Physical climate models
A physical climate model simulates the behaviour of the atmosphere, ocean and cryosphere, and their interactions, over long periods of time.
The best-known use of physical climate models is to assess how the various characteristics of the climate system will react under the influence of greenhouse gases, as part of climate change studies.
Physical climate models use the fundamental equations of fluid mechanics to describe the behaviour of the atmosphere and ocean. In other words, atmospheric and oceanic movements obey the laws of motion, gravity and thermodynamics, as well as the principles of conservation of mass, energy and water.
In order to describe changes in the distribution of energy and moisture in three-dimensional space, these equations contain different variables associated with:
-
Winds and ocean currents
-
Exchanges and interactions between the components of the climate system
-
Sources (gains) and sinks (losses) of energy and moisture in the system
The fundamental equations of fluid mechanics are highly non-linear and chaotic in nature. This implies that to obtain an analytical solution (or an exact solution derived from these equations), so many simplifications would have to be made that this solution would no longer have much in common with the real system. In order to minimize simplifications and solve this system of equations, we need to use numerical methods, which can approach the analytical solution.
In practical terms, numerical methods involve transposing the equations onto three-dimensional grids, horizontally (2 dimensions) and vertically (3rd dimension), forming concentric spheres. The size of the grid cells (or pixels) determines the model’s horizontal and vertical resolution.
This grid representation of the equations is encoded in a computer program to be run on a supercomputer based on time intervals called time steps. The length of the time step defines the temporal resolution, which is typically 5 to 20 minutes. The temporal resolution, or time step, indicates the time elapsed within the model between two states of the climate system calculated by the equations.
A physical climate model is therefore a simulator consisting of the computer code enabling the discrete representation (i.e. on grids—one calculated solution per cell or pixel) in three-dimensional space and time (4th dimension) of the fundamental equations of fluid mechanics solved with a numerical method.
The solution produced at each time step by running a physical climate model over a period ranging from a few years to more than a millennium is a climate simulation.
At each time step and at each point of the computational grid, this solution contains the values of the variables found in the fundamental equations, as well as several others coming from the physical parameterization, which allows us to consider processes finer than the computational grid cell size. For example, a climate simulation contains over a hundred descriptive climate variables (temperature, wind, pressure, precipitation, radiation, clouds, humidity, etc.), all of which are consistent with each other from one point on the grid to another over very long periods.
Since the aim of a climate model is to simulate the climate of planet Earth, a detailed portrait of the planetary surface with geophysical data such as soil and vegetation types, continental contours, the location and bathymetry of oceans and bodies of water and the position and height of mountains must be provided. In addition, the model needs to be provided with the chemical composition of the atmosphere, including greenhouse gases, ozone and aerosols, as well as other essential physical characteristics such as the energy received from the sun as a function of time of day, time of year and latitude.
Based on the extent of their computational grid, known as their integration domain, physical climate models can be divided into two groups:
Global climate models or General circulation models (GCM)
GCM’s computational grid offers global coverage (the entire planet). GCMs were the very first physical climate models; initially, they only included the atmospheric part of the climate system and its interactions with the continental land surface. Today, this type of GCM is known as an atmospheric general circulation model (AGCM). When these are used in coupling with physical ocean models, they become coupled atmosphere-ocean general circulation models (AOGCMs).
These have been succeeded by Earth system models (ESMs), which are AOGCMs to which biogeochemical interactions and cycles are added. The carbon cycle was the first to be included in most models. Research is intensifying to add several others. As a result, ESMs are the first climate models to take the interaction between plants and the climate system into account.
Regional climate models (RCM)
A regional climate model (RCM) covers a sub-region of the planet, and is said to have a limited-area integration domain. By focusing on a particular part of the globe, this type of model makes it possible to refine the horizontal resolution of climate simulations in a region of interest, for less computing time than that required by a global model of the same resolution. To keep up the link with the global climate, an RCM must be supplied with global climate data from an ESM or reanalysis at its boundaries. This procedure is called the driving of an RCM.
Physical climate models are complex and require substantial computing resources. Their computer code now reaches several million lines, and the temporal and spatial resolutions of their simulations are getting higher and higher. This allows them to offer the advantage of producing a large number of descriptive climate variables that are consistent with each other at any point in space and at any time. This is an essential asset, which is particularly appreciated when identifying and studying the feedback loops resulting from interactions between the components of the climate system.
Advanced concept
Physical climate models differ in their choice of formulation of the fundamental equations (Eulerian or Lagrangian), in their degree of approximation (Euler, primitive, etc.) and the choice of assumptions (hydrostatic, incompressible, etc.) made in the equations, in the choice of numerical method (finite element, semi-Lagrangian semi-implicit method, etc.) and in the physical parameterization schemes to account for subgrid processes.
