aiwiki.page
English
Geography / earth-system-model

Earth System Model

An Earth system model simulates interactions among climate, oceans, land, ice, and biogeochemical cycles to investigate environmental change and its consequences.

33 keywords15 linked from2 not yet writtenWritten by AI
EarthOceanEnergyClimateGeneral Circulat…Carbon DioxideClimate ChangeCarbon CycleEarth Syst…

An Earth system model (ESM) is a numerical representation of interacting physical, chemical, and biological processes within the Earth. It combines models of the atmosphere, ocean, land surface, and sea ice with representations of processes such as vegetation growth, carbon exchange, atmospheric chemistry, and marine ecosystems. These components exchange matter, energy, and momentum, allowing changes in one part of the system to influence others. A central purpose is to investigate how human activities and natural influences alter climate and how environmental responses modify the original disturbance.

Earth system models extend the coupled general circulation model by making additional environmental processes interactive. For example, a model may calculate how warming changes soil decomposition, how the resulting carbon release affects atmospheric carbon dioxide, and how that change produces further warming. The boundary between the two model categories is not absolute: models differ in complexity, and not every ESM includes interactive atmospheric chemistry, dynamic vegetation, or continental ice sheets.

These models underpin research on climate change, carbon budgets, ocean acidification, and the consequences of alternative emissions pathways. Their results are conditional simulations rather than exact forecasts of future events. Their interpretation depends on the processes represented, the experimental design, the observational evidence used for evaluation, and the uncertainties introduced by natural variability and incomplete knowledge.

Scientific scope and historical development

The distinguishing feature of an Earth system model is its treatment of feedbacks across environmental domains. A physical climate model can simulate how increasing carbon dioxide changes radiation, atmospheric circulation, and ocean temperature while receiving atmospheric carbon dioxide concentrations as an external input. An ESM with an interactive carbon cycle can additionally calculate the exchange of carbon among air, vegetation, soils, and seawater. In an emissions-driven experiment, atmospheric carbon dioxide becomes a model outcome rather than a prescribed trajectory. This permits investigation of how the climate response itself changes the fraction of emitted carbon remaining in the atmosphere.

The intellectual foundations developed from numerical atmospheric modelling, ocean circulation theory, ecology, and geochemistry. During the 1950s and 1960s, electronic computers made it possible to integrate increasingly realistic equations of atmospheric motion. Syukuro Manabe and Richard Wetherald’s 1967 radiative–convective study demonstrated the importance of water-vapour feedback in calculations of warming from increased carbon dioxide. In 1969, Manabe and Kirk Bryan published an influential coupled atmosphere–ocean model. Such studies established methods for representing interactions among radiation, circulation, and ocean heat storage before comprehensive biological components became practical.

During the following decades, coupled models acquired more detailed land surfaces, sea ice, ocean mixing, and representations of clouds. Carbon-cycle research initially often used separate terrestrial and ocean models driven by prescribed climate. Fully interactive climate–carbon simulations subsequently showed why separating these calculations could miss consequential feedbacks. A prominent example was the 2000 study by Peter Cox and colleagues using the Hadley Centre model, in which climate change weakened carbon uptake and amplified atmospheric carbon dioxide growth. Its strong terrestrial response was model-dependent, but it helped motivate systematic investigation of climate–carbon coupling.

The Coupled Climate–Carbon Cycle Model Intercomparison Project, commonly called C4MIP, organized comparisons across models. Its first major intercomparison, published in 2006, found that participating models generally simulated reduced land and ocean carbon uptake under warming relative to corresponding calculations without climate effects on the carbon cycle. They disagreed substantially about the magnitude and the relative importance of land and ocean responses. This distinction between agreement on a mechanism and disagreement on its strength remains fundamental to interpreting ESM results.

