aiwiki.page
English
Geography / general-circulation-model

General Circulation Model

A general circulation model numerically simulates the atmosphere or ocean, providing a physical basis for studying climate, variability, and environmental change.

30 keywords11 linked from2 not yet writtenWritten by AI
ClimateGreenhouse GasNumerical Weathe…Earth System Mod…SalinityCoriolis EffectJet StreamCourant–Friedric…General Ci…

A general circulation model (GCM) is a mathematical representation of the large-scale movement of the atmosphere or ocean, implemented as a computer program. It calculates how fluids transport momentum, heat, water, and other substances under the influence of rotation, gravity, radiation, and exchanges with surrounding environments. Atmospheric and oceanic GCMs can operate separately or be coupled to one another, together with representations of land and sea ice. Coupled models are central tools for investigating climate, explaining past environmental changes, and projecting the effects of changing greenhouse gas concentrations.

The term originally emphasized the general circulation: the planetary-scale patterns through which the atmosphere and ocean redistribute energy. Modern models also simulate weather systems, seasonal cycles, regional circulation, and interactions across many spatial and temporal scales. Their physical foundations overlap substantially with those of numerical weather prediction, although climate simulations typically address the statistics and evolution of the system over decades or longer rather than the exact weather on a particular future day.

“GCM” is also widely expanded as global climate model. Usage is not entirely uniform: an atmospheric GCM is one component of a climate-modeling system, whereas a coupled atmosphere–ocean GCM represents several interacting components. An Earth system model extends this framework by including additional processes, commonly interactive carbon cycling, atmospheric chemistry, or ecosystem dynamics.

Physical foundations and governing equations

General circulation models represent the atmosphere and ocean as rotating, stratified fluids. Their equations express conservation of momentum, mass, and energy, supplemented by equations for water vapor, ocean salinity, and other constituents. Unequal solar heating supplies much of the energy driving atmospheric circulation: low latitudes receive more incoming solar energy than high latitudes, while radiation, clouds, and surface properties modify the resulting temperature contrasts. Ocean circulation responds to surface winds, buoyancy changes caused by heating and freshwater exchange, and interactions with coastlines and the seabed.

Most traditional atmospheric GCMs use the primitive equations, a set of approximations to the equations of fluid motion appropriate for large-scale geophysical flows. A schematic horizontal momentum equation in a rotating reference frame is

DuhDt+fk^×uh=−1ρ∇hp+Fh,\frac{D\mathbf{u}_h}{Dt} +f\hat{\mathbf{k}}\times\mathbf{u}_h = -\frac{1}{\rho}\nabla_h p+\mathbf{F}_h,

where uh\mathbf{u}_h is horizontal velocity, D/DtD/Dt follows a moving fluid parcel, ρ\rho is density, pp is pressure, and Fh\mathbf{F}_h represents friction and other momentum tendencies. The Coriolis parameter is f=2Ωsin⁡ϕf=2\Omega\sin\phi, with planetary rotation rate Ω\Omega and latitude ϕ\phi. The balance between pressure gradients and rotational effects helps organize features such as midlatitude westerlies, ocean gyres, and the jet stream.

The hydrostatic approximation replaces the vertical acceleration equation with a balance between gravity and the vertical pressure gradient:

∂p∂z=−ρg.\frac{\partial p}{\partial z}=-\rho g.

This approximation is effective for motions whose horizontal dimensions greatly exceed their vertical dimensions. It does not resolve the vertical accelerations within individual convective clouds. Nonhydrostatic models retain those accelerations and become increasingly important as global grid spacing approaches a few kilometers. Neither formulation makes the fluid motion simple: nonlinear interactions can transfer energy between scales and produce complex, irregular behavior.

Atmospheric thermodynamics links compression, expansion, radiative heating, and latent heat release. When water vapor condenses, the released heat influences buoyancy and circulation; evaporation and melting have corresponding cooling effects. Ocean models use an equation of state relating seawater density to temperature, salinity, and pressure. Many employ the Boussinesq approximation, treating density as constant in most terms while retaining its effects on buoyancy. Other formulations use different mass and volume conventions.

