Choosing from multiple sources of projections

The state of the climate in a future period depends mainly on three factors: the pathway of human development (e.g. described by Representative Concentration Pathways), the response of the climate system to that pathway and natural variability.

So, along with considering a range of RCPs and allowing for natural variability, it is necessary to consider the range of possible climate response. The range of results from various Global Climate Model (GCM) simulations gives some indication of this response, where all results must be considered plausible. However, GCMs produce output at a coarse spatial scale and users often desire information at finer spatial scales. The production of finer spatial scale data has two main goals:

  1. Goal 1: To produce datasets that look like the observations we are familiar with (such as gridded data like AWAP ) that can be used to study regional impacts in detail.
  1. Goal 2: To reveal any finer-scale details in the pattern of climate change (both spatial patterns and change to variability in time).

These will not be relevant to every study or impact question, so it is important to examine the context before choosing data inputs. Before producing (or obtaining) fine-resolution datasets, it is important to consider whether they are really needed. Also, different variables have a different potential for regional differences in climate change. For example, the change in average temperature is fairly uniform over broad regions whereas the change in average rainfall can vary between different rainfall districts. If a fine spatial resolution is useful, there are a range of methods to produce fine resolution projections, including:

  1. 'Simple scaling of relevant observations by the changes indicated by global climate models. This method can produce realistic looking datasets that can be used to illustrate the effect of the broad changes on local conditions, so achieves goal 1 but not goal 2. This method is simpler and more widely applicable than statistical and dynamical downscaling so is a very attractive option.
  1. Statistical and dynamical downscaling methods. These are also likely to achieve goal 1, but potentially also achieve goal 2. If there is a compelling reason to expect regional detail in the change ‘signal’ then this may be a good option. However, downscaling is more complex and is often not available from all GCMs. It is also not guaranteed to produce reliable finer-scale detail in the change signal. Ideally, the plausibility of these regional patterns should be demonstrated and linked to an underlying factor that has been introduced through the process (e.g. mountain ranges or coastlines).

There are pros and cons of different approaches to producing fine-scale datasets to study climate change impacts in detail, and there is no one best method. Different downscaling techniques give different results, thus introducing another source of range in the final analysis. In practice, the other limiting factor is availability. Results from only a limited number of downscaling methods using a limited set of GCMs as input exist for each region of Australia.

When selecting data inputs for climate change impacts analysis it is important to choose the right tool or tools for the purpose (if they are available). Each study should consider the spatial and temporal dimensions. On the spatial dimension:

  1. General information can be used for an assessment of change over a region, and fine-scale spatial data is often not required. This assessment should cover the broad perspective of natural variability, the various RCPs and the range in climate response to each RCP. The range of climate response can be gauged by the range between GCMs, but it should be acknowledged that changes outside this range are also possible (especially at the regional scale). Once this is complete, further analysis may be justified.
  1. ‘Simple scaling’ of observations may be useful to illustrate what these broad changes mean at the local scale and give a spatial representation (while being clear that the change signal applied was produced at the broad scale). In these cases, scaled datasets available from this website, CliMond (see below) or the Tropical Data Hub (see below) may be useful.
  1. More sophisticated downscaling may be useful if there is reason to think that there could be regional differences in the climate change signal at scales of less than about 200 km (not local differences in the current climate, but differences in the change signal). This can be hard to judge, but such cases include regions where there is very different rainfall drivers on one side of a mountain range compared to another. Cases where this can be useful include:
    1. NSW, where the Eastern Seaboard shows differences form inland, addressed by NARCliM (see below) 
    2. Tasmania, where there are strong climatic gradients, addressed by Climate Futures for Tasmania (see below)

On the temporal dimension

  1. General information can be used in the first instance, without the need for detailed time series. This assessment can include descriptions of changes to climate averages as well as changes to variability and the incidence of extremes. This can inform an assessment of the likely impacts from known relationships. Again, the context of natural variability, various emission pathways and the spread of response from different GCMs is important to consider. Once this is complete, further analysis may be justified.
  1. ‘Simple scaling’ using the average change given by GCMs onto observed datasets can be used to produce locally-relevant time series. These represent the projected change in the average of that variable and do not contain persistent biases in data compared to observations. This is useful to illustrate the effect of a change in an analysis that requires temporal inputs, but only illustrates the change in the average and is generated at the scale of GCM. Datasets are available on the Map Explorer when configuring the options to display the 'change applied to observational data'.
  1. Scaling of observations using more sophisticated methods may be useful to represent projected changes to the average and also variability. Also, sophisticated scaling techniques have been used to produce internally-consistent datasets of multiple variables for use in biophysical models by the Consistent Climate Scenarios project (see below).
  1. Statistical downscaling can produce data that is locally relevant and simulates a change to the average and variability, and produce outputs of several variables. Produced by Goyder for South Australia (see below).
  1. Statistical and dynamical downscaling outputs may be adjusted to exactly match the statistics of observed datasets and so be used directly. Where downscaling is likely to produce valuable regional detail in the climate change signal, these datasets may offer useful insights. This has been done for the NARCliM for and Climate Futures for Tasmania projects (see below).
Project name Agencies Link Models used Scenarios or RCPs Domain Methods to produce fine resolution Notes
NSW and ACT Regional Climate Modelling (NARCliM) NSW government and UNSW AdaptNSW Subset of CMIP3 models chosen by performance and to span the range of change SRES A2 Eastern and southeast Australia Dynamical downscaling to ~10 km grid Expect regional differences in climate change, especially for rainfall (e.g. Eastern Seaboard compared to inland, high altitude compared to coasts)
Climate Futures fro Tasmania ACE CRC and Tasmanian Government TPAC THREDDS Data Server

Climate Futures for Tasmania
Subset of CMIP3 models chosen by performance that also span a range of change SRES A2 and B1 Tasmania Dynamical downscaling to ~10 km grid Regional differences in climate change, especially rainfall changes (e.g. western Tasmania compared to eastern)
Goyder Institute for Water Research Goyder Institute Goyder Institute Subset of CMIP5 chosen on performance RCP8.5 and RCP4.5 Sites in South Australia Statistical downscaling for sites Downscaling to sites, tailored for hydrological analysis (e.g. analysis of runoff and riverflows)
Consistent Climate Scenarios Long Paddock Long Paddock 19 CMIP3 models chosen by performance Sites across Australia Simple scaling of observations to make daily time series Daily time series suitable for biophysical models for periods centred on 2030 and 2050. Internally consistent rainfall, evaporation, minimum and maximum temperature, solar radiation and vapour pressure deficit. Uses the SILO observed dataset
Tropical Data Hub James Cook University All future climate layers for Australia Majority of CMIP3 models All SRES scenarios, and scaled to RCPs Australia Simple scaling of ~5 km gridded AWAP data Simple scaling technique tailored for ecological applications. Large range of models and scenarios available and formats such as BIOCLIM variables
CliMOND (uses the resources of WorldClim) Various CliMOND Two CMIP3 models chosen by performance SRES Global Simple scaling Simple scaling and interpolation to produce high resolution climate surfaces tailored for ecological applications, e.g. 35 BIOCLIM variables, CLIMEX format, Koppen- Geiger climate classification scheme. Uses the CRU observed datasets