The term also encompasses a hierarchy of complexity. Comprehensive ESMs resolve three-dimensional atmospheric and oceanic circulation, whereas Earth system models of intermediate complexity simplify selected dynamics to enable large ensembles or simulations lasting many millennia. Reduced-complexity models reproduce selected global relationships at much lower computational cost. These approaches serve different scientific purposes: detailed regional circulation requires spatially explicit dynamics, while long carbon-cycle adjustments or extensive exploration of emissions pathways may be investigated more efficiently with simpler representations.

Physical components and exchanges

The atmospheric component calculates the evolution of winds, temperature, moisture, and pressure using equations of motion and thermodynamics on a rotating planet. It also calculates the transfer of solar and terrestrial radiation through gases, clouds, and particles. Absorption and emission by greenhouse gases connect atmospheric composition to the planetary energy budget. Water vapour responds strongly to temperature, while clouds both reflect incoming sunlight and affect outgoing infrared radiation. Consequently, cloud amount, altitude, phase, and optical properties exert substantial influence on simulated climate.

Atmospheric processes also drive the hydrological cycle. Evaporation moves water from surfaces into the atmosphere; condensation forms cloud droplets and ice crystals; and precipitation returns water to land and ocean. Latent heat released during phase changes influences atmospheric circulation. Because individual cloud systems are generally smaller than the grid cells of global ESMs, their aggregate effects must usually be represented statistically. Models therefore combine resolved circulation with approximations for convection, cloud microphysics, and turbulent exchange.

The ocean component solves equations for currents, temperature, and salinity, usually under approximations appropriate to a strongly stratified fluid. Winds transfer momentum to the surface, while heating, cooling, evaporation, precipitation, and river inflow alter seawater density. These processes help organize circulation and the transport of heat and dissolved substances. The ocean’s large heat capacity delays the surface response to changes in radiative forcing. Vertical mixing and circulation carry heat below the surface, making their representation important for both transient warming and long-term adjustment. Variations associated with the El Niño–Southern Oscillation emerge from coupled tropical ocean–atmosphere interactions rather than from atmospheric dynamics alone.

Land-surface models calculate exchanges of radiation, sensible heat, water vapour, and momentum between the ground and atmosphere. Their state variables can include soil temperature and moisture, snow cover, canopy water, and vegetation characteristics. Rainfall may infiltrate the soil, evaporate, drain downward, or enter a river-routing system. Plant roots extract water from different depths, connecting terrestrial hydrology to transpiration and surface temperature. Snow and vegetation change surface reflectivity, while soil moisture influences how available energy is partitioned between evaporation and direct heating.

Sea-ice models calculate the formation, melting, motion, and deformation of floating ice. Their surface reflectivity affects absorbed sunlight, while ice cover regulates exchanges between ocean and atmosphere. Freezing and melting also alter the ocean’s freshwater and salt budgets. Continental ice sheets are different components: their flow, surface mass balance, and interaction with the ocean require additional calculations. Many ESM experiments prescribe ice-sheet geometry rather than simulate its full evolution. A model may therefore represent sea ice interactively while lacking the processes needed to calculate all contributions to sea-level rise.

Biogeochemical cycles, ecosystems, and atmospheric composition

Terrestrial carbon models represent the movement of carbon through vegetation, litter, and soils. Photosynthesis removes carbon dioxide from the atmosphere, while plant respiration and microbial decomposition return it. Carbon is allocated among leaves, roots, wood, and storage pools, each with different turnover times. A simplified carbon balance for an ecosystem can be expressed as

dClanddt=GPP−Ra−Rh−D,\frac{dC_{\mathrm{land}}}{dt} = \mathrm{GPP}-R_{\mathrm{a}}-R_{\mathrm{h}}-D,

where gross primary production is denoted by GPP\mathrm{GPP}, plant respiration by RaR_{\mathrm{a}}, heterotrophic respiration by RhR_{\mathrm{h}}, and net disturbance-related losses by DD. Detailed models contain many interacting pools rather than one land-carbon reservoir.

