Geoscience Reference
In-Depth Information
Models and Analysis Methods
The primary source of GCM results used is the CMIP3 multi-model ensemble
(Meehl et al., 2008), but additionally results from a large multi-member
perturbed physics ensemble (climateprediction.net, here referred to as 'CPDN')
are also considered and compared (Sanderson et al., 2008).
Analysis of the CMIP3 model archive employs the probabilistic method,
developed by Watterson (2008) and Watterson and Whetton (2011) and used
in the last national projections for Australia (CSIRO and BoM 2007). Based on
this approach, a probability distribution is fitted to the projected local changes for
4°C global warming from across the 23 climate models. This provides the 10th,
50th and 90th percentile thresholds for each case, with 90, 50 and 10 per cent
(high, medium and low) chances of exceeding these thresholds. Watterson
(2008) applies a weighting function based current climate performance of the
models (although this is weak and has minor impact on the projected changes).
In CSIRO and BoM (2007) and Watterson (2008), the probability distributions
were scaled to match scenarios of global warming for various time slices and
emission scenarios. Here the results are simply scaled by a global warming of 4°C
(4GW). Note that Watterson (2008) also applied a log-linear scaling to improve
the realism of scaled rainfall decreases where these might otherwise exceed
2
100 per cent for cases of large global warming.
Analysis of the CPDN model ensemble includes formation of raw frequency
distributions for area-average temperature and rainfall changes for regions of
Australia for those ensemble members that reach at least a 4°C global warming
in the last decade of the simulations (around 1,000 ensemble members).