Fact-checked by Grok 2 weeks ago

General circulation model

A general circulation model (GCM) is a numerical framework that applies the fundamental equations of , , and to replicate the large-scale circulation patterns of a planetary atmosphere or . These models discretize continuous physical processes into a three-dimensional of computational cells, enabling the prediction of patterns over short timescales and states over decades or centuries by integrating laws for , , , and moisture. Pioneered in 1956 by Norman Phillips, who demonstrated the feasibility of such simulations using a two-level quasi-geostrophic model on early computers, GCMs evolved from efforts to understand atmospheric dynamics and extend principles to global scales. Early models focused on atmospheric components alone, but coupled atmosphere-ocean GCMs emerged in the 1980s to capture interactions across Earth's , including land surface and processes. These advancements have supported projections of phenomena like global temperature changes and precipitation shifts, though empirical validation remains constrained by computational limits and observational gaps. Despite their foundational role in climate science, GCMs exhibit significant uncertainties arising from the parameterization of subgrid-scale processes—such as , , and —that cannot be explicitly resolved due to grid resolutions typically spanning tens to hundreds of kilometers. Structural differences across models lead to divergent simulations of key feedbacks, like tropical responses, contributing to wide ranges in equilibrium estimates. Peer-reviewed assessments highlight that while GCMs reproduce broad observed features, such as seasonal cycles, their long-term predictive skill is limited by incomplete representation of natural variability and effects, underscoring the need for ongoing empirical scrutiny over reliance on ensemble averages.

Definition and Fundamentals

Terminology and Scope

A general circulation model (GCM) is a numerical representation that approximates the three-dimensional, time-dependent solutions to the equations governing fluid motion in planetary atmospheres or oceans, discretized on a global grid to compute variables such as temperature, velocity components, pressure, and precipitation. These models incorporate physical laws derived from , , and , driven primarily by spatial gradients in solar insolation, planetary rotation via the Coriolis effect, and surface boundary conditions like and land-ocean contrasts. The terminology "general circulation" specifically denotes the simulation of large-scale, statistically steady patterns of , , and transport, as opposed to localized or transient phenomena. In scope, GCMs encompass global domains spanning from the surface to the upper atmosphere or depths, resolving explicit for grid-scale processes while parameterizing unresolved subgrid-scale phenomena such as , , and cloud microphysics. Atmospheric GCMs (AGCMs) focus solely on tropospheric and stratospheric circulation, often coupled to prescribed sea surface temperatures for climate studies; oceanic GCMs (OGCMs) analogously simulate currents, , and ; and coupled atmosphere-ocean GCMs integrate these with land surface and components to capture feedbacks in the full , emphasizing Earth's energy balance over multi-year to centennial timescales. Unlike models, which apply similar dynamical cores but prioritize high-resolution initial-value forecasts over days using real-time observations, GCMs generate ensemble statistics for long-term means, variability, and projections under forcing scenarios, such as altered concentrations. This distinction arises from computational constraints and the chaotic nature of atmospheric flows, where GCMs average over ensembles to isolate forced responses from internal variability. The foundational coupled GCM, developed at the Geophysical Fluid Dynamics Laboratory in the , marked the shift toward comprehensive Earth system simulations, enabling attribution of observed climate changes to natural versus drivers. Modern GCMs, as used in assessments like those from the , typically feature horizontal resolutions of km and vertical layers numbering –100, balancing fidelity to observations with feasible computation on supercomputers.