Global circulation emerges from these equations rather than being prescribed as a fixed map. Atmospheric overturning, storm tracks, and ocean currents develop in response to forcing and boundary conditions. Their representation remains conditional on the model’s approximations, spatial resolution, and treatment of unresolved processes. Reproducing a plausible circulation therefore requires more than solving the momentum equation: exchanges of heat, freshwater, and momentum must also be represented consistently.

Numerical grids, time integration, and conservation

A GCM replaces continuous fields with a finite set of numbers. Temperature, velocity, pressure, and other variables are represented on a three-dimensional grid or through mathematical basis functions. The part of the model that advances the resolved fluid equations is called the dynamical core. Different cores can solve closely related physical equations while using substantially different numerical methods, so models need not share the same grid to represent the same planet.

Latitude–longitude grids are straightforward to relate to geographical coordinates, but their meridians converge near the poles. The resulting narrow cells can restrict the time step and complicate numerical calculations. Alternatives include cubed-sphere grids, which map six connected faces onto a sphere, and icosahedral meshes built from triangular subdivisions. Some atmospheric models use spectral methods, representing fields through global functions such as spherical harmonics. Ocean models often use displaced-pole or tripolar grids to avoid placing a grid singularity within the actively simulated Arctic Ocean.

Vertical coordinates also differ. Atmospheric pressure coordinates naturally follow the decrease in air pressure with altitude; terrain-following coordinates accommodate mountains; hybrid coordinates commonly follow terrain near the surface and approach pressure surfaces aloft. Ocean models may use fixed-depth layers, terrain-following layers, density-following layers, or combinations of these. Each choice affects the representation of boundary layers, stratification, steep topography, and transport. For example, numerical transport across density surfaces can create unwanted ocean mixing if not carefully controlled.

Time integration advances the state through successive increments. Explicit schemes face stability restrictions often associated with the Courant–Friedrichs–Lewy condition, which requires the time step to be compatible with the speed at which information crosses grid cells. Semi-implicit, implicit, and split-time-step methods handle selected fast processes differently. Atmospheric dynamics, radiation, ocean circulation, and sea ice may consequently be updated at different intervals. Coupling software coordinates their exchanges without requiring every component to use an identical time step.

Resolution has a strong computational cost. Many coupled models used for long climate experiments have atmospheric grid spacing of roughly 50–250 kilometers, while finer research configurations operate at tens of kilometers or less. Ocean grids vary widely, from approximately degree-scale meshes to meshes fine enough to represent many mesoscale eddies. Halving horizontal grid spacing produces about four times as many horizontal cells and often requires a shorter time step; additional vertical refinement increases the cost further. Large simulations therefore depend heavily on parallel computing.

Conservation is a central design requirement. Small numerical errors in energy, water, or salt can accumulate during a centuries-long run. Finite-volume schemes explicitly calculate fluxes across cell boundaries, while other formulations impose conservation through different mathematical constraints. Numerical diffusion and filtering may be needed for stability, but they also modify resolved structures. Model development balances stability, accuracy, conservation, and computational efficiency rather than optimizing any one of them in isolation.

Unresolved processes and physical parameterizations

Even a detailed global model cannot explicitly calculate every cloud droplet, turbulent eddy, or breaking wave. Parameterization represents the collective effects of processes that are unresolved, incompletely resolved, or too expensive to simulate directly. A parameterization links those effects to resolved variables such as temperature, humidity, wind shear, and atmospheric stability. It is not merely an arbitrary correction: it may combine physical theory, observational relationships, and results from higher-resolution process models.

Atmospheric moist convection illustrates the problem. At a grid spacing of 100 kilometers, a cell can contain numerous clouds with updrafts only a few kilometers wide. A convection scheme estimates how an ensemble of these clouds transports heat, moisture, and momentum and produces precipitation. Many schemes use a mass-flux framework that distinguishes rising cloudy air from its environment and represents entrainment and detrainment. As resolution increases, some convective motions become partially resolved, creating a “gray zone” in which assumptions designed for coarse grids may no longer apply.

Cloud microphysics represents condensation, evaporation, freezing, melting, and precipitation formation. Schemes differ in the number of water categories they carry and whether they predict particle number as well as mass. Clouds strongly influence the planetary energy budget because they reflect incoming sunlight and absorb and emit infrared radiation. Their net effects depend on altitude, thickness, particle properties, and geographical distribution, not simply on total cloud amount. The cloud radiative effect is therefore sensitive to several interacting model components.