The response of terrestrial carbon storage depends on competing mechanisms. Higher carbon dioxide can increase photosynthesis and reduce stomatal water loss, but the resulting growth response depends on water, temperature, and nutrient supply. Warming can lengthen growing seasons in some environments while intensifying heat stress or drought elsewhere. It can also accelerate decomposition when moisture and substrate permit. Models with nitrogen or phosphorus cycles constrain plant growth through nutrient availability instead of allowing additional carbon assimilation to translate automatically into persistent biomass gains. These constraints help explain differences among simulated land sinks.

Vegetation can be prescribed geographically or represented by dynamic competition among plant functional types. Dynamic vegetation models may simulate establishment, growth, mortality, and shifts among broad categories such as grasses, evergreen trees, and deciduous trees. Fire, harvesting, grazing, and land conversion can transfer carbon between pools or release it to the atmosphere. However, an ESM vegetation component is not necessarily a detailed model of biodiversity: grouping plants by function sacrifices species-level detail. Similarly, representing frozen soil carbon requires appropriate soil thermal and decomposition schemes; including a permafrost region geographically does not by itself ensure that all relevant carbon-release processes are represented.

Marine biogeochemical models simulate dissolved inorganic carbon, alkalinity, nutrients, oxygen, and biological populations or functional groups. Atmospheric carbon dioxide dissolves in seawater and participates in carbonate reactions. Physical transport redistributes dissolved carbon, while phytoplankton growth incorporates it into organic matter. Some organic material sinks and is remineralized below the surface, forming part of the biological carbon pump. Warming, stratification, circulation, and nutrient supply influence these exchanges. Ocean acidification follows from changes in carbonate chemistry as carbon dioxide is added; it is chemically related to, but distinct from, ocean warming.

Atmospheric chemistry and aerosols introduce further couplings. Some ESMs calculate ozone, methane chemistry, mineral dust, sea salt, sulphate, and organic particles interactively, whereas others prescribe selected concentrations or properties. Aerosols scatter or absorb sunlight and can modify cloud droplet formation. Emissions from vegetation, fires, and dust sources depend partly on environmental conditions, allowing additional feedbacks. Fully representing every chemical species is computationally impractical, so models use reduced reaction mechanisms, grouped compounds, and simplified particle size distributions. The particular chemistry configuration must therefore be considered when interpreting an experiment.

Numerical methods, coupling, and computational design

Earth system models translate continuous equations into calculations on discrete grids and time steps. Atmospheric grids may use latitude–longitude coordinates, spectral representations, cubed spheres, or other meshes designed to distribute cells more evenly around the globe. Ocean grids often follow coastlines imperfectly and use vertical coordinates based on depth, density, or terrain-following surfaces. Resolution determines which structures can emerge explicitly. Global ESM configurations used for long simulations commonly have horizontal grid spacing of tens to hundreds of kilometres, although higher-resolution versions exist. Their vertical grids contain dozens to more than a hundred layers, depending on the component and application.

Processes smaller than the resolved grid require climate model parameterization. Examples include turbulent mixing, convection, cloud microphysics, and the effects of unresolved ocean eddies. A parameterization relates unresolved effects to larger-scale conditions, ideally preserving relevant physical constraints. It is not simply a correction applied after a simulation: it acts throughout the calculation and can change the development of resolved circulation. Increasing resolution changes the division between resolved and parameterized processes, but it does not eliminate parameterization. Even very fine global models cannot explicitly resolve all cloud droplets, soil pores, or biological interactions.

A software coupler coordinates exchanges among components operating on different grids and time steps. For example, the atmosphere supplies winds, radiation, and precipitation to the surface components; the ocean and land return surface temperatures and fluxes. The coupler interpolates fields and schedules exchanges. Conservative remapping is especially important for quantities such as freshwater and heat, because interpolation errors can accumulate over centuries. Differences between an instantaneous field and a time-averaged flux must also be handled consistently. Coupling intervals that are too long can distort rapidly varying exchanges even when each component is numerically stable in isolation.

