11.7.2 Skill of Models in Simulating Present Climate

There are relatively few studies of the quality of the MMD global model simulations in the Australia-New Zealand area. The ensemble mean of the MMD model simulations has a systematic low-pressure bias near 50°S at all longitudes in the SH, including the Australia-New Zealand sector, corresponding to an equatorward displacement of the mid-latitude westerlies (see Chapter 8). On average, mid-latitude storm track eddies are displaced equatorward (Yin, 2005) and deep winter troughs over southwest Western Australia are over-represented (Hope 2006a,b). How this bias might affect climate change simulations is unclear. It can be hypothesised that by spreading the effects of mid-latitude depressions too far inland, the consequences of a poleward displacement of the westerlies and the storm track might be exaggerated, but the studies needed to test this hypothesis are not yet available.

The simulated surface temperatures in the surrounding oceans are typically warmer than observed, but at most by 1°C in the composite. Despite this slight warm bias, the ensemble mean temperatures are biased cold over land, especially in winter in the southeast and southwest of the Australian continent, where the cold bias is larger than 2°C. At large scales, the precipitation also has some systematic biases (see Supplementary Material Table S11.1). Averaged across northern Australia, the median model error is 20% more precipitation than observed, but the range of biases in individual models is large (–71 to +131%). This is discouraging with regard to confidence in many of the individual models. Consistent with this, Moise et al. (2005) identify simulation of Australian monsoon rainfall as a major deficiency of many of the AOGCM simulations included in Phase 2 of the Coupled Model Intercomparison Project (CMIP2). The median annual bias in the southern Australian region is –6%, and the range of biases –59 to +36%. In most models, the northwest is too wet and the northeast and east coast too dry, and the central arid zone is insufficiently arid.

The Australasian simulations in the AOGCMs utilised in the TAR have recently been scrutinised more closely, in part as a component of a series of national and state-based climate change projection studies (e.g., Whetton et al., 2001; Cai et al., 2003b; McInnes et al., 2003; Hennessy et al., 2004a,b; McInnes et al., 2004). Some high-resolution regional simulations were also considered in this process. The general conclusion is that large-scale features of Australian climate were quite well simulated. In winter, temperature patterns were more poorly simulated in the south where topographic variations have a stronger influence, although this was alleviated in the higher-resolution simulations. A set of the TAR AOGCM simulations was also assessed for the New Zealand region by Mullan et al. (2001a) with similar conclusions. The models were able to represent ENSO-related variability in the Pacific and the temperature and rainfall teleconnection patterns at the Pacific-wide scale, but there was considerable variation in model performance at finer scales (such as over the New Zealand region).

Decadal-scale variability patterns in the Australian region as simulated by the CSIRO AOGCM were considered by Walland et al. (2000) and found ‘broadly consistent’ with the observational studies of Power et al. (1998). At smaller scales, Suppiah et al. (2004) directly assessed rainfall-producing processes by comparing the simulated correlation between rainfall anomalies and pressure anomalies in Victoria against observations. They find that this link was simulated well by most models in winter and autumn, but less well in spring and summer. As a result of this, they warn that the spring and summer projected rainfall changes should be viewed as less reliable.

Pitman and McAvaney (2004) examine the sensitivity of GCM simulations of Australian climate to methods of representation of the surface energy balance. They find that the quality of the simulation of variability is strongly affected by the land surface model, but that simulation of climate means, and the changes in those means in global warming simulations, is less sensitive to the scheme employed.

Statistical downscaling methods have been employed in the Australian region and have demonstrated good performance at representing means, variability and extremes of station temperature and rainfall (Timbal and McAvaney, 2001; Charles et al., 2004; Timbal, 2004) based on broad-scale observational or climate model predictor fields. The method of Charles et al. (2004) is able to represent spatial coherence at the daily time scale in station rainfall, thus enhancing its relevance to hydrological applications.