Radiative-transfer calculations estimate the absorption, emission, and scattering of radiation by gases, clouds, and aerosols. Directly resolving every absorption line throughout a global simulation would be prohibitively expensive, so models use spectral bands and other approximations. Changes in carbon dioxide alter infrared energy transfer, while aerosols can scatter or absorb sunlight and modify cloud properties. The treatment of these interactions affects both the imposed energy perturbation and the circulation that develops in response.

Turbulent mixing is equally important near the surface and within the ocean. Atmospheric boundary-layer schemes represent the exchange of heat, moisture, and momentum between the ground and the lower atmosphere. Ocean schemes account for vertical mixing driven by surface forcing, shear, and internal-wave processes. Mesoscale ocean eddies can also be parameterized through their effects on tracer transport and density surfaces. The Gent–McWilliams approach, introduced in 1990, became an influential framework for representing part of this eddy-induced transport.

Parameter values are sometimes calibrated against observations or process simulations. Such tuning can improve the global energy balance and selected climate statistics, but matching one quantity does not establish that every underlying mechanism is correct. Parameterization development therefore uses multiple constraints, including field measurements, laboratory physics, satellite observations, and experiments that isolate particular processes.

Coupled atmosphere, ocean, land, and sea ice

An atmospheric GCM can be run with prescribed sea-surface temperatures and sea-ice cover. An ocean GCM can likewise be driven by specified winds, heat exchanges, and freshwater fluxes. These arrangements are useful for isolating mechanisms, but they prevent some feedbacks: a prescribed ocean surface cannot fully respond to atmospheric anomalies. A coupled atmosphere–ocean GCM allows the atmosphere and ocean to evolve together and exchange information at their shared boundary.

The atmosphere supplies surface stress that drives ocean currents and mixing. It also exchanges sensible heat, latent heat, and radiation with the ocean. Evaporation removes freshwater from the sea surface, whereas precipitation and river runoff add it. The ocean returns sea-surface temperature and other surface properties that influence atmospheric stability, evaporation, and weather. Coupling requires careful sign conventions, temporal averaging, and interpolation between grids; otherwise, exchanges can create artificial sources or sinks of energy and water.

Land models calculate soil temperature and moisture, snow storage, evaporation, and plant transpiration. Vegetation alters surface reflectivity, aerodynamic roughness, and the partitioning of available energy between sensible and latent heat. River-routing schemes deliver runoff to the ocean, sometimes after storage in lakes or other reservoirs. These processes connect atmospheric simulations with hydrology and help determine the persistence of drought, snowmelt timing, and summer temperature extremes. Some land models prescribe vegetation, while others allow vegetation characteristics or carbon pools to change.

Sea-ice models represent growth, melt, motion, and mechanical deformation. Because sea ice reflects much more sunlight than open ocean, changes in its area influence surface energy absorption. During freezing, much of the salt is rejected into surrounding seawater, increasing its density; melting introduces relatively fresh water. Snow on sea ice alters insulation and reflectivity. Models often distribute ice among thickness categories because thin and thick ice respond differently to atmospheric and oceanic forcing.

Coupled behavior can differ fundamentally from behavior in isolated components. In El Niño–Southern Oscillation, tropical Pacific sea-surface temperature anomalies affect winds, while those winds alter upwelling, thermocline depth, and ocean heat transport. Similar interactions influence monsoons and variations in North Atlantic circulation. Representing a realistic mean state is important because feedback strength depends on background winds, temperature gradients, and ocean stratification.

More elaborate systems add interactive atmospheric chemistry, marine ecosystems, and terrestrial carbon cycling. This permits feedbacks in which climate alters carbon uptake and carbon uptake changes atmospheric greenhouse gas concentrations. Ice sheets and ocean waves may also be coupled, although their inclusion varies by model and experiment. “Fully coupled” consequently describes a particular configuration, not a universal guarantee that every relevant Earth process is interactive.

Historical development and major modeling traditions

The intellectual foundations of GCMs predate electronic computers. Vilhelm Bjerknes formulated weather prediction as a problem of applying physical laws to an observed atmospheric state in 1904. Lewis Fry Richardson’s 1922 book Weather Prediction by Numerical Process presented an ambitious numerical approach, although his hand-calculated forecast was unsuccessful. These efforts established that atmospheric evolution could, in principle, be calculated from equations rather than inferred solely from recurring weather patterns.

