Oral Presentation AMOS Annual Meeting and International Conference on Tropical Meteorology and Oceanography

Simulated rainfall variability over Australia: evaluation of the latest ACCESS-CM2 AMIP/C20C+ model results (#219)

Roger W Bodman 1 , David J Karoly 2
  1. The School of Earth Sciences, The University of Melbourne, Melbourne, Victoria, Australia
  2. CSIRO, Aspendale, Victoria, Australia

Key features of the Australian climate have been evaluated based on AMIP/C20C+ simulations produced using the ACCESS-CM2 (Australian Community Climate and Earth System Simulator – Coupled Model version 2) climate model, including surface temperatures, rainfall and atmospheric circulation. Historical forcings for the period 1951 to 2014 were utilised along with configurations for two different land surface models, the UK MetOffice JULES model and the Australian CABLE model, with multiple realisations for each. 

The ACCESS-CM2 AMIP results provide an opportunity to evaluate the performance of the atmospheric model in the context of Australia’s late twentieth-/early twenty-first century observed climate. Model results have been compared to observations and reanalysis data as well as an earlier version of the model used for the CMIP5 AMIP simulations.

The new simulations represent the general pattern of surface air temperature quite well, although there are issues at higher latitudes and with the amount of variability. The mean rate of precipitation is over estimated, with particular problems around the equatorial region. Overall, the new ACCESS-CM2 AMIP results show improvements over ACCESS 1.3 AMIP as measured by global RMSE values.

For the Australian continent, the simulated surface temperature results are similar to observations, although the model performance varies between minimum, mean and maximum annual temperatures as well as for seasonal temperatures. Simulated annual rainfall shows a weaker correlation to observations than for temperature. The ENSO pattern correlation for rainfall reveals broad agreement with observations, but the correlations are typically weaker and the patterns less distinct, and the mean value based on multiple realisations performs better than individual realisations.