|
An El-Nino Prediction Model ( Intermediate Ocean - Statistical Atmosphere Coupled Model )
An El Nino prediction model is developed based on an intermediate ocena model similar to the Can and Zebiak (CZ) and a statistical atmosphere Model (Hereafter called ISCM). Atmosphere model is statistical model using the relationship between wind stress and SST. The present ocean model differs from CZ in the parameterization of subsurface temperature and the basic state. The parameterization of subsurface temperature is replaced by a statistical relationship constructed based on the SVD singular vectors of the themocline depth and subsurface temperature of NCEP assimilation data. The basic state of CZ model is modified to consider more recent climatology.
2. Atmosphere Model (Schematic Diagram) A statistical atmosphere model is developed based on the singular value decomposition (SVD) of wind stress and SST The atmosphere model produces wind stress using model produces SST and SVD results. The zonal and meridional wind stresses are separately computed based on SVD. The present statistical model uses the two leading singular vectors since other singular vectors explain negligible fractions of covariance. Experiments with more singular vectors produces similar results. It is also noted that the first and second modes explain the mature and transition phases of ENSO, respectively
3. The Predictability The present model combines the observed SST and wind stress in the initialization. The initial conditions are produced for each month from January 1980 to December 1998. Starting from the initial condition, the present model is integrated up to 24 months Figure 1 shows observed (thick curve) and forecasted (thin curve) Nino3 index. Each forecast is 12 months long, starting from the beginning of each month. The ISCM could forecast the interannual SST variation of the eastern Pacific such as El Nino and La Nina, comparatively. The prediction results of the ISCM re compared with those of observation and the CZ model with an initialization of Chen et al. (1995). This version of CZ model is called to as LDEO2 [Chen et al., 1998]. Figure 2 compares the predictability skills of the present model and LDEO2. The skills are measures by the correlation and the RMS error between monthly model forecast and observed SST anomalies in the tropical Pacific. For the Nino3 SST case, the present model gives a higher correlation and lower RMS error in most of lead times, particularly in the earlier lead times before one year. The better performance of the present model is more distinctive when the correlation and RMS are compared for the Nino4 SST, the anomaly averaged over the central equatorial Pacific (Figs. 2c and 2d). Noted is that the present model has a predictive skill, the correlation exceeding 0.5, for both Nino3 and Nino4 up to 2 yrs. It is also noted that the correlations over tropical central and eastern Pacific exceed 0.7 for a 6 month lead and 0.5 for a 12 month lead. Comparing to LDEO2, the present model has a better predictability skill (with a correlation about 0.1 larger than that of LDEO2) in most of the domain, particularly in the central tropical Pacific. It is pointed out that the better performance of SST prediction in the central Pacific is mainly due to a better representation of subsurface temperature in the present model. A most striking difference between the present model and LDEO2 is in their forecasts of the 1997/98 El-Nino, as illustrated in Fig. 3. LDEO2 totally missed the onset of this big event, and the initial conditions were much weaker than the observation. On the other hand, the present model produced a better prediction. The initial condition followed observation closely. The model predicted the warming in early 1997 when the SST anomaly is very weak, and the rapid decay of warm condition after late 1997 is well captured by the model. . |