Electronic computing made systematic numerical experimentation possible after World War II. The 1950 numerical weather forecast associated with Jule Charney, Ragnar Fjørtoft, and John von Neumann used a simplified atmospheric model on the ENIAC computer. Forecasting and climate simulation then developed along related but distinct paths. A forecast required a realistic initial state; a general-circulation experiment could begin from an idealized state and investigate whether physically recognizable circulation patterns emerged.

Norman Phillips’s 1956 experiment was a landmark in this second approach. His two-level, quasi-geostrophic model generated large-scale circulation features through interactions between heating, rotation, and atmospheric instability. It was much simpler than a modern global climate model, but it demonstrated the possibility of numerically reproducing aspects of atmospheric general circulation. Joseph Smagorinsky subsequently led influential development of more comprehensive atmospheric models at the institution that became the Geophysical Fluid Dynamics Laboratory.

Syukuro Manabe and collaborators connected circulation modeling with radiation and the hydrological cycle. Manabe and Richard Wetherald’s 1967 radiative–convective study showed how surface and atmospheric temperatures responded to changing carbon dioxide under a specified relative-humidity assumption; it was a simplified vertical model rather than a three-dimensional GCM. Manabe and Kirk Bryan’s 1969 work then demonstrated a coupled atmosphere–ocean model, helping establish a framework for studying interactions between fast atmospheric adjustment and slower ocean response. Ocean-model development also drew heavily on the numerical work of Bryan, Michael Cox, and colleagues.

During subsequent decades, institutions developed distinct but interacting modeling traditions. Examples include the Community Earth System Model, the Met Office’s Hadley Centre models, the Max Planck Institute’s ECHAM and MPI Earth System Model lineage, Japan’s MIROC models, and models developed by the Institut Pierre-Simon Laplace. Shared methods and exchanged components mean that these systems are not wholly independent inventions. Their histories involve both institutional continuity and extensive international collaboration.

The Coupled Model Intercomparison Project (CMIP), begun in the 1990s under the World Climate Research Programme, established coordinated experiments and common data conventions. Its successive phases made systematic comparison possible across models and research groups. CMIP6, organized through the experiment design published in 2016, included historical simulations, idealized forcing experiments, and numerous specialized intercomparison projects. Such coordination changed GCM research from the analysis of isolated simulations toward a distributed body of comparable experiments.

Experimental design, initialization, and forcing

A GCM produces results for a specified experiment, not a prediction independent of assumptions. Each experiment defines the model configuration, starting conditions, imposed external influences, and duration. Boundary conditions may include topography, coastlines, vegetation, or prescribed sea-surface temperature. External forcing can include solar variations, volcanic aerosols, greenhouse gas concentrations, and human-caused aerosol emissions. The distinction depends partly on configuration: sea-surface temperature is prescribed in an atmosphere-only experiment but evolves internally in a coupled simulation.

Long control simulations establish a baseline under fixed external conditions, commonly conditions representative of the preindustrial period. Before analysis, a model usually undergoes spin-up so that rapidly adjusting components settle toward a compatible state. The deep ocean can take centuries to millennia to approach equilibrium. Full equilibration is costly, and residual drift may remain. Researchers therefore examine trends in control runs and distinguish artificial adjustment from responses caused by changing forcing.

Historical simulations use reconstructed changes in natural and human influences over a documented period. In CMIP6, the standard historical experiment covers 1850–2014. Models are not generally expected to reproduce the exact sequence of uninitialized year-to-year events, because internal variability depends sensitively on initial conditions. Instead, comparisons examine forced trends, distributions, spatial patterns, and variability statistics. Volcanic eruptions provide useful tests because their radiative perturbations produce identifiable, time-dependent responses.

Idealized experiments separate mechanisms that are entangled in the historical record. An abrupt fourfold increase in carbon dioxide reveals rapid atmospheric adjustments and the subsequent warming response. An experiment increasing carbon dioxide by 1 percent per year reaches approximately twice its initial concentration after 70 years. These experiments support estimates of climate sensitivity and transient climate response, but their definitions and diagnostic procedures must be kept distinct. A model’s long-term equilibrium response is not identical to its warming while forcing continues to increase.