Governing Physical Principles

General circulation models (GCMs) derive their foundational dynamics from the conservation laws of physics, including mass, momentum, and energy, applied to fluid motion on a rotating sphere. These principles are encapsulated in the primitive equations, a set of partial differential equations that approximate the compressible Navier-Stokes equations under the hydrostatic balance assumption, which holds for large-scale flows where vertical accelerations are negligible compared to gravitational forces. The primitive equations thus prioritize horizontal momentum balance influenced by Coriolis forces, pressure gradients, and frictional effects, while treating vertical structure through hydrostatic equilibrium: \partial p / \partial z = -\rho g, where p is pressure, \rho is density, g is gravity, and z is height. The horizontal momentum equations in the primitive set are: \frac{D\mathbf{u}}{Dt} + f \mathbf{k} \times \mathbf{u} = -\frac{1}{\rho} \nabla_p \phi + \mathbf{F}, where \mathbf{u} is the horizontal vector, D/Dt is the , f = 2\Omega \sin\phi is the Coriolis parameter (\Omega being Earth's rotation rate and \phi ), \nabla_p is the horizontal gradient on pressure surfaces, \phi is , and \mathbf{F} represents viscous and other forces. The ensures mass conservation: \frac{\partial \omega}{\partial p} + \nabla \cdot \mathbf{u} = 0, with \omega = dp/dt as vertical velocity in pressure coordinates. The thermodynamic equation governs : \frac{D\theta}{Dt} = Q, where \theta is potential temperature and Q includes heating terms like release and , linked via the equation of state p = \rho R T (). These equations neglect sound waves through the anelastic or hydrostatic approximations, enabling efficient computation of synoptic-to-global scales without resolving acoustic timescales. For oceanic GCMs, analogous primitive equations apply, incorporating the Boussinesq approximation to filter gravity waves and treat density variations primarily through : \frac{D\mathbf{u}}{Dt} + f \mathbf{k} \times \mathbf{u} = -\nabla \phi + b \mathbf{k} + \mathbf{F}, where b = -g \delta \rho / \rho_0 is , alongside incompressibility \nabla \cdot \mathbf{u} = 0 and a temperature-salinity equation for evolution. Radiation and phase changes enter as source terms, but their explicit resolution is limited by grid scales, necessitating parameterizations elsewhere; the framework ensures dynamical consistency with observed circulations like Hadley cells or gyres when forced by realistic boundary conditions. Empirical validations, such as numerical convergence studies to resolutions below 10 km, confirm that solutions approach physical limits under dry adiabatic conditions, underscoring the robustness of these principles despite computational constraints.

Model Architecture

Spatial Discretization and Grids

Spatial in general circulation models (GCMs) involves approximating the continuous partial differential equations of atmospheric and oceanic dynamics on a set of points, transforming the spherical domain of into a computational grid. This process is essential for , as it enables , finite volume, or methods to solve the governing equations while preserving key properties like and where possible. Horizontal discretization typically occurs on quasi-uniform or structured grids to handle the sphere's curvature, while vertical discretization uses coordinate transformations such as terrain-following levels or levels to resolve atmospheric layers from surface to upper . The most traditional horizontal grid is the latitude-longitude (lat-lon) system, where points are spaced uniformly in (e.g., 1° to 2.5° intervals) and at fixed latitudes, resulting in rectangular cells that converge toward the s. This grid simplifies implementation for transform methods but introduces the "pole problem": grid cells shrink to zero size at the s, violating the Courant-Friedrichs-Lewy (CFL) stability criterion due to excessively short time steps required near the s, and causing numerical noise from grid-point singularities. To mitigate this, models apply semi-Lagrangian , polar filtering, or reduced Gaussian grids that omit points near the s, allowing resolutions like T159 (approximately 125 km) in operational GCMs. Gaussian grids address some lat-lon limitations by selecting latitude points as roots of , enabling exact quadrature for spectral expansions in global GCMs and avoiding interpolation errors in transform methods. These grids pair with spherical harmonic basis functions for horizontal representation, computing derivatives analytically in space before transforming to space for nonlinear terms, which enhances accuracy for smooth large-scale flows but can suffer from that requires dealiasing techniques. Spectral methods on Gaussian grids have been foundational in models like those from ECMWF, supporting resolutions up to T799 (about 25 km) while maintaining computational efficiency through fast Fourier transforms. To overcome uniformity issues in lat-lon grids, quasi-uniform alternatives like icosahedral and cubed-sphere grids have gained adoption. Icosahedral grids subdivide the faces of a projected onto the sphere, yielding hexagonal or triangular cells with nearly equal areas (e.g., spacing of 100 km), which eliminate pole singularities and support scalable on Voronoi tessellations for finite-volume schemes. Cubed-sphere grids tile the sphere with six quadrilateral faces from a , providing quasi-uniform resolution (e.g., 0.25° effective spacing) and benefits for , as used in NASA's GEOS model and CESM, though they introduce seams requiring careful flux reconstruction. These grids improve conservation and reduce compared to lat-lon systems, particularly for high-resolution (sub-10 km) simulations, but demand more complex coding and higher memory for .

