NARCliM Downscaling Methodology

This summarizes the steps to be taken in performing the downscaling for this project. The aim of the methodology is to produce a high resolution regional climate ensemble that samples the uncertainty of future climate found in the CMIP3 GCM ensemble [Meehl et al., 2007] and in the dynamical downscaling technique, as well as spanning as much of the range of future climate projections found in the CMIP3 ensemble as possible. While acknowledging that ideally all GCMs would be downscaled using multiple RCMs, limitations of computing time and data storage space require pragmatic choices to be made.

The project is limited to a twelve member GCM/RCM ensemble. This will be created by choosing four GCMs and downscaling each of these with three different RCMs. In each case three 20-year time-slices will be simulated. For future projections the SRES A2 emission scenario [IPCC, 2000] will be used. All RCM simulations will be performed at 10km resolution over NSW/ACT. This high resolution domain will be embedded within a 50km resolution domain that covers the CORDEX-AustralAsia region. Choosing this larger domain ensures that a future stage of the project focused on CMIP5 results can take advantage of simulations performed for the CORDEX initiative [Giorgi et al., 2009].

The project methodology proceeds in six main steps.

1. A series of stakeholder meetings have been held to determine the climate related variables of most interest to both the project partners and the larger stakeholder community focused on climate change impacts.
2. The three RCMs that will be used to perform the downscaling must be identified.
3. These RCMs will then be used to perform long historical simulations driven by reanalysis that will facilitate an extensive evaluation of the RCM performance including their ability to downscale the effect of inter-decadal variability.
4. The four GCMs from which to downscale must be chosen.
5. These will then be used to drive the three RCMs to simulate three time-slices that represent the present-day, the near-future and the far-future.
6. Once the ensemble has been created it will be evaluated for its "present-day" performance and analysed to produce ensemble best-estimates of the future change and uncertainty range around that change.

Each of these steps is examined in more detail below.

1. Incorporating stakeholder desired model outputs

A series of stakeholder meetings have been held to both inform them of the planned project and to illicit feedback concerning various aspects of the project in order to ensure the outcomes will be useful for them. This feedback has been incorporated into the project plan described here. In particular, a number of output variables not currently built into the RCM output procedures have been requested. Incorporating these variables requires RCM code development which must be completed before the climate simulations can be performed. In some cases, this will be the first time these variables have ever been produced in climate model output creating a truly unique dataset that will enable world leading research.

The extra variables identified through the stakeholder process include:

1. Maximum 5, 10, 20, 30 and 60 minute precipitation accumulations each day
2. Peak 10 minute wind gust each day

All variables will be output at a 3-hourly time step except for the following which will be output hourly:

1. Precipitation
2. 2-metre temperature
3. 2-metre humidity
4. 10-metre winds

2. Choosing the RCMs to perform the downscaling

The RCMs to be used will be based on the Weather Research and Forecasting (WRF) modelling system [Skamarock et al., 2008]. This system facilitates the use of many RCMs by allowing all model components to be changed and hence many structurally different RCMs can be built. The aim of this methodology is to choose three RCMs from a large ensemble of adequately performing RCMs, such that they retain as much independent information as possible while spanning the uncertainty range found in the full ensemble. Due to computational limitations, the RCM performance and independence will be evaluated based on a series of event simulations rather than using multi-year simulations.

2.1. Evaluate RCM performance for a series of important precipitation events

By limiting the evaluation period to a series of representative events for NSW, a much larger set of RCMs can be tested. In this case an ensemble of 36 RCMs will be created by using various parametrizations for the Cumulus convection scheme, the cloud microphysics scheme, the radiation schemes and the Planetary Boundary Layer (PBL) scheme. Each of these RCMs will be used to simulate a set of 7 representative storms that cover the various NSW storm types discussed in the literature [Shand et al., 2010; Speer et al., 2009]. An eighth event focused on a period of extreme fire weather will also be analysed. In each case a two week period is simulated centred around the peak of the event. Subsequent analysis then includes pre and post-event climate as well as the event itself.

Evaluation will be performed against daily precipitation, minimum and maximum temperature from the Bureau of Meteorology's (BoMs) Australian Water Availability Project [Jones et al., 2009]. Evaluation will also be performed against the mean sea level pressure and the 10m winds obtained from BoMs MesoLAPS analysis [Puri et al., 1998]. Any RCMs that perform consistently poorly will be removed from further analysis. The overall spread in these results provides a measure of the uncertainty in the RCM.

