NARCliM model selection

Projections on the AdaptNSW website use the NSW and ACT Regional Climate Model (NARCliM) data released in 2014 and comprise a twelve model ensemble. Three configurations of the Weather Research and Forecasting (WRF) regional climate model (RCM) have been used to downscale projections from four global climate models (GCMs) providing a total of 12 models. This ensemble approach is used to ensure robust climate projections that are at the regional scale and span the likely future climate.

NARCliM uses dynamical downscaling to generate fine scale climate projections but there are other techniques to downscale GCMs.  

Why was dynamical downscaling used?

Global climate models typically have grid cells that are hundreds of kilometres wide. This reduces their usefulness in projecting future climate conditions at regional scales, as they are not able to accurately represent the features of the Earth’s surface that influence local climates, such as coastlines and mountain ranges.

To overcome these limitations GCMs are ‘downscaled’. There are two main techniques for downscaling GCMs:

  • Statistical downscaling uses the existing relationships between local observations and large-scale climate features, and it applies these relationships to projections from GCMs. The benefit of this method is that it requires relatively small computing resources. However, the method assumes that the relationships between large-scale and local climates will not change in the future.
  • Dynamical downscaling uses high-resolution regional climate models (RCMs). These are built by using the same principles as GCMs but contain additional information about a range of factors (such as topography) that are important in determining regional climate. Because they are built on physical principles, dynamical RCMs allow for changes in the existing relationship between weather variables and climate drivers. Dynamical models create a new time series, rather than adjusting an existing one.

Many climate processes take place on much smaller scales than a GCM can resolve. This can be overcome by using regional climate models to downscale global climate models, so that the influence of these processes on the local climate is captured. Additionally, unlike statistical downscaling, dynamical downscaling is able to project climate variables that have poor, or no, observational data and can be run at finer time scales.

Why were CMIP3 GCMs used?

The Coupled Model Intercomparison Project (CMIP) was established under the World Climate Research Program to support climate model analysis, validation, intercomparison, documentation and access to data. The project helps climate scientists to systematically evaluate and improve climate models.

Phase three of CMIP (CMIP3) used the IPCC Special Report on Emission Scenarios (SRES) and informed the IPCC Fourth Assessment Report (AR4, published in 2007). Climate projections in the IPCC Fifth Assessment Report (AR5, published in 2014) are based mainly on the fifth phase of CMIP. CMIP5 incorporates the latest versions of climate models and uses updated emissions scenarios or Representative Concentration Pathways.

Work on NARCliM started in 2010, before the introduction of RCPs and the completion of CMIP5; therefore, the 2014 NARCliM projections relied upon the CMIP3 models from the IPCC Fourth Assessment Report.

The CMIP3 models have been widely used to investigate global climate system processes and climate change projections. A number of CMIP3 evaluation publications are available, and several studies have evaluated the performance of CMIP3 GCMs over south-eastern Australia. This made it possible for NARCliM to select the GCMs that performed best over NSW.1

Why use an ensemble of simulations?

Models can’t perfectly represent the Earth’s climate. For example, models may have some regions that are slightly hotter or drier than the actual climate. These differences are known as error, or 'bias'. To reduce the bias and uncertainty from a single model, a collection or ensemble of models is used to cover a range of plausible future climates.

The 12 models used in the 2014 NARCliM projections were chosen to provide high-resolution climate projections that span the likely range of future climate uncertainty across all CMIP3 models.

GCM selection

Regional climate modelling requires substantial computing and data storage facilities, restricting the number of GCMs that could be used for NARCliM. This meant that a systematic approach was required to ensure the best choices of GCMs for NARCliM at the time.

GCMs were chosen such that the full range of uncertainty within the CMIP3 GCMs was captured by models that simulated the NSW climate well. Even if a GCM can simulate the current climate, it may not necessarily be able to simulate the future climate well. Choosing GCMs that span the range of projected future climate reduces the uncertainty of the projections.2

The GCMs selected for the 2014 NARCliM projections were selected by using a three-stage process:1

  1. The performance of a total of 23 CMIP3 GCMs was evaluated, and models that did not adequately simulate the historical climate were removed.
  2. The set of GCMs that performed well was then ranked on the basis of a measure of their independence (i.e. whether they exhibited the same strengths and weaknesses in their representation of the climate).
  3. The GCMs were then evaluated on the basis of their projections of future climate change. The most independent models that spanned the largest range of plausible future climates were chosen.

The selection process referred to the published scientific literature to identify GCMs that performed poorly over south-eastern Australia. The performance of each of the 23 CMIP3 models was ranked on the basis of the findings of 12 independent, peer reviewed studies. Further information on individual model performance and study references is provided in Tech Note 1, Table 1: Summary of model assessments. The six worst-performing models were removed from further analysis.

