FAQ

Users of the climate change data provided by the Goyder Institute are likely to need different elements of the data provided, depending on their application.

Two types of climate change information are provided:

A. Mean amounts of projected change in the variables of rainfall, temperature and PET for each NRM region

These provide the projected amounts of change in rainfall, temperature and PET for each region according to an average of the projections of the selected GCMs, averaged across all the weather stations in the region that have been used for downscaling. These regional summaries indicate the average amount of change (in each weather variable) that is incorporated in the downscaled climate datasets for the weather stations in a region. These data will be essential for regional climate change adaptation planning in South Australia and for NRM planning that incorporates consideration of climate change at the regional scale.

B. Multiple time series datasets of projected values of weather variables, downscaled to individual weather stations in daily time steps, provided as multiple text files

These datasets are intended for use in numerical modelling and statistical analyses where variations in climate variables on a daily (or greater) timescale are influential on the model outcome. The datasets are useful for climate change impact modelling studies, commonly in hydrological disciplines, but also in agriculture and infrastructure planning and quantitative risk assessment for many sectors. Typically, users of these datasets will be skilled in numerical modelling, data processing and/or statistical analysis.

Users should note that detailed climate projection datasets such as these are not forecasts of an exact sequence of events in the future. Rather, they are a set of synthetic climate data that are statistically realistic but include projected change in the climate. These datasets can be used to explore what future conditions may look like in a climate sense, and explore the impacts of that projected change.

For regional climate change adaptation planning purposes, the regionally summarised climate change projections provided in the regional summary brochures (www.goyderinstitute.org) and in the Goyder Institute project’s final downscaled projections report (Charles and Fu 2014) provide the most appropriate information source and are recommended for use wherever possible for climate change adaptation planning in South Australia. These reports provide summaries of the change projected by the suite of selected GCMs, for each of the projected climate variables, averaged across the weather stations within each South Australian NRM region.

For location-specific modelling of climate impacts, such as for prediction of impacts on runoff in a particular reservoir catchment, or predictions of changes in heatwave frequency and characteristics for a particular city, the downscaled future climate time series datasets are provided for 200 weather stations in South Australia.

The RCP4.5 and RCP8.5 scenarios are representative of increases in atmospheric greenhouse gas concentrations through the 21st century that result in global radiative forcings in 2100 of 4.5 W/m2 and 8.5 W/m2 respectively corresponding to CO2 levels of approximately550 ppm and 940 ppm by 2100. The downscaled datasets for these scenarios represent individual GCM projections of climate at each weather station under the change in global climate that is projected by the GCM with the respective concentration pathway (and corresponding radiative forcing) applied.

The downscaled datasets for the historic (filename suffix ‘his’) scenario represent individual GCM recreations of climate at each weather station from 1961 to 2005 under the GCM’s historical climate of that period (i.e. they do not replicate the sequences of the observed historical record).

Because the downscaled projections of future climate are affected by both the GCM and the NHMM downscaling process, any study that compares results of modelling with synthetic climate data with the results of modelling of current or historic climate conditions must use historic climate data that is derived from the same GCM and NHMM downscaling process. It would not be an appropriate comparison if results of future climate scenario modelling were compared with results from models using actual historic data from the same location.

A combination of 15 GCMs, two concentration pathways and 100 realisation of each results in 3000 future climate time series for each weather station. Many users may find this an unwieldy number of climate datasets to process through a model or on which to undertake statistical analysis and may wish to select a subset of either GCMs, RCPs or realisations to work with.

Whether to use one or two RCPs?

The RCP4.5 and RCP8.5 scenarios represent, respectively, reasonable lower and upper bounds of greenhouse gas concentration pathways for the 21st century. In view of the factors that introduce a range of uncertainty into projections of future climate, it is common in climate impact studies to select two scenarios which may be considered ‘bookends’ to the range of possible outcomes. The RCP4.5 and RCP8.5 scenarios represent reasonable bookends to the likely pathways of atmospheric greenhouse gas concentrations. Hence an approach that considers the future impacts of both of these scenarios may be adopted in many cases. By adopting this approach, modelling or statistical analysis results can demonstrate the range of future climate impact outcomes that are possible while future greenhouse gas concentrations remain uncertain.

