parameters produces a sustainable ice cover with realistic thickness distribution, enhancing AMOC moderately. 4) Improved CFSv2 sea ice simulation also reduces the warm SST bias in the North Pacific during summer. A more realistic freshwater balance may lead to a major improvement of CFSv2

37. *Removed the model bias in (b) Slow Annual Cycle OBSCFSv2 OBS Total Annual Cycle How well is the Slow Annual Cyrcle simulated by CSFv2? OLR in the East Asian Monsoon Region (120E-140E) target period & area for FAC

38. OBS CFSv2 How is the Fast Annual Cycle simulated by CSFv2? FAC of OLR (120E-140E) Pre Monsoon Dry Phase Pre Mei-yu Wet Phase Mei-yu front Subtropical ridge Western North Pacific Monsoon Monsoon Gyre Repeating dry and wet phases are very well simulated in the CFSv2, i.e., Pre-Monsoon Dry Phase, Pre-Meiyu Phase, Grand Onset, and Monsoon Gyre. Northward propagating signal for both dry and wet phases are also well simulated in the CSFv2. However, the intensity is much weaker in the CFSv2 and the timing of both phases are later than observation.

39. Zhu et al. (GRL, 2012) 39

40. 40