Model evaluation determines how well climate models represent historical climate and forms an integral part of the confidence building exercise for climate change projections (Section 6.4). The assumption is that the better models perform over the historical period the more confidence can be assigned to their projected changes.
Global climate models are able to reproduce the major climatic features and modes of variability affecting Australia (Section 5.2), such as major pressure systems, seasonal and annual cycles of rainfall and temperature, the monsoon, El Niño - Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD).
The ability of individual CMIP5 models to simulate Australian climate varies depending on the climatic variable, region and season under consideration (for details seeSection 5.2). No small subset of models can be identified that outperforms the remaining models consistently. However, there is a small subset of models that perform poorly across a number of metrics used to assess their performance.
Geographically finer details around pronounced topography and coastlines such as in Tasmania or the Cape York Peninsula are more difficult to simulate. Several models show biases in these regions (seeSection 5.2).
Broad-scale temperature, rainfall and surface wind climatologies are well captured by global climate models on seasonal and annual time scales. A few models were identified as being somewhat deficient across several assessments (Table 5.6.1 inChapter 5) and are therefore highlighted as such in the and in Box 9.1 in the .
Most of the CMIP5 models exhibit biases in the central Pacific sea surface temperatures, which impact on the El Niño Southern Oscillation (ENSO) relationship with Australian rainfall. Some of the models also show biases in the onset of the Australia monsoon (Section 5.2.4).
Selecting a subset of better performing models did not provide more clarity on the direction of the expected rainfall changes. Using the full range of available models, rather than a performance-based selection or weighting, is the chosen method for regional applications (Chapter 6 and Box 6.2.1). Model evaluation results, however, may influence the choice of individual models in some applications.
Page updated: 28th September 2016