Parameterizations for Subgrid-Scale Processes

In general circulation models (GCMs), spatial resolutions of approximately 50–250 km horizontally preclude explicit resolution of subgrid-scale processes, necessitating parameterizations to approximate their aggregate effects on resolved variables such as , , and fluxes. These processes, including deep , boundary-layer , and formation, operate on scales of 1–10 km or smaller and exert critical influences on large-scale dynamics, yet their representation relies on empirical or closures rather than direct . Traditional parameterizations introduce structural uncertainties, as evidenced by inter-model spreads in and feedbacks, often requiring to observational datasets for realism. Convection parameterizations predominantly adopt the mass-flux approach, decomposing subgrid updrafts and downdrafts into organized transports with prescribed , detrainment, and assumptions like convective quasi-equilibrium, where is rapidly relieved. Schemes such as the original Arakawa-Schubert formulation or its derivatives, including Tiedtke's bulk mass-flux variant, compute cloud-base based on and inhibition, thereby simulating vertical redistribution of heat and moisture. These methods capture essential features of organized but struggle with scale transitions in higher-resolution "gray-zone" simulations (around 10 km), where partial resolution of plumes leads to double-counting or underestimation of transports, prompting scale-aware modifications that reduce as grid spacing decreases. Turbulence in the and free is parameterized via diffusion closures, with first-order schemes applying eddy viscosities for vertical mixing, often augmented by nonlocal terms for convective boundary layers. Higher-order closures, such as those prognosticating turbulent or using probability density functions (PDFs) for subgrid variability, provide more comprehensive representations; for instance, the Cloud Layers Unified By Binormals (CLUBB) scheme unifies treatment of , shallow , and boundary-layer clouds by modeling joint PDFs of velocity and buoyancy. These approaches address non-local mixing but remain computationally intensive and sensitive to stability functions, contributing to biases in surface fluxes and low-level winds when validated against large-eddy simulations. Cloud and microphysics parameterizations handle subgrid condensate formation, often diagnostically linking cloud fraction to relative humidity exceedance or convectively detrained moisture, with overlap assumptions (e.g., random or maximum) affecting . Prognostic schemes track water/ice paths, incorporating autoconversion and sedimentation for , but their coupling to and schemes frequently underpredicts low- cover and , exacerbating shortwave biases in midlatitudes. Overall, these parameterizations' foundations—relying on bulk assumptions rather than scale-invariant physics—underscore persistent challenges in faithfully reproducing observed variability, with ongoing refinements targeting improved process interactions for coupled atmosphere-ocean GCMs.

Numerical Methods and Flux Conservation

Finite difference methods, pioneered in early atmospheric models such as those developed by in 1956, approximate spatial derivatives via expansions on structured grids like latitude-longitude or cubed-sphere configurations, enabling straightforward implementation but prone to issues like the pole problem in polar regions where grid points converge. Finite volume methods, as implemented in dynamical cores like GFDL's, integrate the governing equations over discrete control volumes, computing fluxes across cell faces to inherently enforce local , , and , which is essential for long-term stability in climate simulations. Spectral methods transform variables into global basis functions, such as or , offering high accuracy for smooth flows and efficient handling of spherical geometry but requiring dealiasing techniques to mitigate Gibbs oscillations and ensure . Flux conservation in GCMs prevents artificial accumulation or depletion of conserved quantities, such as dry mass and total , which could otherwise induce spurious trends over multi-decadal runs; for instance, non-conservative schemes have been shown to cause energy drifts exceeding observational uncertainties in atmospheric models. In finite volume and finite difference approaches, is achieved by designing monotonic, positivity-preserving flux limiters (e.g., van Leer or schemes) that reconstruct variables at interfaces while satisfying the telescoping property of integrated fluxes, as demonstrated in operational models like ECMWF's IFS. Spectral models enforce global through quadrature rules that integrate exactly over the sphere and post-processing adjustments, though they may violate local , necessitating hybrid schemes for coupled systems where ocean-atmosphere interfaces demand precise flux matching. Advanced techniques, including discontinuous Galerkin methods, further enhance flux by using flux integrals along element boundaries, reducing errors in high-resolution simulations. Time-stepping schemes, typically explicit or semi-implicit, must couple with spatial to maintain overall ; for example, schemes with Asselin filters control computational modes in GCMs, while implicit treatments of gravity waves in models (e.g., via the ECMWF semi-implicit scheme since 1975) allow larger time steps without violating flux balances. Validation of these methods against benchmarks, such as Held-Suarez tests, confirms that conservative formulations yield statistically steady circulations with minimal drift, whereas non-conservative variants exhibit unphysical warming or cooling rates. In coupled GCMs, interfacial flux is often enforced via adjustments like those in coupling software, mitigating biases from mismatched grids and ensuring consistency with empirical energy budgets derived from satellite observations.

