The atmosphere in each grid box of a climate model is defined by a set of characteristics, such as the temperature, air pressure, wind speed and direction, and cloud cover. Exchanges of energy and moisture with neighbouring grid boxes – and with higher or lower atmospheric levels – are also modelled, as is the chemical composition of the air in the grid box. Similar characteristics define the ocean component of the model. The typical size of fluid characteristics, such as eddies and waves, is much smaller in the oceans than in the atmosphere, so ocean grid boxes are made smaller than atmospheric ones to help the models simulate ocean behaviour. Models also calculate the interactions between the atmosphere and ocean, with each component affecting the other.

As well as the atmosphere and oceans, climate models include the solid components of the Earth system – ice, land and vegetation. The continents are built into the model grid, along with ice sheets at the poles. Grid boxes covering land are characterised according to the type of terrain they represent, such as deserts, mountains, forests, scrubland and other types of vegetation. Some models even specify different soil types. The amount of detail that can be incorporated depends largely on the size of a model’s grid boxes, which typically vary from 100 to 400 km across. More complex ecosystem behaviour and vegetation responses to changes in climate are increasingly being included in the models.

Scientists provide climate models with inputs of different scenarios, and then run the models on computers to see how the modelled climate behaves. Models are first run with stable inputs, where the only changes are the regular daily and seasonal cycles of sunlight. If the model produces a realistic simulation of the climate under these stable conditions, scientists proceed to run it with a changing input, such as an increase in greenhouse gas concentrations in the model’s atmosphere. The model then simulates the climate’s response to this change, enabling scientists to explore features of the climate’s behaviour and how it might change under different conditions. Scientists also use the models to investigate the climate’s response to other changes, such as solar variations and volcanic eruptions.

It’s impossible to predict exactly how much carbon dioxide and other greenhouse gases humans will emit over the coming decades. So scientists input the models with a selection of future possibilities. The projected emissions rates in each scenario are based on different assumptions about global economic and technological development. Under a low-emissions scenario, the global average surface temperature is predicted to increase by somewhere in range 1–3 °C by 2100. This compares with a likely warming of 2–6 °C under a high-emissions scenario. This doesn’t sound much, but a 5 °C change in the opposite direction is the difference between the current warm period and an ice age. These predictions are used to provide information to governments making decisions about future emissions.

Carbon dioxide (CO2) has a long atmospheric lifetime, so the CO2 emitted by human activities will remain in the atmosphere for over 100 years on average. Combined with the inherent time-lag in the climate system, this means that a certain amount of future global warming is inevitable regardless of further emissions. Measurements show an increase of about 0.7 °C in global average surface temperature over the 20th century. Scientists have run climate model simulations under the ‘what if’ assumption of constant atmospheric CO2 levels after the year 2000. The models predict that in this scenario, global temperatures would still rise by about 0.6 °C above 1990 levels by 2100 before stabilising. In reality, the CO2 increase continued after 2000 and atmospheric levels are still rising.

Comparisons of climate model output with observed trends and patterns gives scientists confidence in the models’ ability to simulate the climate system. However, the models still have limitations and their complexity is constrained by the available computing power. Scientific understanding of the Earth system is advancing, resulting in better model representations of processes such as ice-sheet dynamics and ecosystem behaviour. Computer performance is also improving and supercomputers used for climate modelling can now carry out trillions of calculations per second. Some of these – such as the Earth Simulator in Japan – are among the most powerful computers in the world. These advances mean that climate models are becoming more and more sophisticated.

Scattered across the Earth’s oceans, 3000 floating data collectors are constantly taking measurements. Lorna McLean uses information from these Argo floats to improve how the ocean is represented in climate models.

The best way to test a climate model is to see how closely it matches observations. Argo floats take measurements at a pinpoint location. But climate models give predictions over large areas. This makes it difficult to see whether a model prediction matches an Argo float measurement.

‘I have developed a way of estimating how the properties of the ocean vary between Argo floats,’ says Lorna. ‘This helps scientists check the accuracy of their models and improve predictions for the future.’

• Climate research and modelling at CEA
• NASA: An Introduction to Modelling
• World Meteorological Organisation