Multiple Ocean Analysis Initialization for Ensemble ENSO Prediction using NCEP CFSv2 B. Huang 1,2, J. Zhu 1, L. Marx 1, J. L. Kinter 1,2 1 Center for Ocean-Land-Atmosphere Studies 2 Department of Atmospheric, Oceanic, and Earth Sciences George Mason University M. A. Balmaseda European Centre for Medium-Range Weather Forecasts R.-H. Zhang Earth System Science Interdisciplinary Center, University of Maryland Z.-Z. Hu Climate Prediction Center, NCEP/NOAA 1
Multiple-ocean Analysis Ensemble (MAE) ORA-S4, ECMWF (Balmaseda et al. 2012) COMBINE-NV, ECMWF (Balmaseda et al. 2010) ORA-S3, ECMWF (Balmaseda et al. 2008) CFSR, NCEP (Saha et al. 2010) GODAS, NCEP (Behringer 2005) SODA2.1.6, UM/TAMU (Carton and Giese 2008) ECDA, GFDL (Zhang et al. 2007) Cause of Uncertainty Sparse and uneven distribution of obs. (structural uncertainty) Different forecast systems Different assimilation schemes Different surface and ocean data inputs
Heat Content Anomaly moderate high low Zhu et al. (Clim. Dyn., 2012) ODA Heat Content Uncertainty (1979-2007) DATA SOURCE ECMWF: ORA-S3, COMBINE-NV NCEP: GODAS, CFSR UM/TAMU: SODA GFDL : ECDA 3
Scientific Questions What are the effects of uncertainty in upper ocean heat content on seasonal-tointerannual (SI) prediction? Will ocean initialization with multipleocean analysis ensemble (MAE) improve SI predictive skill in tropics? 4
An Example of Ocean HC Uncertainty CFSR HCA ECMWF HCA Heat content anomaly (HCA) from ODA analyses shows high uncertainty 1st EOF modes from different analyses give different patterns Projection spread is large among analyses Zhu et al. (Clim. Dyn., 2012)
Why MAE? Ensemble average reduces noise effectively Leading EOF patterns become physically meaningful S/N ratio improves significantly Signal exists in all analyses (masked by high internal noise) Zhu et al. (Clim. Dyn., 2012)
Heat Content Anomaly Uncertainty is relatively low in Tropical Pacific 7
Heat Content Anomaly Uncertainty is relatively low in Tropical Pacific Slight phase shift 8
Heat Content Anomaly Uncertainty is relatively low in Tropical Pacific Slight phase shift But not negligible in prediction 9
Experiment Design 12-month hindcasts initialized in April 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) A discontinuity in surface fluxes over high-lat. North Atlantic (tropical effect negligible) MAE Initialization Experiments (1979-2007) 1) Ocean initial condition (OIC): Monthly means from COMBINE-NV, ORA-S3, CFSR, GODAS 2) Anomaly initialization in OIC 3) Perturbed Atmosphere-land IC (4-member with each OIC, Apr 1-4, CFSR) Additional Hindcast Experiments 1) AVEoci --- Average OIC of COMBINE-NV, ORA-S3, CFSR, GODAS 2) ORA-S4 ---instantaneous OICs from ORA-S4 (1982-2009) with full Initialization 3) CFS Reanalysis and Reforecast (CFSRR, Provided by NCEP, 9-month, 24-member, 1982-2009) 10
Jun Sep Dec Zhu et al. (GRL., 2012) 11
Prediction skill of the Nino3.4 is sensitive to OICs (April ICs: 1979-2007) Predictive skills of individual OICs have substantial differences ES_Mean is comparable to the best of individual predictions Perturbing OICs gives a better ensemble spread than perturbing AICs only 12
CFSR initial states seem slightly different from others 13
Ensemble Mean OIC vs Ensemble Ocean Prediction AVEoci Features 1) Ensemble mean OIC from COMBINE-NV ORA-S3 CFSR GODAS 2) Anomaly Initialization 3) 4 ensemble members (April ICs: 1979-2007) Ensemble ocean prediction is superior to ensemble mean OIC 14 NINO3.4 Prediction Skill
Q1: Does anomaly initialization help? Maybe Q2: Does monthly mean OIC lower skill? No (April ICs: 1982-2007) CFSRR (NCEP) vs. CFSR (COLA) Differences in Initialization 1) Full vs. Anomaly 2) Instantaneous vs. Monthly 3) Ensemble sizes: 24 vs. 4 ORA-S4 vs. COMBINE-NV Differences in Initialization 1) Full vs. Anomaly 2) Instantaneous vs. Monthly 3) ORA-S4 is more updated 15 NINO3.4 Prediction Skill
Another Region of Potential Gain from MAE Initialization Southwestern Indian Ocean SST Prediction, JJAS 1982-2007 MAE initialization achieves higher skill than individual OIC cases and CFSRR near Madagascar 16
HC Subsurface Memory Ensemble mean EOF patterns similar to individual analyses The subsurface projection of Indian Ocean dipole mode Westward propagation of offequatorial Rossby waves
MAE Effects on Rainfall Prediction LD=2-5 Mons
LD=2-4 Mons Model shows some skill in the northwestern US Zhu et al. (Clim. Dyn., 2013)
Correlation with SSTA in JJA The enhanced precipitation is associated with ENSO and PDO Model overestimates positive correlations with Indian and Atlantic Oceans
Summary There is considerable uncertainty in upper ocean heat content anomalies from different ocean analyses The uncertainty in ocean initial state causes a noticeable spread in ENSO prediction Multiple-ocean analysis ensemble (MAE) initialization improves ENSO prediction skill and reliability Southwestern tropical Indian Ocean SST prediction benefits from MAE initialization Precipitation prediction using the MAE initialization shows improvement in certain areas, probably due to improved ENSO forecast 21
References Zhu, J., B. Huang, L. Marx, J. L. Kinter III, M. A. Balmaseda, R.-H. Zhang, and Z.-Z. Hu, 2012: Ensemble ENSO hindcasts initialized from multiple ocean analyses. Geophy. Res. Lett., 39, L09602, doi:10.1029/2012gl051503. Zhu, J., B. Huang and M. A. Balmaseda, 2012: An ensemble estimation of the variability of upper-ocean heat content over the tropical Atlantic Ocean with multi-ocean reanalysis products. Clim. Dyn., 39, 1001-1020. Zhu, J., B. Huang, Z.-Z. Hu, J.L. Kinter III, and L. Marx, 2013: Predicting US summer precipitation using NCEP Climate Forecast System version 2 initialized by multiple ocean analyses. Clim. Dyn.,. DOI: 10.1007/s00382-013-1785-x (in press). Zhu, J., B. Huang, M. A. Balmaseda, J. L. Kinter III, P. Peng, Z.-Z. Hu, and L. Marx, 2013: Improved reliability of ENSO hindcasts with multi-ocean analyses ensemble (MAE) initialization. Clim. Dyn., in revision. 22
23
CFSv2 SST Climatology and Standard Deviation