Types and Configurations

Atmospheric-Only GCMs

Atmospheric-only general circulation models (AGCMs) simulate the dynamics and physics of the Earth's atmosphere by numerically solving the Navier-Stokes equations in spherical coordinates, along with equations for , continuity, and , while prescribing time-varying lower boundary conditions such as observed or modeled sea surface temperatures (s) and sea ice concentrations. These models typically operate on global grids with horizontal resolutions ranging from 50 to 250 km and vertical levels extending from the surface to the or lower , incorporating parameterizations for sub-grid processes like , cloud formation, and turbulence. By excluding interactive ocean and land components, AGCMs enable controlled experiments to isolate atmospheric responses to specified forcings, such as SST anomalies associated with El Niño-Southern Oscillation (ENSO). AGCMs trace their origins to early numerical weather prediction models developed in the 1950s, evolving into comprehensive atmospheric simulations by the 1960s through efforts at institutions like the and the Geophysical Fluid Dynamics Laboratory (GFDL). Notable early examples include the GFDL spectral models, which advanced from barotropic to primitive equation formulations, enabling the first multi-year integrations of global in the late 1960s. Modern implementations, such as NASA's GEOS-5 AGCM, build on these foundations with enhanced resolution and physics, supporting configurations for both free-running and nudged simulations aligned to reanalysis data. Key examples of operational AGCMs include the Australian Community Climate and Earth-System Simulator (ACCESS) version 1.0 atmosphere-only configuration, which uses prescribed SSTs to constrain 70% of the surface temperature field to observations, and the UCLA AGCM, employed in coupled and uncoupled modes for ENSO prediction experiments since the 1990s. These models often employ finite-volume or spectral dynamical cores to ensure conservation of mass, momentum, and energy, with horizontal resolutions as fine as 25 km in high-resolution variants for studying phenomena like tropical cyclones. AGCMs are applied in seasonal-to-interannual forecasting by forcing ensembles with predicted or observed SSTs, revealing atmospheric teleconnections such as the Pacific-North American pattern during ENSO events, and in paleoclimate studies by imposing proxy-reconstructed SSTs to assess shifts. They also facilitate attribution studies, such as evaluating the atmospheric impact of volcanic aerosols or forcings under fixed oceanic boundaries. Despite their utility, AGCMs exhibit limitations due to the absence of ocean-atmosphere , resulting in unrealistic flux biases in midlatitudes and inadequate representation of coupled modes like the Madden-Julian Oscillation's full variability. For instance, AGCM predictions of midlatitude oceanic fluxes diverge from coupled general circulation models (CGCMs) by up to 20 W/m² in seasonal means, underscoring the need for coupled systems in long-term climate projections. Validation against satellite-derived fields and reanalyses often highlights systematic errors in tropical and stratospheric circulation, attributable to parameterization uncertainties.

Oceanic GCMs


Oceanic general circulation models (OGCMs) numerically simulate the three-dimensional movement of , including fields, , and distributions, to represent basin-scale to dynamics. These models solve the of motion, comprising prognostic equations for horizontal , tracer ( and ), and a diagnostic for hydrostatic , typically under the Boussinesq that treats as constant except in terms. The hydrostatic assumes vertical accelerations are negligible compared to gravity, simplifying the vertical to a balance between and weight.
OGCMs discretize the ocean domain on structured grids, such as latitude-longitude or , with vertical levels using z-coordinates (fixed depth), terrain-following sigma coordinates, or hybrid schemes to resolve and . Sub-grid-scale processes, including turbulent mixing, mesoscale eddies, and air-sea fluxes, are parameterized due to limits that prevent explicit ; for instance, eddy viscosities and diffusivities are applied to mimic unresolved lateral and vertical transports. Initial spin-up integrates the model from rest under climatological forcing to achieve quasi-equilibrium circulation, often requiring decades of simulated time. Key implementations include the Modular Ocean Model (MOM), a flexible hydrostatic code supporting generalized vertical coordinates and mass-conserving formulations, developed at NOAA's Laboratory for process to planetary-scale studies. The Parallel Ocean Program (POP) version 2 uses a z-level grid with an implicit , optimized for in global simulations. NEMO (Nucleus for European Modelling of the Ocean) provides a framework configurable for regional or global domains, incorporating advanced options for biogeochemical tracers and sea-ice coupling. These models have evolved since early global efforts in the late 1970s, with refinements in resolution and physics enabling hindcasts of observed circulations like the thermohaline conveyor.