Some users may choose to reduce the number of scenarios to be dealt with by only considering only one RCP. This is a valid approach, providing the RCP adopted and the uncertainty in whether it will be realised in future are clearly articulated. RCP4.5 and RCP8.5 do not start to diverge significantly until after 2020. For impact studies that project no further into the future than 2030, use of only the RCP8.5 scenario may be considered a reasonable and precautionary approach.

Which GCMs to select?
The Goyder Institute climate change project selected GCMs from the Coupled Model Intercomparison Project Phase 5 (CMIP5) (Taylor et al., 2012) These CMIP5 GCMs were used in the most recent IPCC AR5 report (IPCC, 2013). Fifteen GCMs were selected on the basis that they had all the variables available to produce the atmospheric predictors required for input to the calibrated NHMM downscaling process. Their names and host institutions are listed in Table 1.

The fifteen selected GCMs provide a range of projections of future climate. There are risks inherent in selecting individual GCMs from this group for climate change impact studies, since it is not possible to identify which is likely to provide a more accurate projection of the future climate.

To enable comparison of projections from GCMs assessed as better performing, in terms of their ability to simulate the large-scale processes of relevance to South Australian climate drivers, the Goyder Institute project assessed a set of CMIP5 GCMs that included 12 of the 15 GCMs available for downscaling. Based on this assessment it was possible to classify the 12 into a subset of six better and six poorer performing GCMs, based on their ability to reproduce drivers of relevance to South Australian climate. These subsets are identified in Table 1. Details of the GCM assessment are summarised in Cai et. al. (2014a,b).

Users of the downscaled datasets for modelling and statistical studies are advised to apply the projections of a number of GCMs to ensure that the range and uncertainty among the projects of differing GCMs is represented and conveyed in the resulting products. The six better performing GCMs identified by the Goyder Institute project may for many users be a suitably small number of GCMs to convey a justifiable range of climate change projections for South Australia.

Users needing to sub-select GCMs further may also take a bookending approach to GCM selection. In this, the two or more GCMs that provide the least and most amount of temperature and/or rainfall change for the location of interest would be selected. This process should be undertaken with care for a number of reasons:

• The GCM that projects the most or least amount of change for one location may not do so for another location. Similarly for differing time periods, a GCM that projects the most or least amount of change for one time period (such as 2030-2050) may not do so for another time period (such as 2070-2070).
• A GCM may project an amount of change that is significantly different to the projections of the majority of the group of 15 GCMs. This will not be apparent unless the changes projected by the whole group for the location of interest are assessed to determine the spread of their differing projections.
• A GCM that projects the most or least amount of change in temperature may not be the same as the GCM that projects the most or least amount of change in rainfall.
• The GCMs that project the highest/lowest values for one variable (whether it be annual average rainfall, annual maximum rainfall and so on) will not necessarily lead to the highest/lowest values in the outcomes of a model they are applied to. For example, the downscaled data from a GCM that projects the greatest reduction in rainfall may not result in the greatest reduction in runoff when the projections of a number of GCMs are applied to a hydrological catchment runoff model.

The CSIRO ‘Climate Futures’ analysis method¹ provides a means that allows an analysis of the spread of change in both rainfall and temperature projected for a location by a group of GCMs. This method entails plotting the mean change projected by the individual GCMs for a given location and time horizon, with the temperature and rainfall variables plotted on axes of a single scatter graph. The resulting graph illustrates the relative spread of the projected changes in rainfall and temperature among the group of GCMs. From this graph, the GCMs that project, for example, the warmest-driest and coolest-wettest future climates can be selected for use in modelling and statistical applications. The cautions listed above should be carefully considered if this approach to GCM selection is taken.