2.2. Determine RCM independence

Using the method of Abramowitz and Bishop [2010] the level of independence between the RCMs will be quantified. This method uses the correlation of model errors as an indicator of model independence. In combination, more independent models provide more robust estimates of the climate. Quantification of the model independence provides an indicator of which models contribute the most independent information and hence should be retained in the three chosen RCMs.

2.3. Choose the RCMs

The ensemble subset of adequately performing models, which is anticipated to be most of the 36 member ensemble, is identified in section 2.1 This ensemble subset is then evaluated for model independence (section 2.2) The most independent RCMs that span the subset ensemble variance will be chosen.

3. Perform and analyse historical RCM simulations

The aim of this section is to provide a comprehensive analysis of the performance of the RCMs over the recent past. Forty year historical simulations (1970 - 2010) of the chosen RCMs driven by the NCEP/NCAR reanalysis [Kalnay et al., 1996] will be performed. This will allow evaluation of the RCM performance on time scales ranging from hourly through to decadal. While previous work suggests that good performance is likely [Evans and McCabe, 2010], this evaluation will identify strengths and weaknesses of the RCMs that will be considered when analysing the future simulations. At least one of these simulations will begin in 1950 and extend for 60 years. This will capture the very wet decade of the 50s and allow for an entire Inter-decadal Pacific Oscillation (IPO) cycle to be investigated.

4.1. Evaluate GCM performance

Many evaluation studies of CMIP3 GCMs focused on south-east Australia have been performed. A comprehensive literature review will be performed to extend the meta-analysis of these studies that was done by Smith and Chandler [2010] and will provide a comprehensive evaluation that uses a suite of evaluation techniques and metrics. This and similar studies have shown that identifying the overall best performing GCMs is a difficult, if not impossible, task. The evaluation performed here will not aim to do this but rather it will aim to identify GCMs that produce consistently poor results. These GCMs will be removed from the subsequent analysis.

4.2. Determine the GCM independence

Similar to section 2.2, the method of Abramowitz and Bishop [2010] will be used to determine the level of independence of the adequately performing GCMs. In contrast to section 2.2, this will be determined by examining the model results over several recent decades rather than a series of events. The relative level of model independence will be an important factor with GCMs that demonstrate greater independence being chosen preferentially.

4.3. Examine the future changes projected by the GCMs

4.4. Choose the GCMs

Focusing only on the GCMs that perform adequately for south-east Australia and using the information from sections 4.2 and 4.3, the most independent GCMs that span the projected future change matrix will be chosen. Practical considerations such as the availability of 6 hourly data to drive the Regional Climate Models (RCMs) will also need to be considered.

5. Perform GCM/RCM simulations

Three 20 year simulations will be performed with each GCM/RCM combination, for the present-day (1990-2010) and two future periods (2020-2040 and 2060-2080). The process will be staggered with each GCM being downscaled by the three RCMs before the next GCM is downscaled. When the downscaling for each GCM is completed the data will be made available to project partners through the NARCliM data server.

While the second GCM is being downscaled, results from the first GCM will be analysed, and so forth. The analysis will start with an evaluation of the performance for the present-day using statistics of the observations. A comparison with the reanalysis driven RCMs will also be performed. This provides an opportunity to understand which errors are derived from the GCM and which from the RCM (e.g. Evans, 2010). Following this the projected future changes will be examined.

6. Ensemble best estimate and uncertainty

Given the collection of 12 GCM/RCM simulations, ensemble best estimates and the related uncertainty will be calculated to facilitate ease of use for impacts assessments. A number of factors will be considered when producing the ensemble best estimate. First the evaluation against observations will be examined to ensure no GCM/RCM simulation performs unacceptably poorly. The model independence will again be examined and a combination of the level of model independence and the likelihood of the GCM future climate change matrix category will be used to produce an ensemble best estimate.

The uncertainty in these future climate projections will be quantified in a number of ways. These techniques will range in complexity from simply investigating the spread in the future climate projections through to employing a Bayesian analysis of the changes given the prior knowledge of the simulation performance for present-day and the position in the GCM future change matrix.

6.1. Evaluate GCM/RCM performance

Using the same suite of evaluation techniques and metrics as used for the GCMs, the regional simulations will be evaluated thus providing a measure of the level of confidence in each GCM/RCM combination. If any particular ensemble member performs unacceptably poorly it will be removed from further analysis. This evaluation step will also establish a basis for bias correction of variables commonly used in impacts assessments.