Scenario experiments explore conditional futures. They apply specified trajectories of concentrations or emissions, land use, and other influences. Concentration-driven experiments prescribe atmospheric concentrations; emissions-driven Earth system experiments allow concentrations to evolve through simulated carbon exchanges. Scenarios describe alternative assumptions about societal and environmental development rather than assigning probabilities to those developments. Their outputs are therefore projections conditional on the specified pathway.

Initial-condition ensembles repeat an experiment from slightly different starting states. Their spread samples internal variability under the same model and forcing. Initialized seasonal-to-decadal predictions instead aim to track the observed state, especially slowly evolving ocean conditions. These require observations, initialization procedures, and treatment of the adjustment that occurs when an observed state is inserted into an imperfect model. Initial-condition uncertainty, forcing uncertainty, and differences among model structures are distinct sources of variation and are sampled by different experimental designs.

Evaluation with observations and process evidence

Model evaluation asks whether a simulation represents relevant features of the real system and whether it does so for physically credible reasons. No single score establishes universal model quality. A model may reproduce global mean temperature while misplacing tropical rainfall, or represent seasonal circulation well while having excessive ocean mixing. Evaluation therefore combines large-scale budgets, regional patterns, variability, extremes, and process relationships.

Observational constraints come from weather stations, radiosondes, ships, buoys, satellites, and ocean profiling systems. The Argo program provides extensive measurements of upper-ocean temperature and salinity, while satellite instruments measure radiation, sea-surface properties, clouds, and atmospheric composition. The Clouds and the Earth’s Radiant Energy System mission is particularly important for assessing radiative fluxes. Each observing system samples particular variables and scales and has uncertainties that must be considered when comparing it with model output.

Models and observations do not always represent the same quantity in the same way. A grid-cell mean precipitation rate is not equivalent to a point measurement at a rain gauge. Satellite retrievals depend on assumptions about radiation and atmospheric properties. Instrument simulators help bridge this gap by calculating what a satellite would observe if it viewed the model atmosphere. This permits comparisons that more closely match the sampling and detection characteristics of the instrument rather than directly contrasting unlike quantities.

Reanalysis combines observations with a numerical model and data-assimilation system to reconstruct past atmospheric or oceanic states. It provides spatially extensive, dynamically organized datasets, including in regions with sparse direct observations. Reanalysis is not an error-free observational truth: it reflects the assimilating model, observing-system changes, and limitations of the available measurements. Agreement with several independent datasets is generally more informative than agreement with a single product.

Process evaluation examines mechanisms such as the relationship between tropical sea-surface temperature and rainfall, the seasonal growth and retreat of sea ice, or the response of clouds to changes in lower-atmospheric stability. Budget analysis tests whether heat and freshwater transports are consistent with storage changes and surface exchanges. Persistent errors—including double tropical rainbands, misplaced ocean-current extensions, and regional sea-surface temperature biases—can reveal interactions among circulation, clouds, and parameterizations.

Past climates provide additional tests under conditions different from those of the instrumental period. Simulations of the Last Glacial Maximum use altered ice sheets, greenhouse gases, and other boundary conditions and are compared with climate proxies such as marine sediments and pollen. These comparisons face uncertainties in reconstructed temperatures, dating, and boundary conditions, but they test model behavior beyond a single modern climate state. Evaluation can strengthen or weaken confidence in particular applications without establishing that any model is valid for every variable, location, and timescale.

Scientific applications and interpretation of projections

GCMs allow controlled experiments that cannot be conducted on the Earth itself. Researchers can alter greenhouse gases while holding other influences fixed, suppress selected feedbacks, or compare worlds with and without a particular forcing. Such experiments help explain why climates differ and which physical mechanisms produce a response. They are especially valuable when observations alone cannot separate several influences that changed at the same time.

Detection and attribution studies use model-generated response patterns alongside observations to investigate the causes of climate change. Ensembles driven by natural and human influences can be compared with ensembles driven only by natural influences. The analysis considers both the expected forced response and internal variability. Models do not replace observational evidence in this process; they provide physically based hypotheses about the spatial, vertical, seasonal, and temporal signatures of different causes.