The remaining GCMs were then ranked on the basis of independence,3 focusing on daily mean temperature and precipitation, as these variables are of the most interest to users of climate change information. Further information on individual model performance is provided in Tech Note 2Table 2: The absolute GCM independence coefficient for each model.

Each of the models’ projected changes in future temperature and rainfall were then mapped and the four most independent models spanning the extent of future climate change were chosen. Further information is provided in Tech Note 1, Figure 1: Future change space for the GCMs.

The GCMs selected for were MIROC3.2 (medres), ECHAM5, CCCM3.1 and CSIRO-Mk3.0.

RCM selection

To further ensure that the 2014 NARCliM projections spanned the range of uncertainty in regional climate models, a process similar to that used for GCM selection was used to choose the best regional climate models:4, 2

  1. The performance of RCMs was evaluated over south-east Australia and models that did not adequately simulate the climate were removed.
  2. From the set of RCMs that performed well over south-east Australia, a subset was chosen so that each RCM was as independent as possible from the others.

The WRF modelling system has been demonstrated to be effective in simulating temperature and rainfall in NSW5 and provides good representations of local topography and coastal processes. This model was jointly developed by several major research centres in the United States of America and is widely used internationally.6, 7

WRF has different settings that reflect uncertainties in our understanding of some physical processes. Varying these settings among model runs and using several different settings with each GCM provides a more realistic set of projections. Performing multiple model runs also captures more reliable information on rare, extreme weather events, such as heatwaves, heavy rain and drought.

For evaluation, the WRF model was run with 36 different physics scheme settings using combinations of the cumulus convection scheme, the cloud microphysics scheme, radiation schemes and the Planetary Boundary Layer scheme.

Because of computational limitations, not all RCM combinations could be included in the full NARCliM projections. The performance and independence of each RCM were evaluated and used to simulate a range of storm types that affect the state, such as East Coast Lows, with performance testing for precipitation, minimum and maximum temperatures, mean sea level pressure and wind.2, 8, 9

The RCM ensemble was evaluated over south-east Australia in order to remove any models that were not able to adequately simulate the climate. From the set of RCMs that performed well, a subset was chosen, so that each was as independent as possible from the others. On the basis of model performance and independence, three WRF model configurations were selected for use in the downscaling4 (see table).

Regional climate model configuration


Planetary boundary layer physics / surface layer physics

Cumulus physics


Short-wave / long-wave radiation physics


MYJ/ Eta similarity


WDM 5 class



MYJ/ Eta similarity


WDM 5 class



YSU/MM5 similarity


WDM 5 class


Further information on RCM selection is provided in Tech Note 2 


  1. Evans JP and Ji F (2012). Choosing GCMs, NARCliM Technical Note 1, 7 pp., NARCliM Consortium, Sydney, Australia.
  2. Evans JP, Ji F, Lee C, Smith P, Argüeso D, and Fita L (2014.) Design of a regional climate modelling projection ensemble experiment (PDF 1.43 MB) – NARCliM, Geosci. Model Dev., 7, 621–629.
  3. Bishop CH and Abramowitz G (2013). Climate model dependence and the replicate earth paradigm. Clim. Dynam., 41, 885–900, doi:10.1007/s00382-012-1610-y.
  4. Evans JP and Ji F (2012). Choosing the RCMs to perform the downscaling (PDF 298 KB). NARCliM Technical Note 2, 8 pp., NARCliM Consortium, Sydney, Australia.
  5. Evans JP and McCabe MF (2010). Regional climate simulation over Australia's Murray-Darling basin: A multi-temporal assessment  (PDF 6.02 MB). J. Geophys. Res., 115(D14114), doi:10.1029/2010JD013816.
  6. Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Duda MG, Huang XY, Wang W, and Powers JG (2008). A Description of the Advanced Research WRF Version 3, NCAR Technical Note, National Center for Atmospheric Research, Boulder, CO, USA.
  7. National Center for Atmospheric Research(NCAR) 2004, National Center for Atmospheric Research, USA, viewed 7 August 2014.
  8. Evans JP, Ji F, Abramowitz G, and Ekstrom M (2013). Optimally choosing small ensemble members to produce robust climate simulations. Env. Res. Letters, 8(4), doi:10.1088/1748-9326/8/4/044050
  9. Ji F, Ekström M, Evans JP, and Teng J (2014). Evaluating rainfall patterns using physics scheme ensembles from a regional atmospheric model. Theor. Appl. Clim., 115, 297–304, doi:10.1007/s00704-013-0904-2