6.2. Determine GCM/RCM independence

Again the method of Abramowitz and Bishop [2010] will be used to determine the level of independence between the GCM/RCMs. In this case the independence quantification can be used as a set of weights to produce an optimum independence weighted ensemble mean.

6.3. Produce regional climate change best estimates

These model independence derived weights will be combined with the likelihood of the GCM projected climate derived from the GCM future climate change matrix, to produce an ensemble best estimate that accounts for both the model independence of the GCM/RCM simulations and the range of the projected future climates from the whole CMIP3 GCM ensemble.

6.4. Produce regional uncertainty estimates

Uncertainty estimates of the future regional climate changes will be produced by examining the spread in GCM/RCM simulations, the change in probability distribution functions for various climate variables in a manner similar to Deque and Somot [2010], and a Bayesian analysis of the changes simulated given the simulation of present climate and the position in the GCM future change matrix (e.g. Buser et al., 2010; Tebaldi and Sanso, 2009).

7. References

Abramowitz, G., and C. Bishop (2010), Defining and weighting for model dependence in ensemble prediction, AGU Fall meeting, San Francisco, USA.

Buser, C., H. Kunsch, and C. Schar (2010), Bayesian multi-model projections of climate: generalization and application to ENSEMBLES results, Climate Research, 44(2-3), 227-241, doi:10.3354/cr00895.

Deque, M., and S. Somot (2010), Weighted frequency distributions express modelling uncertainties in the ENSEMBLES regional climate experiments, Climate Research, 44(2-3), 195-209, doi:10.3354/cr00866.

Evans, J. P. (2010), Global warming impact on the dominant precipitation processes in the Middle East, Theoretical and Applied Climatology, 99(3-4), 389-402.

Evans, J. P., and M. F. McCabe (2010), Regional climate simulation over Australia's Murray-Darling basin: A multitemporal assessment, J. Geophys. Res., 115(D14114), doi:10.1029/2010JD013816.

Giorgi, F., C. Jones, and G. R. Asrar (2009), Addressing climate information needs at the regional level: the CORDEX framework, WMO Bulletin, 58(3), 175-183.

IPCC (2000), IPCC Special Report on Emissions Scenarios, edited by N. Nakicenovic and R. Swart, Cambridge University Press, UK.

Jones, D. A., W. Wang, and R. Fawcett (2009), High-quality spatial climate data-sets for Australia, Australian Meteorological Magazine, 58(4), 233-248.

Kalnay, E. et al. (1996), The NCEP/NCAR 40-year reanalysis project, Bulletin of the American Meteorological Society, 77(3), 437-471.

Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney, J. F. B. Mitchell, R. J. Stouffer, and K. E. Taylor (2007), The WCRP CMIP3 multimodel dataset - A new era in climate change research, Bull. Amer. Meteor. Soc., 88(9), 1383-1394.

Puri, K., G. Dietachmayer, G. Mills, N. Davidson, R. Bowen, and L. Logan (1998), The new BMRC limited area prediction system, LAPS, Australian Meteorological Magazine, 47(3), 203-223.

Shand, T. D., I. D. Goodwin, M. A. Mole, J. T. Carley, I. R. Coghlan, M. D. Harley, and W. L. Peirson (2010), NSW Coastal Inundation Hazard Study: Coastal Storms and Extreme Waves, WRL Technical Report, UNSW Water Research Laboratory, Sydney, Australia.

Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, M. G. Duda, X.-Y. Huang, W. Wang, and J. G. Powers (2008), A Description of the Advanced Research WRF Version 3, NCAR Technical Note, NCAR, Boulder, CO, USA.

Smith, I., and E. Chandler (2010), Refining rainfall projections for the Murray Darling Basin of south-east Australia-the effect of sampling model results based on performance, Climatic Change, 102(3-4), 377-393, doi:10.1007/s10584-009-9757-1.

Speer, M., P. Wiles, and A. Pepler (2009), Low pressure systems off the New South Wales coast and associated hazardous weather: establishment of a database, Australian Meteorological and Oceanographic Journal, 58(1), 29-39.

Tebaldi, C., and B. Sanso (2009), Joint projections of temperature and precipitation change from multiple climate models: a hierarchical Bayesian approach, Journal of the Royal Statistical Society A-Statistics in Society, 172, 83-106, doi:10.1111/j.1467-985X.2008.00545.x.

Whetton, P., and K. Hennessy (2010), Potential benefits of a "storyline" approach to the provision of regional climate projection information, International Climate Change Adaptation Conference, NCARF, Gold Coast, Australia.