Coupled Atmosphere-Ocean GCMs

Coupled atmosphere-ocean general circulation models (AOGCMs) integrate an atmospheric general circulation model with an general circulation model, enabling bidirectional exchanges of , heat, freshwater, and radiative fluxes at the air-sea interface. These interactions simulate the coupled dynamics essential for phenomena like El Niño-Southern Oscillation (ENSO) and decadal climate variability, which cannot be adequately captured by uncoupled models using prescribed s (SSTs). Initial efforts to develop AOGCMs occurred in the late 1960s and early 1970s, with pioneering work by Manabe and Bryan in demonstrating basic coupled simulations, though limited by coarse resolutions and computational constraints. By the mid-1980s, coupled models supplanted atmospheric-only GCMs as the standard for studies, incorporating dynamics to address deficiencies in SST variability representation. Development accelerated in the , with models run synchronously to produce multi-century integrations for equilibrium states. A primary challenge in early AOGCMs was climate drift, arising from mismatches in simulated meridional heat and freshwater transports between the atmosphere and components, leading to unrealistic SST trends. To mitigate this, flux adjustments—artificial corrective fluxes derived from uncoupled model differences—were introduced in models like those from the Hadley Centre and GFDL, ensuring stable pre-industrial despite underlying parameterization errors. Critics argue flux adjustments obscure physical deficiencies rather than resolving them, prompting later generations to prioritize improved subgrid-scale parameterizations and higher resolutions for drift-free . Prominent examples include HadCM3, developed by the UK Met Office in the late , which operates without flux adjustments at 2.75°×3.75° atmospheric and 1.25°×1.25° oceanic resolutions, simulating realistic tropical biases and ENSO variability. GFDL's CM3, introduced around 2006, couples the AM3 atmosphere with MOM4 ocean, emphasizing refined physics for ocean circulation and , contributing to CMIP5 assessments. These models participate in frameworks like the (CMIP), standardizing evaluations across institutions for projections and variability studies. Ongoing advancements focus on resolving coupled feedbacks, such as air-sea interactions in the Maritime Continent, using nested regional models or enhanced resolutions to reduce biases in and circulation. Despite progress, persistent issues include high computational demands—requiring supercomputers for simulations—and incomplete representation of ocean mesoscale eddies, which influence uptake. AOGCMs thus provide a hierarchical tool for dissecting system responses, though validation against paleoclimate proxies reveals limitations in capturing low-frequency variability without additional forcings.

Computational Framework

Software and Algorithms

General circulation models (GCMs) are predominantly coded in , valued for its efficiency in handling large-scale numerical arrays and vectorized operations essential for simulating over global grids. This choice stems from the field's origins in the mid-20th century, when enabled early computational experiments, leading to extensive legacy codebases that prioritize and performance on systems over modern language features. Complementary languages include or C++ for low-level optimizations, such as in parallel environments, and for scripting, data I/O, and post-processing tasks. Key software frameworks orchestrate GCM components, such as the Community Earth System Model (CESM), an open-source system that couples atmospheric, oceanic, land, sea ice, and biogeochemical modules through the CPL7 flux coupler. CPL7 manages asynchronous data exchanges, interpolates fluxes (e.g., heat, momentum, freshwater) between disparate grids using methods like bilinear or conservative remapping, and enforces conservation laws to prevent spurious energy drifts in long simulations. Similarly, the NOAA Laboratory (GFDL) employs the Flexible Modeling System (FMS), which supports modular assembly of GCMs with built-in parallelism via (MPI) for distributed-memory architectures. For European models, the OASIS3-MCT coupler enables parallel, conservative interpolation across model domains, reducing interpolation errors in coupled atmosphere-ocean simulations by up to 10-20% compared to non-conservative schemes. These frameworks abstract low-level I/O and communication, allowing scientists to focus on physics while leveraging libraries like for gridded data storage. Core algorithms emphasize numerical stability and efficiency. Time-stepping routines often employ explicit schemes, such as the leapfrog method for prognostic variables in atmospheric GCMs, which alternates between time levels to damp computational modes while advancing solutions with time steps on the order of 10-30 minutes for typical resolutions. Implicit or semi-implicit solvers handle stiff terms like or oceanic baroclinic modes, enabling larger effective time steps via techniques like distorted physics, where horizontal is amplified to relax constraints without altering long-term equilibria. Coupling algorithms in frameworks like prioritize flux through Schwarz or great-circle mappings, minimizing artificial sources/sinks that could bias budgets by 0.1-1 W/m². Emerging hybrid approaches, such as NeuralGCM, integrate surrogates for subgrid processes within traditional dynamical cores, accelerating simulations by factors of 10-100 while preserving skill in mid-latitude weather patterns. However, these remain experimental, as core GCMs rely on deterministic physics-based solvers verified against observational benchmarks.

