Climate models

What is a climate model?

The atmosphere, the sea and the rest of the earth are constantly changing. This happens as a result of interactions between these different components, and through external influences and cycles such as diurnal and annual cycles. All these processes follow scientific laws. Therefore, in principle, it will be possible to calculate the changes that occur forward in time, if one has sufficiently detailed knowledge of the conditions at a particular point in time. This is why development and use of weather forecasting models makes sense. Use of a weather forecasting model is simply to perform a gigantic calculation.

Weather forecasting models have been around for a long time and have been used by weather forecasting services around the world to deliver weather forecasts. The first usable weather forecasts were made by a weather forecasting model around 1950 [L32].

Climate models were developed on the basis of existing weather forecasting models. Thus, the climate models could benefit from a long period of experience from weather forecasting models and their development. However, there is an important difference between weather forecasting models and climate models. While a weather forecasting model has to calculate the specific development of the weather a few days ahead in time, a climate model will try to calculate how the weather, and especially the temperature, on average will develop many years ahead in time. Variations from day to day will be of no interest when the results from a climate model are analysed. On the other hand, the average temperature over a longer period of time (for example, a month or longer) will be interesting.

In addition to the usual processes that control pressure, temperature, wind, humidity and cloud formation, the climate models take into account greenhouse gas emissions. Greenhouse gases affect how much heat radiation hits the earth's surface through the greenhouse effect. The IPCC has defined different scenarios for the amount of greenhouse gases that are released from now and throughout the rest of this century. Every time a climate model is run on a computer, one of these scenarios provides input to the model in terms of the size of day-to-day emissions.

In climate models (and weather forecasting models), the earth's surface and atmosphere are modelled using a grid that covers the entire planet. Above each square there is a column with three-dimensional cells going upwards in the atmosphere. Attached to each cell is a set of variables that describes the average physical state of the cell. A square is typically 50x50 km². The height of each cell varies with how far up in the atmosphere the cell is situated. It is smallest at the bottom and increases with height above the surface. The physical conditions of all cells are recalculated for each time step that is simulated. The length of such time steps is typically 30 minutes. The following website has a decent description of how a climate model is structured [L33].

Uncertainty

Climate models are inherently uncertain. There are many different processes that ideally should be simulated. Since there are always limitations in terms of computing capacity on the machines running the models, only a selection of these processes is included in a model. Which processes are selected is made on the basis of many decades of research in geophysics, meteorology, etc., and as computing power increases, new processes will be included in the models.

Some processes are not fully understood by scientists, but there is often observational data about the processes on a large scale. Such processes, and other processes that require too much computing power, are often included in the models using parameters stipulated on the basis of experience. For example, we do not have a good theory for how much light is reflected by direct sunlight falling on snow, ice and other bright surfaces (albedo). Instead of simulating this directly on the basis of physical laws, observations are used to stipulate what percentage of sunlight is reflected back into the atmosphere ([L33]: look for "How are climate models parameterised and tuned?").

There is often uncertainty about the exact value that should be chosen for such parameters. Such values are therefor often determined by trial and error. It will always be possible to estimate a range within which the parameter value must be found. If the model is set up to run over a time period before the present (a so-called hindcast), observations will exist that can be compared with output from the model. Then it will be possible to adjust the parameter value (within the range of realistic values) so that model results are as consistent as possible with observations.

Climate sceptics present such tuning of models as adapting the models so that they provide a desired result (namely that they show global warming). This is a pretty gross allegation about reputable researchers with long experience, and is bordering on allegations of fraud. Such tuning is in fact necessary for the model to work over a longer simulation period. There is no alternative way to do this. However, the research community itself acknowledges that such adaptations should be better documented.

Uncertainty in the models is due to many factors: too low time resolution, too rough grid, too few altitude levels, too few processes that are modelled and uncertainty around parameterisation to name a few. The uncertainty surrounding parameterisation is not a dominant part of this.

Results

Climate models are developed by several different research groups around the world. In 1980, some international organisations created the World Climate Research Programme (WCRP) to coordinate the efforts of the various research communities. The WCRP created in turn the project Coupled Model Intercomparison Project (CMIP), which was given the responsibility of comparing results from the different climate models. The CMIP has collected model results in several different phases. The results from the last phase (CMIP5) were collected in the period 2010-2014, and a new phase (CMIP6) is planned where the deadline for submission of results is set to 2020. This collaboration includes approximately 30 climate models from different research communities [L36].

The models have been developed in different research communities and therefore show some variation in the results. But common for all models is that greenhouse gas emissions must be included in the calculations for the models to be able to simulate the global temperature development observed over the last 30 years. The following figure taken from [L37] (page 6) shows model results from the CMIP3 where the models have been run with and without man-made greenhouse gas emissions:

The red curve shows the observed temperature development in all three sub-figures. The grey "ribbon" represents the simulated temperature development from several different models. In sub-figure (a), the simulations are done with only natural influences (variations in solar activity and ash/dust from volcanic eruptions). Sub-figure (b) shows simulations with only man-made influences (greenhouse gas emissions and sulphur from industry). In sub-figure (c) both types of influences are included (natural and man-made).

There are already some results from the CMIP6. An article in the journal Earth System Dynamics from November 2020 [L140] summarises the results from 33 models used in this project. The models were weighted based partly on how they managed to simulate the historical climate development, and partly to what extent the models were related to each other. The article estimates that this weighting gave 17% more reliable results (skill) than an alternative analysis without the use of weighting. This weighting resulted in estimates for heating that were slightly lower than the corresponding unweighted analysis.

These models analyse climate change based on two different emission scenarios, SSP1-2.6 and SSP5-8.5. These are scenarios with an optimistic (SSP1-2.6) and a pessimistic (SSP5-8.5) prescription of greenhouse gas emissions. They have been created for the CMIP6 project and represent a further development of scenarios used in CMIP5 and earlier (RCP scenarios). An overview of the SSP scenarios can be found in the following article [L141].

The optimistic scenario gave a warming of between 0.7 and 1.4°C in 2100 compared to average temperatures in 1995-2014. This corresponds to a warming between 1.5 and 2.2°C compared to pre-industrial times. The pessimistic scenario gave a warming between 3.9 and 5.4°C compared to pre-industrial times. Both of these estimates are uncertain (66% probability that the temperature increase will be within the specified range).

Climate models have been used for many years. Results are available where the temperature for as early as 1970-1980 was simulated. An article in Geophysical Research Letters from January 2020 [L110] examines how well these climate models were able to predict the observed temperature. Of the 17 models investigated, 9 were able to predict a temperature increase within error margins compared to the observed temperature. The uncertainty range for the observed temperature increase was around 0.11-0.25 °C per 10 year period depending on which period was simulated. The models used different emission scenarios for greenhouse gases, and some of these scenarios have been shown to prescribe either too low or too high emission intensity. The article used a method to compensate for these erroneous emission scenarios, and then found that 13 of the 17 models predicted a temperature development within the uncertainty range. The researchers behind these models could not possibly have adapted the models to give such good agreement with future observations.

Latest update: 2021-07-14