Journal of Southern Hemisphere Earth Systems Science (Jan 2020)

Evaluation of CMIP6 AMIP climate simulations with the ACCESS-AM2 model

  • C. Mackallah,
  • P. B. Dobrohotoff,
  • J. Srbinovsky,
  • I. N. Harman,
  • M. R. Dix,
  • D. J. Karoly,
  • R. W. Bodman

Journal volume & issue
Vol. 70, no. 1
pp. 166 – 179

Abstract

Read online

The most recent version of the ACCESS-AM2 atmosphere-only climate model is introduced with results from the CMIP6 Atmospheric Model Intercomparison Project (AMIP) experiments configured with two land-surface models: CABLE and JULES. AMIP simulations are required as part of the CMIP6 core experiments. They are forced by prescribed time-varying observed sea surface temperature and sea-ice variations as well as variations in natural and anthropogenic external forcings. We evaluate the performance of the two configurations using three historical realisations for each. Model biases are estimated both globally and for the Australian region. The model shows close agreement with observed interannual variations of global-mean temperature across the latitude range 65°N–65°S. This is also true for the land-only temperature for 65°N–65°S, and a more stringent test of the model is driven by specified observed sea surface temperatures. Patterns of mean precipitation are simulated reasonably well, although there are biases in the amount and distribution of precipitation, typical of longstanding problems in representing this aspect of the climate. Selected features of the atmospheric circulation are discussed, including air temperatures and wind speeds. For the Australian region, in addition to examining the climatological patterns of temperature and precipitation, important drivers of climate variability are reviewed: El Niño-Southern Oscillation, the Indian Ocean Dipole and the Southern Annular Mode. In general, the correlation patterns for precipitation simulated by ACCESS-AM2 are somewhat weaker than in observations, although the ensemble means show better agreement than individual ensemble members. Overall, the two different land-surface schemes perform similarly. ACCESS-AM2 has reduced root mean square errors for both temperature and precipitation of around 15–20% at the global scale compared to the older CMIP5 versions of the model: ACCESS 1.0 and ACCESS 1.3.

Keywords