Hardware Requirements and Parallel Computing

General circulation models demand substantial hardware resources due to the computational intensity of solving coupled partial equations for atmospheric, oceanic, and other components over global three-dimensional , often involving billions of cells and millions of time steps per . High-resolution configurations, such as those targeting sub-100 km spacing and dozens of vertical levels, typically require supercomputers with peak performances exceeding 1 petaflop (10^15 floating-point operations per second) and memory capacities in the petabyte range to handle the data volume from prognostic variables like , , and . For context, a single century-long at moderate (e.g., 1° ) can consume 10^5 to 10^6 core-hours on multi-core clusters, scaling to months of wall-clock time on dedicated partitions of national facilities like those in the list. Parallel computing frameworks are essential for feasibility, leveraging domain decomposition to partition the spatial into subdomains assigned to individual processors or nodes, minimizing communication overhead while maximizing load balance. In distributed-memory systems, the (MPI) standard facilitates data exchange across nodes for boundary updates and global reductions, with hybrid MPI plus approaches combining distributed and shared-memory parallelism for intra-node efficiency on multi-core CPUs. Spectral transform methods, common in atmospheric GCMs, exhibit strong scalability due to their separable computations in and grid spaces, achieving near-linear speedup up to thousands of processors in models like those used for CMIP. Grid-point formulations, prevalent in oceanic components, employ similar 1D or 2D decompositions but require careful halo exchanges to maintain locality. Advancements in hardware, including graphics processing units (GPUs) for accelerating parameterizations and linear solvers, offer potential speedups of 2-10x over CPU-only runs for certain kernels, though full model porting remains challenged by irregular memory access patterns in dynamical . Coupled atmosphere-ocean GCMs amplify requirements, with between components necessitating asynchronous strategies to sustain parallelism across heterogeneous architectures. For CMIP6, aggregate computational demands across participating models totaled over 10^9 for core experiments, underscoring reliance on exascale-capable systems for future higher-fidelity ensembles.

Simulation Timescales and Resolution Limits

Spatial resolutions in general circulation models (GCMs) are constrained by computational resources, with typical horizontal grid spacings in Coupled Model Intercomparison Project phase 6 (CMIP6) atmospheric components ranging from 25 km in high-resolution variants to 250 km in coarser configurations. Vertical resolutions commonly feature 30 to 100 levels to represent atmospheric layers from the surface to the or higher. These choices balance fidelity in resolving large-scale dynamics against the exponential growth in computational demands; finer grids increase the number of grid points cubically for three-dimensional domains, scaling roughly as (\Delta x)^{-3} for fixed vertical extent. Temporal resolution, dictated by numerical stability criteria such as the Courant-Friedrichs-Lewy condition, requires time steps on the order of minutes (typically 10-30 minutes) to prevent instability in explicit schemes for advection-dominated processes. Doubling spatial resolution not only multiplies grid points but also halves allowable time steps, yielding an overall computational cost increase of approximately one order of magnitude per simulation unit time. This scaling precludes routine global simulations below 10-25 km horizontal resolution for extended periods, as such efforts demand supercomputing resources equivalent to thousands of CPU-years, often limiting applications to short-term weather-like forecasts or regional domains rather than century-scale climate integrations. Simulation timescales in GCMs prioritize long-term averages for climate statistics, with standard runs spanning 150 years for historical hindcasts (e.g., 1850-2000) and projections extending to 2100 or 2300 under emission scenarios. At coarser resolutions (e.g., 100-250 km), multi-century or millennial simulations are feasible on modern supercomputers, enabling equilibrium assessments, though they rely heavily on parameterizations for unresolved sub-grid phenomena like deep , which require resolutions below 5-10 km for explicit resolution. Higher-resolution models, while improving representation of mesoscale features such as storm tracks, face trade-offs: extended runs become impractical, restricting ensemble sizes (often to 1-5 members versus dozens at low resolution) and hindering robust . For paleoclimate or very long-term studies, coarser grids or hybrid approaches (e.g., emulators) are employed to circumvent these limits.