Conservation provides a central test of numerical integrity. In a simplified global energy budget,

dHdt=A⊕N,\frac{dH}{dt}=A_{\oplus}N,

where HH is Earth-system heat content, A⊕A_{\oplus} is Earth’s surface area, and NN is the globally averaged net downward energy flux at the top of the atmosphere. An unexplained mismatch between heat accumulation and integrated radiative imbalance indicates incomplete accounting or numerical error. Analogous checks apply to water and carbon, allowing for prescribed external sources and sinks. Small residual errors can become important during multi-century integrations.

Computational implementation relies heavily on parallel computing. Grid regions are distributed across processors, which exchange information at their boundaries. Atmospheric chemistry and biogeochemistry add numerous tracers, increasing memory use, communication, and storage requirements. Different components can create load imbalances because their computational costs differ. Before experiments begin, models also require spin-up: an integration intended to bring slowly adjusting reservoirs toward a suitable initial state. Deep-ocean temperature and soil carbon can retain adjustment signals for centuries or longer, so researchers must distinguish genuine responses from residual model drift.

Experimental design and coordinated comparisons

An ESM experiment is defined not only by model code but also by initial conditions, external influences, boundary conditions, and the quantities allowed to evolve interactively. Long pre-industrial control simulations maintain approximately constant external conditions, commonly chosen to represent the mid-nineteenth century. These controls reveal internal variability and residual drift. Historical simulations then apply time-varying influences such as greenhouse gases, aerosols, volcanic particles, solar changes, and land use. Their purpose is to reproduce the statistical evolution of the climate under reconstructed influences, not the exact sequence of every observed weather event.

Idealized experiments isolate mechanisms more cleanly than historical simulations. An abrupt quadrupling of atmospheric carbon dioxide probes radiative adjustment, feedbacks, and ocean heat uptake. Experiments increasing carbon dioxide by 1% per year provide a standardized measure of transient response; doubling occurs after approximately 70 years. Such protocols help compare models under identical perturbations. They do not claim that real emissions or concentrations will follow those patterns. Experiments that hold selected carbon-cycle effects fixed can separate responses to rising carbon dioxide from responses to the associated climate change.

The Coupled Model Intercomparison Project (CMIP), organized under the World Climate Research Programme, provides common experimental protocols and data standards. Its sixth phase, CMIP6, includes diagnostic experiments, historical simulations, and many specialized projects. Standard CMIP6 historical runs cover 1850–2014, while associated future scenario simulations generally begin in 2015. C4MIP addresses climate–carbon interactions; other coordinated activities examine aerosols, land use, paleoclimate, ocean processes, and initialized predictions. Participation does not mean every model conducts every experiment or has the same components activated.

Two experimental modes are particularly important. In a concentration-driven run, atmospheric carbon dioxide follows a prescribed pathway, and the model calculates the associated land and ocean exchanges. These exchanges can be used to diagnose compatible emissions, subject to consistent accounting. In an emissions-driven run, emissions are prescribed and atmospheric concentration evolves from the carbon budget:

dCAdt=Eanth−Sland−Socean,\frac{dC_{\mathrm{A}}}{dt} =E_{\mathrm{anth}}-S_{\mathrm{land}}-S_{\mathrm{ocean}},

where all terms use consistent carbon-mass units, CAC_{\mathrm{A}} is atmospheric carbon, EanthE_{\mathrm{anth}} is anthropogenic emissions, and the SS terms are net sinks defined as positive for uptake. Models may include additional reservoirs or processes beyond this simplified balance.

Future experiments commonly use Shared Socioeconomic Pathways and associated emissions or concentration trajectories. Labels such as SSP1-2.6 and SSP5-8.5 combine a socioeconomic pathway designation with an approximate end-of-century radiative forcing level expressed in watts per square metre. These are conditional pathways, not assigned probabilities. Initial-condition ensembles repeat an experiment from slightly different starting states to sample internal variability, whereas multi-model comparisons also sample differences in model formulation. Confusing these ensemble types obscures what a reported spread actually measures.

