Climate Model Comparison Tool

Resourcesncdc noaa map

Choosing Climate Models

Climate Model Intercomparison Project

Climate Model Intercomparison Project (CMIP)

The Intergovernmental Panel on Climate Change (IPCC) assesses multiple global climate models (GCMs) as part of the international climate change assessment reports. The World Climate Research Programme creates the framework for the models, making the analysis of multiple models much simpler; the aggregation of these models is called the Climate Model Intercomparison Project (CMIP).

Two Climate Model Intercomparison projects are currently in use: CMIP3 (released 2010) and CMIP5 (released 2013). The CMIP5 and CMIP3 datasets both contain output from a large number of GCMs, but are based on two different sets of emission scenarios. CMIP3 is based on the IPCC’s Special Report on Emissions Scenarios that is a collection of story lines that describes possible scenarios of the development of the human population along with different plausible scenarios of energy consumption and the proportion of renewable energy to fossil fuels. CMIP5 is based on a simpler set of scenarios that use varying paths to reach different levels of greenhouse gas concentrations. Both collections of GCMs have scenarios that range from a lower to higher degree of climate change.

If starting a project from scratch, one should definitely use the CMIP5 models that are the newer versions of the CMIP3 models, with the addition of more new models. CMIP3 models could be to continue a project already started with CMIP3 models or if downscaled or other post-processed output is needed that is only available based on CMIP3 models.

CMIIP3 and CMIP5 models  spreadsheets are available here. More information on the spreadsheets is presented below. These spreadsheets include model name, a link to the reference, model type, data format, and atmosphere resolution, among other categories.

Choosing GCM’s for Pilot and Research Projects

In a perfect world, all the appropriate GCM’s with multiple runs would be used. Unfortunately, both time and financial concerns sometimes eliminate this possibility. Instead, pilot projects and full research projects, if enough finances are not available, must make choices on which models to use. The section below is designed to help you make appropriate choices.

A pilot project or study is often conducted to develop a preliminary understanding of a specific subject, usually with the possibility of a future full research study. Pilot projects are limited in scope and are of a shorter duration than a full study. As such, pilot projects use a limited number of GCMs, typically one to three models. Please reference our Research Guide for more information on pilot projects. For a full research project at least eight or nine GCMs are recommended, with the full list of available models being optimal.

Choosing one model or a collection of models is no simple task. It depends greatly on the output format needed for the study (temporal and spatial resolution, variables needed, output downscaled or not?) since not all models are available in the same format. First, one must decide what kind of model output and format is needed. Then one should choose a set of models based on the availability for that format.

If users are just interested in doing a quick pilot study using one to three robust GCMs, one can choose to use one or all of these three models: CCSM4, HadGEM-AO, and MIROC5, which are all well-established models that have shown to be reliable in previous research. These models were the overall recommendation by several climate scientists if “they could only pick three,” but again this depends on the availability of the model and the output format that is needed.

Another method for determining which climate models to use is based on climate model genealogy and a “family tree” of model comparisons. In this method, models’ similarity is documented. When choosing models for the pilot project, you want to pick models that are not closely related. Please reference the following paper for more information: Climate Model Genealogy: Generation CMIP5 and How We Got There. In Figure 1), Column a) “Control State” is used for historical conditions while column b) “Projected Change RCP8.5” is used for future conditions. Models in the same branch are more closely related. Models with branches further to the left in each column have a greater similarity. When selecting models, it is important to capture the most branches on the diagram. In other words, if you want 3 models, you move from the right hand side of the column to the left until you have 3 subgroups. For example, when using Panel b, RCP8.5, you could select a MIROC model, which is farthest to the right and therefore the model with the highest independence. You could then select CSIRO as the next model. For the 3rd model, technically you could choose from any that are left since they all fall in the same “family”. Since IPSL stands on it’s own (only two models on that branch), it seems something from the CCSM4/CESM family would be the next logical choice as “most representative” of the remaining models. If using this method, it will be important to affirm the models picked have the model output and format needed for the study.

The CMIP3 and CMIP5 spreadsheets mentioned earlier will helpful when choosing a set of models. Each available GCM is listed along with their temporal and spatial resolution and available future scenarios. The models have been grouped into three categories: Most Reliable, New and Interesting, and Experimental, based on which models have been around for a long time and shown consistent results, which models are from new modeling groups, and models that include different processes, such as the bio-geo-chemical cycle, that were not in previous versions.

If downscaled GCM output is needed, statistically downscaled output and regional model output, containing information about downscaled GCMs and their sources, can also assist in choosing models for a study. Each spreadsheet gives an overview of a collection of downscaled output using different methods of downscaling. Note that they do not all have the same temporal and spatial resolution, they may have different output variables, and they may cover different regions. Each description also has information on the downscaling method used as well as links to publications.

As can been seen from the spreadsheets, there are two fundamentally different ways to create downscaled output, statistical downscaling and regional climate modeling. Statistical downscaling uses statistical relationships between observations and large-scale GCMs to create a higher-resolution version of the GCM output that more closely resembles the observations. Generally there is more of this type of output available since it is an efficient way of generating high-resolution products; however often there are only few variables available such as minimum and maximum temperature, precipitation, relative humidity, and solar radiation. Regional models are high-resolution climate models that cover a certain region, such as North America. These models use GCM output as their input and then solve equations for the physical interactions between e.g. surface processes and atmospheric circulation, and generally have many more variables available. However they take a long time to run and do not have as many GCMs and scenarios available. It is up to the user to decide which kind of downscaled output is more useful for their particular study. Output for both methods can be found in the links above and the information on the spreadsheets states which method has been used to create it.

If gridded observational datasets of climatic variables are needed, Observational Reanalysis Data give an overview of different options available, along with their temporal and spatial resolution, variables, and references. Station observations are available here.

Please note: the very limited number of climate models used in a pilot study is not extensive enough to provide conclusive results or to develop policies or guidelines. Pilot studies can, however, show trend lines or clarify the research questions for a full study or project.