Validation Against Empirical Data

Observational Benchmarks

General circulation models (GCMs) are benchmarked against observational data to evaluate their fidelity in simulating Earth's , focusing on climatological means, variability, and spatial patterns from instrumental records spanning decades to centuries. Empirical datasets include surface station measurements from networks like the Global Historical Climatology Network (GHCN), satellite-derived products such as those from NASA's for radiation budgets, ARGO floats for , and reanalyses like ERA5 for upper-air fields. Validation typically involves control runs (equilibrium simulations under constant forcing) or historical simulations driven by observed forcings, assessing metrics like root-mean-square error (RMSE) and pattern correlations against these observations. Near-surface air serves as a primary , where multi-model ensembles from phases like CMIP5 and CMIP6 reproduce the observed global mean warming of approximately 0.85°C from 1880 to 2012, but with regional discrepancies; for instance, models often overestimate mid-tropospheric warming in the compared to radiosonde and records showing slower rates. Precipitation patterns reveal persistent biases, including the double-ITCZ problem, where CMIP6 models still simulate spurious equatorial rainfall maxima in both hemispheres, contrasting with asymmetric observations dominated by a single ITCZ shifted northward. GCMs perform better on large-scale fields (pattern correlations often exceeding 0.9 globally) than on , where wet biases over oceans and dry biases in subtropical lands yield lower skills, with RMSE values for annual means typically 20-50% higher relative to errors. Upper-air and circulation benchmarks highlight additional challenges; for example, GCMs underpredict observed stratospheric cooling trends post-1979, linked to , and exhibit errors in jet stream positions, such as weakened over the mid-latitudes. Ocean surface temperatures (SSTs) show good agreement in annual cycles but discrepancies in variability, with models like those in CMIP5 failing to capture the observed slowdown in rates during the early 2000s period. Sea ice extent validations indicate overestimation of summer minima in historical runs, while trends are mismatched, with models projecting decline against observed expansion until 2014. These benchmarks underscore that while GCMs capture first-order and energy balances, subgrid-scale processes like clouds and drive systematic errors, necessitating ongoing refinements.

Historical Hindcasting Performance

General circulation models (GCMs) undergo hindcasting by simulating historical conditions using observed forcings such as concentrations, solar variability, volcanic aerosols, and emissions, typically spanning the instrumental record from 1850 onward or the . These simulations are validated against empirical datasets like HadCRUT for surface temperatures or GPCP for to assess fidelity in reproducing observed trends, variability, and spatial patterns. Multi-model ensembles, such as those from the (CMIP), aggregate outputs from dozens of GCMs to evaluate collective performance, revealing improvements over successive phases (e.g., CMIP3 to CMIP5) in capturing large-scale features while highlighting persistent systematic errors. In global mean surface temperature (GMST), CMIP5 ensembles demonstrate high skill, with the multi-model mean aligning closely with observations over the —tracking within 0.5°C and achieving pattern correlations exceeding 0.95 for large-scale fields—and effectively reproducing post-1950 warming, volcanic cooling episodes, and interannual variability linked to phenomena like El Niño-Southern Oscillation (ENSO). Earlier individual GCMs from 1970–2007 also exhibit skillful hindcasts of multidecadal GMST trends, with an average skill score of 0.69 when adjusted for forcing uncertainties, showing no systematic over- or underestimation across 17 models evaluated against post-publication observations. CMIP6 models similarly capture decadal trends, such as 1901–1940 warming and 1941–1970 cooling, but display greater spread and overestimation of post-1998 warming in 90% of simulations, with the ensemble mean implying trends ~0.1–0.2°C per decade higher than observed rates of ~0.18°C per decade. Regional biases persist across phases, including cold anomalies of 1–2°C in zones (e.g., eastern Pacific), overestimation of seasonal cycle amplitude over land, and errors in high-topography areas like the and . Precipitation hindcasts show moderate skill at large scales, with CMIP5 pattern correlations improving to 0.82 (from 0.77 in CMIP3) for features like tropical maxima, subtropical dry zones, and monsoon systems, alongside better representation of intense extremes in higher-resolution variants. However, systematic errors include a spurious double (ITCZ), overestimation of tropical rainfall (e.g., western Pacific), underestimation of heavy sensitivity to warming, and regional discrepancies exceeding 20% for return values in many areas. CMIP6 simulations exacerbate some issues, such as overestimating historical trends relative to observations, which propagates into projections. Other variables reveal mixed results: ocean heat content changes from 1961–2005 fall within observational ranges, seasonal cycles match with <10% error (though Arctic summer decline is underestimated in ~75% of models), and top-of-atmosphere radiative fluxes align within 2.5 W/m² of satellite data. Cloud and moisture biases remain prominent, contributing to radiative imbalances of tens of W/m² regionally and dry lower-troposphere errors up to 25%, while tropical cyclone intensity is underestimated in standard resolutions. Overall, while GCM hindcasts robustly simulate global-scale historical evolution—bolstered by ensemble averaging—their regional and process-level fidelity is limited by parameterization uncertainties, with CMIP6 introducing hotter baselines that challenge attribution of recent trends.

