Model ensembles to understand climate variability and change

By November 22nd, 2021

A comparison of the projections from just one simulation (r1) from a group of CMIP6 models with many simulations from the ACCESS- ESM1.5 model. This shows outputs for the Southern Australia (SA) ‘super-cluster‘ region used in (as marked on the map). This analysis illustrates how natural variability and forced climate change can be disentangled using a large ensemble in combination with a multi-model ensemble. The plot shows a worst case, high emissions pathway of SSP5-8.5 (and RCP8.5 from CMIP5) to illustrate a strong climate signal. It is not presented as a ‘most likely’ future scenario.

Top panels show the annual temperature and rainfall time series relative to the baseline period of 1950-1999. Bottom left shows a map of the projected change in annual rainfall in the mean of the 40 ACCESS-ESM1.5 members, and bottom right shows a scatter plot of the annual temperature and rainfall change between 1950-1999 and 2050-2099.

Important points on the plot include:

  • ACCESS-ESM1.5 provides a credible ‘dry case’ projection for Australia, and can be used to illustrate this extreme dry storyline for Australia (the model evaluates well in the past and present climate, so the projection is considered plausible).
  • ACCESS-ESM1.5 has mid-range climate sensitivity and warming, in contrast to some models in CMIP6 that have high climate sensitivity and very high warming (well outside the CMIP5 range, as seen on the scatter plot).
  • There is a spread of projected change in the Large Ensemble, illustrating the important effect of natural variability on what we experience even between 50-year periods.
  • The spread in the Large Ensemble is not as large as the spread from the group of CMIP6 models – this is because structurally different models give a different change ‘signal’, for both warming and also rainfall change.
  • If we had a large ensemble for each model, we could analyses the scatter due to natural variability compared to a different climate change signal in each model in a quantitative way. We are moving in this direction, with CMIP6 containing more large ensembles than ever before.

ACCESS realisation 6 is marked in red on the scatter plot – this is the simulation that provides sub-daily outputs that can be used in high-resolution downscaling models such as in CORDEX. The projection from this simulation is in the mid-range of the 40 for this case (note, it is a particularly dry projection over the first half of the 21st century, especially for eastern Australia, making it a useful ‘stress test’ case here).