Presentation on theme: "INTRODUCTION Although the forecast skill of the tropical Pacific SST is moderate due to the largest interannual signal associated with ENSO, the forecast."— Presentation transcript:
INTRODUCTION Although the forecast skill of the tropical Pacific SST is moderate due to the largest interannual signal associated with ENSO, the forecast skill of the tropical Indian and Atlantic SST is still poor. This might be due to the fact that the interannual signal is weak and observations are sparse there. However, ENSO has significant impacts on the SST variability outside the tropical Pacific, which can be forecast indirectly through forecast of ENSO. The Markov model, based upon observations in the tropical Pacific, has a competitive forecast skill for ENSO. Here is an attempt to extend the Markov model to the global tropics so that it not only forecast the tropical Pacific SST associated with ENSO but also ENSO’s impacts on SST variability outside of the tropical Pacific. Prediction of Global Tropical SST Using a Markov Model and Comparison with NCEP’s CFS Forecast Yan Xue Climate Prediction Center/NCEP/NOAA, email@example.com DATA ERSST version 2, 1979-2000, monthly, 2 o x2 o, 20 o S-20 o N Sea level from SODA version 1.2, 1979-2000, monthly, 2 o x2 o, 20 o S-20 o N CMAP precipitation, 1979-2000, monthly, 2 o x2 o, 20 o S-20 o N Reanalysis-2 wind stress, 1979-2000, monthly, 2 o x2 o, 20 o S-20 o N METHODOLOGY Remove annual cycle in 1979-2000 Calculate multiple EOFs of SST and sea level with equal weight Calculate associated patterns of precipitation and wind stress with MEOFs Use Principal Components of MEOFs to construct 12 transition matrixes --- Markov model (Xue et al. 2000) Determine number of PCs to retain in Markov model with cross validation Take one year data out sequentially Build Markov model with remaining data, and forecast the year that is taken out Compare hindcast skill in 1982-2000 with NCEP’s Climate Forecast System (CFS) COMPARISON with NCEP’s CFS FORECAST CONCLUSIONS The warm biases in the Global Ocean Data Assimilation System (GODAS, see poster P2.9) before 1990 hurts the CFS’s hindcast skill in the western Pacific, Indian and Atlantic oceans significantly. The hindcast skill of the Markov model is superior to that of CFS when the model’s climatology in 1982-2000 is removed, while it is comparable to that of CFS when the model’s means in 1982-1990 and 1991-2000 are removed in addition to removing the model’s climatology in 1982-2000. The hindcast skill of the tropical Pacific SST is the lowest in summer due to spring barrier. CFS simulates the recovery of skill in fall in the central-eastern Pacific, but it fails to simulate the recovery of skill in the north-western Pacific. The hindcast skill of the tropical Indian SST is excellent in later winter and spring, and that of the north-western Atlantic is modest in spring, while that of the tropical Atlantic is generally poor. CFS’s forecast has too strong variability in the equatorial western and north- western Pacific, and has significant cold biases since 1999. Six MEOFs appear capture the most significant air-sea coupled modes in the global tropics. P1.16 With 22 degrees of freedom (1979-2000), MEOF 1 and 2 are significant, describing the mature and onset phases of ENSO; MEOF 3 and 4 are mixed, describing asymmetry between warm and cold events and the Atlantic Nino; MEOF 5 and 6 are significant, describing the Atlantic meridional mode, and Southern Ocean variability. Six MEOFs together account for 90% of SST variability in the central-eastern Pacific, 60% in the tropical western Pacific, 50% in the tropical Indian Ocean (IO) except in the southern-eastern IO, and 60% in the north-western and south-eastern Atlantic Ocean. Six MEOFs together account for 80-90% of sea level variability in the equatorial Pacific, 50% in the southern-western Indian Ocean, and 20-30% in the equatorial Atlantic Ocean, which is attributed to the small variability there. Although the SST and sea level variability in the southern Indian Ocean are moderate, they are not represented well by six MEOFs. Due to the pre-1990 warm biases in the Global Ocean Data Assimilation System (see poster P2.9) that is used to initialize the oceanic component of CFS, the CFS’s hindcast skill is evaluated in two ways. One way is to remove the model’s climatology in 1982-2000 (referred as CFS) and another is to remove the model’s means in 1982-1990 and 1991-2000 in addition to removing the climatology in 1982-2000 (referred as CFS_biascorr). Markov model has a superior hindcast skill compared to CFS outside of the tropical Pacific, but its superiority diminishes significantly compared to CFS_biascorr. NINO3.4 (170 o W-120 o W, 5 o S-5 o N), WSST (120 o E-170 o W, 5 o S-5 o N), NWSST(120 o E-160 o E, 5 o N-20 o N), IND (48 o E- 104 o E, 6 o S-6 o N), TNA(80 o W-40 o W,7 o N-20 o N),ATL (40 o W-8 o E, 6 o S-6 o N). MEOF SPACE REDUCTION Xue, Y., A. Leetmaa, and M. Ji, 2000: ENSO prediction with Markov models: The impact of sea level J. Climate, 13, 849-871.