Figure 8.21: Comparison of eigenvectors for the leading EOFs of the SSTs between the ENSO time-scale (<12 years) (upper panels) and the decadal time scale (>12 years) (lower panels) for (a) observation, and (b) the MRI coupled climate model, respectively (Yukimoto, 1999). Numbers in bracket at the upper left show explained variance in each mode.

Figure 8.22: Niño-3 SST anomaly predictions and hindcast made at various times during the 1997 to 1998 El Niño event together with the subsequent observed SST anomaly (solid). Predictions made at ECMWF (Stockdale et al., 1998, long dash), JMA (Ishii et al., 1998, short dash), NCEP (Barnston et al., 1999, long short dash) and the hindcast made at BMRC (Wang et al., 2000, dot dash) are shown.

The atmosphere-ocean coupled system shows various modes of variability that range widely from intra-seasonal to inter-decadal time-scales (see Chapters 2 and 7). Since the SAR, considerable progress has been achieved in characterising the decadal to inter-decadal variability of the ocean-atmosphere system (Latif, 1998; Navarra, 1999). Successful evaluation of models over a wide range of phenomena increases our confidence.

ENSO is a phenomenon resulting from large-scale air-sea inter-actions (see Chapter 7, Section 7.6.5). ENSO modelling has advanced considerably since the SAR (e.g., Yukimoto et al., 1996; Kimoto and Shen, 1997; Knutson et al., 1997; Timmermann et al., 1998). Some models now use enhanced horizontal resolution in the tropics to better resolve equatorial ocean dynamics. Models show SST variability in the tropical Pacific, which has some similarity to observed ENSO as is shown in upper panels of Figure 8.21. However, some aspects of ENSO are still not well captured by present day coupled models (Delecluse et al., 1998). Latif et al. (1999) analysed the SST climatology and interannual variability simulated by twenty four models in the equatorial Pacific. When compared with observations, the models have flaws in reproducing the annual cycle. About half of the models are characterised by too weak interannual variability in the eastern equatorial Pacific, while models generally have larger variability in the central equatorial Pacific (Table 8.2). It was found that the majority of the models show the observed ENSO-monsoon relationship, that is, a weak Indian summer monsoon tends to be associated with El Niño.

Seasonal forecasting with coupled global models has just begun (Barnston et al., 1999; McPhaden, 1999), although few of the models discussed in Chapter 9 are used. While forecast skill of coupled global models is still lower than statistical models (Landsea and Knaff, 2000), coupled global models have better skill than simple models. The 1997 to 1998 El Niño event (Trenberth, 1998b) is a good test of coupled model forecast systems. Figure 8.22 plots the SST anomaly during the 1997 to 1998 El Niño for predictions made with various initial conditions by prediction systems at ECMWF (Stockdale et al., 1998), the Japan Meteorological Agency (JMA) (Ishii et al., 1998), NCEP (Barnston et al., 1999) and for the hindcast made at the Bureau of Meteorology Research Centre (BMRC) (Wang et al., 2000). Those comprehensive models predicted unusually warm tropical Pacific SST for 1997, albeit with underestimation of the strength of the event and the warming speed. A similar conclusion is reached with other global climate models (Oberhuber et al., 1998; Zhou et al., 1998). Unusually strong Madden-Julian Oscillation (MJO, see Section 8.7.4) and westerly wind bursts may have affected not only the timing but also the amplitude of the 1997 to 1998 El Niño (McPhaden, 1999; Moore and Kleeman, 1999), and, in this respect, models may fail to forecast the onset of an El Niño in some circumstances. However, these results suggest an improved ability of coupled models to forecast El Niño if sufficient data to initialise the model are available from a good ocean data assimilation system.

In summary, the higher resolution coupled climate models employed since the SAR are better able to simulate El Niño-like SST variability in the tropical Pacific. However, there still remain common model errors such as weaker amplitude of SST anomalies and westward shift of the variability maximum compared to the observations. Current models can predict major El Niño events with some accuracy, suggesting that, as the resolution increases and the model physics improves, El Niño simulation will also improve.

The leading mode in the Pacific with decadal time-scale is usually called the Pacific Decadal Oscillation (PDO, see Chapter 2, Section 2.6.3). Unlike the well-documented interannual mode (ENSO), the decadal pattern does not have a distinctive equatorial maximum. Several coupled climate models are able to reproduce a pattern of this decadal variability broadly similar to the observed pattern (Latif and Barnett, 1996; Robertson, 1996; Yukimoto et al., 1996, 2000; Knutson and Manabe, 1998; Yu et al., 2000a). These modelling groups have proposed different mechanisms to explain the observed Pacific decadal variability based on analysis of large samples of simulated decadal variability in their coupled models. An example is shown in Figure 8.21b, where larger SST variability in the North Pacific than in the equatorial region is captured (Yukimoto, 1999). While the geographical location of the mid-latitude poles and the amplitude ratio between the tropical and mid-latitude poles vary slightly from one model to another, pattern correlation between the observed and model leading decadal EOFs are quite high.

IPCC Homepage