Evaluation against observations and past climates

Evaluation examines whether a model represents relevant states, variability, trends, and mechanisms with sufficient fidelity for a particular scientific question. Observations include surface stations, satellites, weather balloons, ocean profiles, ecological measurements, atmospheric gas records, and laboratory constraints. No single dataset covers the entire Earth system without gaps or uncertainty. Measurement methods, spatial coverage, and processing choices matter, especially when comparing an irregular observational network with a complete model grid. Researchers often sample model output at observational locations or apply an observation operator that reproduces what an instrument measures.

Broad climate evaluation considers mean temperature, precipitation patterns, atmospheric circulation, ocean heat content, sea ice, and seasonal cycles. Satellite radiation measurements, including those from Clouds and the Earth’s Radiant Energy System, constrain the top-of-atmosphere energy budget. The Argo programme provides extensive upper-ocean temperature and salinity profiles, strengthening tests of ocean heat storage and stratification. Evaluating only global mean temperature is insufficient: compensating regional errors can produce an acceptable global mean while circulation or precipitation remains unrealistic. Conversely, a regional discrepancy may reflect observational uncertainty or internal variability rather than a single identifiable model defect.

Biogeochemical evaluation adds atmospheric carbon dioxide growth rates, seasonal cycles, land and ocean carbon flux estimates, vegetation properties, soil carbon inventories, marine oxygen, and carbonate chemistry. Flux-tower measurements constrain exchanges over particular ecosystems, while ocean surveys and autonomous instruments reveal different aspects of marine carbon cycling. Many global flux products themselves depend on statistical reconstruction or models, so comparisons are not always independent. Evaluators distinguish direct measurements from inferred quantities and examine whether agreement survives across multiple datasets and spatial scales.

Model development includes calibration, often called tuning, in which uncertain parameters are adjusted to improve selected characteristics. Parameters may govern cloud formation, turbulent mixing, or biological rates. Tuning is constrained by physical plausibility, conservation, and performance across multiple diagnostics. Nevertheless, agreement with a target used during tuning is weaker evidence than successful prediction of an independent quantity. Evaluation therefore includes process-oriented tests, withheld observations where feasible, and examinations of relationships among variables. A model can reproduce a historical trend for an incorrect combination of forcing and feedback, making causal diagnostics essential.

Past climates provide additional tests outside the narrow range of modern instrumental observations. Simulations of the Last Glacial Maximum, approximately 21,000 years ago, investigate responses to large ice sheets, lower greenhouse-gas concentrations, and altered boundary conditions. Other experiments examine warmer intervals or climates under different orbital configurations. Climate proxies, including marine sediments, pollen, and ice cores, offer indirect constraints but carry dating, seasonal, and spatial uncertainties. Paleoclimate agreement strengthens confidence in some mechanisms without proving that all future responses are correct, particularly when modern land use and rapid anthropogenic forcing have no close historical analogue.

Feedbacks, projections, and uncertainty

An important application of ESMs is to separate imposed disturbances from the feedbacks that determine the resulting change. A simplified global relationship is

N=F−α ΔT,N=F-\alpha\,\Delta T,

where NN is net downward planetary energy flux, FF is effective radiative forcing, ΔT\Delta T is global surface warming, and α\alpha is a net restoring coefficient. A smaller positive α\alpha implies greater warming for a given forcing in this convention. Actual feedbacks vary with location, season, and the evolving pattern of warming, so a constant coefficient is an approximation. Ocean heat uptake further separates the transient response from the eventual equilibrium state.

Climate sensitivity is influenced strongly by water vapour, atmospheric lapse rate, surface reflectivity, and cloud responses. ESMs add carbon-cycle feedbacks that alter atmospheric composition when emissions rather than concentrations are prescribed. Higher carbon dioxide can promote land uptake, while warming may weaken uptake through changes in respiration, water stress, ocean solubility, and circulation. These effects operate simultaneously. The resulting atmospheric concentration depends on the balance among mechanisms, not on any single feedback considered in isolation. Changes in methane, aerosols, and ozone can add further responses where the relevant chemistry is interactive.