As noted above, the downscaled daily climate projection datasets such are not forecasts of an exact sequence of events in the future. Rather, they are a set of synthetic climate data that are statistically realistic but include projected change in the climate. Within the downscaled climate projections datasets, for each combination of GCM and emissions scenario there are 100 realisations of rainfall from 2006-2100.

This multiple dataset approach is necessary when statistically downscaled daily time series of climate variables are provided. This is because each projection is only one possible realisation of the future outcome of temperature, rainfall, etc. under the climate conditions that are projected by the GCM and downscaled to represent the weather patterns of the individual location; they are not a prediction of what will happen. Repeating the statistical downscaling process 100 times for each GCM/emissions scenario combination provides a group of projections that represent the range of possible daily weather outcomes within the projected future climate.

Each of these groups of 100 realisations is based on the same GCM projection of climate, hence they each have similar decadal-scale trends. However, within each GCM/emissions scenario combination, there is a broad spread of future climate projections. Within each group of 100, users might take an average of the outcomes of the modelling or statistical process they apply the group of 100 datasets to, or might select, for example, the 10th wettest, 10th driest and median, climate realisations the of the impact projections resulting from these.

The following technical FAQs address addresses questions related to the similarities and differences between the climate projections data generated for the SA Climate Ready project and the CSIRO and Bureau of Meteorology project “Climate Change in Australia: Projections for Australia’s NRM regions” (CCIA).

Q5a. What kinds of climate change projection data are these projects providing?

The SA Climate Ready project provides:

1. Projections of average changes in maximum and minimum temperature, rainfall and potential evapotranspiration at the scale of individual SA NRM regions;
   
2. Downscaled projected climate data for individual weather stations in daily time steps for the climate variables of rainfall, maximum temperature, minimum temperature, vapour pressure deficit, areal evapotranspiration and solar radiation. These variables were chosen specifically to assist with anticipated hydrological and environmental modelling applications of the resulting data in South Australia.

The CCIA project provides projected climate change data for the following atmospheric and ocean variables at various spatial scales: mean temperature, maximum daily temperature, minimum daily temperature, rainfall, relative humidity, point potential evapotranspiration, wet areal evapotranspiration, wind-speed, solar radiation, wind speed and fire weather days, mean and extreme sea-level rise, sea surface temperature, sea surface salinity and ocean acidification. Qualitative projected change information is also available for tropical cyclones, snow, run-off and soil moisture.

The CCIA project does not provide data for mean projected climate change for individual SA NRM regions. The CCIA project will produce projected climate data for a subset of individual weather stations within South Australia, but these will be scaled from historic data by a method that does not represent the potential temporal variation in climate variables on a daily time scale.

Users of data from both SA climate Ready and CCIA should note that detailed climate projection datasets are not forecasts of an exact sequence of weather in the future

Q5b. What global climate models were used?

Both projects considered up to 40 available global climate models available from the Coupled Model Intercomparison Project Phase 5 (CM​IP5) of the World Climate Research Program.

The SA Climate Ready project selected 15 GCMs to generate climate change projections. A list of these GCMs is provided in the User Guide. The project also examined the influence on projections of using only the best six or worst six of the 15 GCMs selected according to their ability to represent the influence of the ENSO and the Indian Ocean Dipole, which are large scale climatic drivers that affect the climate, and particularly rainfall, in south eastern Australia.

By comparison, the CCIA project provides summary information and projected change data using up to 40 CMIP5 models. The number of models used varied depending on the climate variable and greenhouse gas concentration scenario at the resolution of the particular host GCM (~67 km to ~333 km). ‘Application-ready’ future climate data is provided to intermediate users (having undergone on-line training) for 8 GCMs, selected to represent the range of projections made by the full set of 40 models. For advanced users, application ready data from all GCM’s and also data produced using two different downscaling methodologies (dynamical and statistical) are available.

Q5c. What concentration pathways were used?

Global climate models generate climate projections based on potential future concentrations of atmosphere greenhouse gases and aerosols. Scenarios of future concentrations are referred to as Representative Concentration Pathways (RCPs).