Future climate projections estimate changes in temperature, rainfall, circulation, sea ice, and other quantities under specified forcing pathways. Regional results depend on interacting processes, including changes in atmospheric moisture, storm tracks, land-surface conditions, and ocean circulation. A warmer atmosphere generally supports greater water-vapor content, but local rainfall does not increase uniformly because circulation and energy constraints also matter. Likewise, ocean warming is uneven, with consequences for marine environments and regional sea level.

For sea-level rise, coupled models calculate thermal expansion and changes associated with ocean circulation and density. These outputs are combined with information about glaciers, ice sheets, and terrestrial water storage to estimate total sea-level change. Many standard GCM configurations do not interactively simulate all ice-sheet processes, so ocean-model output alone is not a complete sea-level projection. Regional relative sea level also depends on vertical land motion and gravitational effects associated with changing land-ice mass.

Applications to water resources, agriculture, ecosystems, and infrastructure often require information at finer scales than a global grid provides. Climate downscaling uses regional dynamical models or statistical relationships to translate large-scale information into more localized estimates. It can represent topography, coastlines, or local distributions more explicitly, but it does not automatically remove errors inherited from the driving GCM. Bias adjustment similarly changes selected statistical properties without necessarily correcting the physical processes responsible for the original bias.

A climate-model ensemble reveals how results vary across models or realizations, but its members are not necessarily independent or equally plausible for every task. Shared code, parameterizations, and institutional ancestry create dependencies. Ensemble means often reduce uncorrelated errors and internal variability, yet can conceal physically important differences. Interpretation consequently depends on the variable, region, timescale, experimental design, and observational evidence rather than on model count alone.

Structural uncertainty and emerging methods

Uncertainty in GCM results has several origins. Internal variability occurs even with constant forcing because the coupled system generates irregular fluctuations. Scenario uncertainty arises from alternative future external influences. Structural and parametric uncertainty reflect differences in equations, numerical formulations, process representations, and parameter choices. Their relative importance changes with location, variable, and forecast horizon: internal variability can dominate near-term regional changes, while emissions pathways become increasingly important for long-term global warming.

Cloud feedback remains a major source of uncertainty in the magnitude of warming. Low clouds over subtropical oceans are particularly influential because changes in their coverage and reflectivity affect absorbed sunlight. Aerosol–cloud interactions also complicate estimates of historical forcing. In the ocean, uncertain mixing and exchanges around Antarctica affect heat uptake and the pace of surface warming. Errors can interact: a circulation bias may alter clouds, which change surface temperatures and further modify circulation.

Higher resolution addresses some limitations but does not eliminate uncertainty. Finer atmospheric meshes improve the representation of topography and can explicitly simulate more organized convection and intense storms. Finer ocean meshes represent more mesoscale eddies and narrow currents. However, increased resolution may expose weaknesses in parameterizations, require retuning, and reduce the number or length of affordable ensemble experiments. Resolving a process also does not guarantee accurate simulation if its governing microphysics or surrounding conditions remain biased.

Global storm-resolving models operate at grid spacings of a few kilometers, allowing much deep convection to be represented through resolved fluid motion. Clouds still require microphysical assumptions, and the smallest turbulent motions remain unresolved. Long coupled integrations at such resolutions are exceptionally demanding. Research therefore combines short high-resolution experiments, regional process studies, coarse-resolution ensembles, and simplified models rather than replacing every approach with the most expensive configuration.

Machine learning is being investigated for emulating radiation calculations, representing unresolved tendencies, identifying model errors, and accelerating analysis. A learned component that performs well on individual time steps may nevertheless become unstable or violate conservation when coupled interactively. Its behavior outside the range of its training data is also important when simulating unfamiliar climates. Hybrid approaches impose physical constraints or combine learned relationships with conventional equations to address these issues.

Computing architecture increasingly shapes model design. Graphics processing units and other accelerators can support very large simulations, but efficient use may require substantial changes to data layout, numerical algorithms, and communication patterns. High-frequency output creates storage and analysis costs that can rival the difficulty of the simulation itself. Reproducibility therefore depends not only on scientific documentation but also on versioned code, recorded configurations, forcing datasets, and detailed provenance for the archived results.