Estimating uncertainties in global and regional climate change projections using a multi-thousand member climate model ensemble

Information on the uncertainties in projections of future climate change is vital for their effective use across a wide range of applications. A multi-thousand member perturbed-physics ensemble of climate model simulations is being used to better estimate model uncertainties in climate change projections for the globe, North American region and sub-regions. Ensemble members have been generated by the distributed computing project climateprediction.net at the University of Oxford, where thousands of simulations have been run on PCs across the globe, each running a different version of the Hadley Centre HadCM3L global coupled ocean-atmosphere climate model with perturbed parameterizations.

The HadCM3L model has horizontal resolution of 3.75° longitude by 2.5° latitude with 19 vertical layers in the atmosphere and 20 vertical layers in the ocean. A 30 minute dynamical integration time step is used with physics parameterization run every 3 hours. Transient climate change simulations are available from 1921 to 2080 along with matching control simulations that use flux adjustments to maintain stable climates. For the period 1921 to 2000, the model is forced by observed changes in both anthropogenic and natural climate forcing factors, including changes in greenhouse gases and aerosols, and changes in solar irradiance and volcanic aerosols. For the period 2001 to 2080, the model is forced by a range of future solar and volcanic forcing scenarios as well as projected changes in anthropogenic greenhouse gases and aerosols according to the IPCC SRES A1B emission scenario, a mid-range scenario. Model physics parameters are varied within their current range of uncertainty for each forcing scenario, providing an ensemble of more than 7,000 climate simulations. Output variables are available in the form of globally gridded 10-year means and time series monthly means for the globe and 51 separate regions over the full time period, 1921-2080.

The simulated 20th century climate variations are compared with observed climate variations to assess their performance. Projections of climate change over the 21st century are obtained by weighting the different ensemble members by their goodness-of-fit to the 20th century observed changes, discarding those members considered to have inadequate simulations. This constrained data set is still composed of several thousand members and can be used to estimate probability distributions of mean temperature and precipitation change for a single future climate forcing scenario, which can provide insight into sources of model uncertainties in projected future climate change.

An assessment of the model control simulations will be presented along with preliminary transient simulation results for changes in global mean temperature and regional mean temperature for North America.