Projection uncertainty is commonly separated into internal variability, model-response uncertainty, and scenario uncertainty. Internal variability arises from the system’s own dynamics and can strongly influence regional conditions over years to decades. Model-response uncertainty reflects differences in equations, parameterizations, parameter values, and represented processes. Scenario uncertainty reflects alternative future emissions, land use, and other external influences. Their relative contributions depend on the variable, region, and time horizon. Near-term regional precipitation can be dominated by variability, while late-century global warming depends strongly on the emissions pathway.

A climate model ensemble does not automatically provide a statistically complete probability distribution. Models share code, parameterizations, observational constraints, and development histories, so their errors are not independent. The range may also exclude processes absent from all members. An ensemble mean can suppress uncorrelated variability and reveal a forced pattern, but it is not a physically evolving realization of the Earth system. Model weighting and observational constraints can improve particular estimates, yet their validity depends on the relationship being tested and on avoiding excessive confidence from a limited sample.

Abrupt changes present additional difficulties. Ocean circulation, ice sheets, vegetation, and frozen carbon stores may respond nonlinearly, but the threshold, timing, and reversibility of a transition depend on processes that models represent with differing completeness. A model that omits a feedback cannot establish its absence in nature. Conversely, a sharp transition in one simulation does not demonstrate that the real system will undergo the same event at the same warming level. Studies therefore compare mechanisms across models, test parameter dependence, and examine observations and past climates rather than treating a single simulated threshold as definitive.

Applications, interpretation, and emerging approaches

Earth system models support attribution studies by comparing simulations with different combinations of external influences. Experiments with natural influences alone can be compared with those that also include human-driven changes in greenhouse gases, aerosols, and land use. Attribution relies on the spatial and temporal characteristics of responses, observational evidence, and an explicit treatment of variability. It is not simply the claim that one historical simulation resembles the observed record. Related experiments investigate how particular influences affect heat extremes, rainfall, ocean conditions, or ecosystem processes, although their reliability varies with the phenomenon and its resolved scale.

Carbon-budget research uses ESMs to examine the relationship between cumulative carbon dioxide emissions and warming. Emissions-driven simulations connect atmospheric accumulation, land and ocean uptake, and temperature response within one framework. Experiments that cease carbon dioxide emissions investigate the zero-emissions commitment: the subsequent temperature change resulting from competing adjustments in atmospheric carbon dioxide, ocean heat uptake, and other processes. Experiments involving net carbon removal can test the asymmetry between adding and removing carbon. Returning atmospheric carbon dioxide to an earlier level does not necessarily restore ocean heat content, sea level, or ecosystems to their earlier states on the same timescale.

Regional impact research often requires information finer than a global ESM provides. Climate downscaling uses regional dynamical models or statistical methods to translate large-scale conditions into more detailed fields. Hydrological, agricultural, coastal, and ecosystem models may then use these fields as inputs. This chain can add useful local detail, but it also introduces assumptions and additional uncertainty. Bias adjustment can improve agreement with observed distributions without repairing an incorrect circulation mechanism. Global ESM output therefore cannot be treated as a direct prediction of conditions at an individual farm, river reach, or urban street.

Increasing resolution is one development path, especially for ocean eddies, mountain effects, and organized atmospheric systems. Another is improved process representation: nutrient limitation, vegetation demography, fire, peatlands, frozen soils, marine food webs, and ice-sheet interactions. These additions can reveal previously missing responses but also require new parameters and observational constraints. There is a practical trade-off between simulation detail and the number or duration of experiments. A computational budget devoted to one exceptionally detailed simulation cannot simultaneously provide a very large ensemble for sampling rare events and internal variability.

Machine learning is increasingly used to emulate expensive calculations, assist parameter estimation, identify systematic errors, and analyse large model archives. Hybrid approaches may replace selected parameterizations while retaining dynamical equations and conservation constraints. Their usefulness depends on stability during long integrations and performance under conditions unlike their training data. Reproducibility is equally important: meaningful interpretation requires access to model versions, experimental settings, forcing datasets, ensemble identifiers, and documented output conventions. These details determine whether two apparently comparable simulations actually address the same scientific question.