Climate Models, Scenarios and Uncertainty for newbies - Part I

Climate models are numerical models that simulate the different components of the climate system (atmosphere, oceans, land and ice-covered regions of the planet) and their interactions. See for instance: https://www.carbonbrief.org/qa-how-do-climate-models-work/, and IPCC AR6 [sections 1.5.3 and 1.6]

‍

The main inputs are the amount of the sun’s energy that is absorbed by the Earth, and how much is trapped by the atmosphere. This depends on the concentration of greenhouse gases (CO2, methane, etc), and aerosols (emitted when burning fossil fuels, forest fires and volcanic eruptions). These external factors are called forcings. 

‍

Climate models can differ substantially in their degree of complexity, or, in other words, how realistically they can simulate the earth system. This covers not only which parts of the climate system a model takes into account, but also the degree to which the different components are coupled. On the low end of the spectrum we typically find relatively simple atmosphere-ocean GCMs (general circulation models). These models can simulate atmosphere-ocean processes and interactions. Often the ocean component takes sea ice into account. On the high end, so-called earth system models can reach a degree of complexity where atmospheric and ocean processes are coupled to sophisticated land surface schemes that include land use,  glaciers, ice caps, interactive vegetation, soil models  and more. The coupling of these components allows for interactive feedback between the different parts of the climate system.

It is important to note that models not only differ  in the number of components but also in the complexity of each component, for example, how complex is the representation of cloud processes in the atmosphere component.

Even though the physical and chemical processes in the climate system follow known scientific laws, the complexity of the system implies that many simplifications and approximations have to be made when modelling the system. The choice of approximations creates a variety of physical climate models [IPCC 2021]. 

There are different sources of uncertainties in climate model projections. Climate forcing or scenario uncertainty is introduced by the fact that to simulate future climate, the models are run using different scenarios of anthropogenic forcings that represent plausible but inherently unknowable future socio-economic development. Climate model and climate variability uncertainties are due to our incomplete knowledge of the climate system, the limitations of computer models to simulate it, and the system’s nonlinearity. The relative and absolute importance of these different sources of uncertainty depends on the spatial scale, the lead-time of the projection and the variable of interest . At shorter time scales, in many cases, the current natural variability of the climate system and other non-climatic drivers of risks will have a higher impact than the climatic changes driven by changes in atmospheric concentrations of greenhouse gases.  

How do you select the climate models? Should they be selected case per case? What are the recommended KPIs to select them?

‍

The selection of models happens on various levels. At the most fundamental, the decision to include a model is simply made by the availability of data for a specific variable. Aspects like time frequency, spatial resolution, available scenarios, play a role here. Second, model output is also checked for technical issues, such as, physically impossible values or large transients between historical and future scenarios.

After this first screening, models could be evaluated with respect to their ability to match key aspects of observed climate. IPCC AR6 states that ‘no universal, robust method for weighting a multi-model projection ensemble is available, and expert judgement must be included, as it did for AR5, in the assessment of the projections’

There are therefore many approaches to this task and no general recipe exists. On a basic level for instance, model biases and temporal variability of the variable in question (e.g., surface wind) can be evaluated against observations. Whether model quality depends on the location can be assessed by examining maps of the evaluation results.
For process based analysis other relevant variables can be included in the evaluation. If necessary, this can be extended to an evaluation of spatial patterns or indices of large-scale drivers of the process of interest. These drivers of variability can be different depending on the location of interest, while NAO can influence the climate in Europe, it is less relevant for North America for instance. It is then definitely necessary to adapt the evaluation to the regional conditions.

How the results of such an evaluation are further used for selecting models depends very much on the specific case.

Weighting individual models by their performance and independence³ (results of the evaluation) can be an alternative to a binary model selection. In the most recent IPCC report, for instance, projections of global surface air temperature are based on weighting models according to their ability to match past warming, their equilibrium climate sensitivity and transient climate response².

¹IPCC Sixth Assessment Report, https://www.ipcc.ch/report/ar6/wg1/

²IPCC Sixth Assessment Report, Section 4.2.6, 4.3.4, IPCC AR6 and Box4.1
‍https://www.ipcc.ch/report/ar6/wg1/chapter/chapter-4/#4.2.6-
https://www.ipcc.ch/report/ar6/wg1/chapter/chapter-4/#4.2.6#box-4.