The SA Climate Ready climate projections data were generated using the intermediate concentration pathway RCP4.5 and high concentration pathway RCP8.5. In contrast, the CCIA project uses all four available concentration pathways: RCP2.6, RCP4.5, RCP6 (only for sea level projections) and RCP8.5.

Q5d. What was the approach to downscaling?

Downscaling is a technique used to convert the results of large scale gridded data (~67 km to ~333 km) from global climate models to generate data at a more local scale. This has two main goals: to make the data locally relevant, and to reveal any regional variations in the projections of change. Scaling of observations by projected changes from GCMs is an alternative method to produce locally relevant data, but this cannot reveal any new regional variations in the projected change.

The SA Climate Ready project’s downscaled climate projections data were generated using Nonhomogeneous Hidden Markov Model (NHMM) statistical downscaling for daily multi-site rainfall. A statistical weather generator model, conditional on the rainfall, was used to generate the non-rainfall variables.

The CCIA climate projections are also produced using GCMs as the primary data source, with data scaling and both dynamical and statistical downscaling used to provide complementary forms of information. GCM outputs are used to describe projected change and produce scaled datasets. The regional pattern of change shown by downscaling were also examined and described in outputs from a dynamical downscaling technique but these outputs are not provided as datasets for use in applications.

Q5e. For what regions or locations is data available?

The SA Climate Ready provides two types of climate information and includes coverage for all of SA’s eight NRM regions:

• Mean amounts of projected change in the variables of rainfall, temperature and PET for each NRM region; and
• Multiple time series datasets of projected values of weather variables, downscaled to 200 individual weather stations in daily time steps, provided as multiple text files.

The data align with and provide broad coverage of all the South Australian NRM regions, with the exception of the Alinytjara Wilurara NRM Region, for which data is only available for a single weather station. The mean changes projected for that region are therefore discussed in conjunction with the South Australian Arid Lands NRM Region.

The CCIA project examines and describes projected climate change in eight ‘clusters’ of natural resource management (NRM) regions, with some of the clusters being subdivided resulting in 15 ‘sub-clusters’ that collectively cover the Australian continent (See CCIA website link). Within South Australia, there are three relevant sub-clusters:

• Rangelands (South), which covers the:
  • Alinytjara Wilurara NRM Region;
  • South Australian Arid Lands NRM Region;
• Southern and South-Western Flatlands (East), which covers the:
  • Eyre Peninsula NRM Region;
  • Northern And Yorke NRM Region;
  • Adelaide and Mount Lofty Ranges NRM Region;
  • Kangaroo Island NRM Region;
• Murray Basin, which covers the:
  • South Australian Murray-Darling Basin NRM Region; and
  • South East NRM Region.

Q5f. What type of local-scale data is available?

The SA Climate Ready climate projections data is available for 200 individual weather stations across South Australia. Two types of climate information are provided:

• Mean amounts of projected change in the variables of rainfall, temperature and PET for each of seven SA NRM regions (Alinytjara Wilurara NRM Region is not represented in the mean projected change summaries); and
• Multiple time series datasets of projected values of weather variables, downscaled to 200 individual weather stations in daily time steps, with coverage of all eight SA NRM regions (note however, only one station is provided for AW NRM region). These are provided as multiple text data files.

The CCIA ‘application-ready’ future climate data has been generated for selected high quality weather stations across Australia (this varies depending on the climate variable of interest as is defined by the Bureau of Meteorology) and also available on a 5 km grid. To produce the future climate data the courser resolution change grid (host GCM ~67 km to ~333 km) is applied to the Bureau of Meteorology’s AWAP baseline gridded climate information. The main difference between CCIA and SA Climate Ready datasets is that CCIA provides station or gridded data scaled by changes projected by GCMs, whereas SA Climate Ready provides projected climate data for individual stations, generated by a statistical model that applies the changes projected by GCMs to the statistical variation of historic measurements of climate variables at each weather station.​