Identified Biases and Systematic Errors

General circulation models (GCMs) exhibit persistent systematic biases in simulating the mean climate state, including a double intertropical convergence zone (ITCZ) bias characterized by excessive precipitation in the southern equatorial Pacific alongside the observed northern ITCZ, observed across CMIP3, CMIP5, and CMIP6 ensembles based on annual mean precipitation metrics. This error stems from deficiencies in representing ocean-atmosphere interactions and convection parameterization, leading to unrealistic equatorial asymmetry that propagates into variability simulations like ENSO. Models without this bias, such as those refined for better Pacific SST gradients, show improved tropical circulation fidelity. Temperature biases include a widespread cold bias in the equatorial Pacific sea surface temperatures (SSTs), with models producing excessively cool and narrow cold tongues due to overestimated trade winds and insufficient warming feedbacks. Polar regions often display a "cold pole" bias, where stratospheric temperatures are underestimated, linked to overly strong polar vortices and errors in gravity wave drag parameterization. CMIP5 models consistently underpredict surface air, ground, and soil temperatures globally, with deviations of several degrees in high-latitude continental areas. Precipitation simulations suffer from overfrequent light rain events, or "drizzling bias," arising from inadequate resolution of subgrid convective processes, which contributes to the double-ITCZ and excessive tropical wetness. Cloud-related errors include misrepresented diurnal cycles, with models failing to capture peak afternoon convection and instead producing nocturnal maxima, tied to boundary layer and microphysics shortcomings. Warm precipitation biases in midlatitudes reflect rapid droplet formation and fallout, exacerbating mean-state errors in vertical structure. These biases accumulate from approximations in unresolved physics, such as convection and clouds, and persist despite tuning to historical observations, indicating structural limitations rather than mere calibration issues. Downscaling amplifies them unless corrected, often resulting in overly wet and cold regional projections. Efforts like quantile mapping address statistical mismatches but do not resolve underlying causal errors in dynamics.

Applications in Prediction

Relation to Numerical Weather Prediction

General circulation models (GCMs) and numerical weather prediction (NWP) models share a foundational reliance on the numerical solution of the same primitive equations governing atmospheric dynamics, including the Navier-Stokes equations for momentum, thermodynamic equations for energy, and continuity equations for mass and water vapor. These shared physical principles enable both approaches to simulate fluid motions in the atmosphere, but they diverge in application: NWP focuses on deterministic short-term forecasts from hours to about two weeks by assimilating current observational data into high-resolution initial conditions, whereas GCMs emphasize statistical long-term climate behavior over months to centuries. Historically, GCMs emerged directly from advancements in NWP during the mid-20th century, as computational capabilities allowed extension of short-range weather simulation techniques to global, multi-year integrations. Early NWP efforts, pioneered by figures like Lewis Fry Richardson in the 1920s and realized computationally in the 1950s at institutions such as the Swedish Meteorological and Hydrological Institute and the U.S. Joint Numerical Weather Prediction Unit, provided the algorithmic basis for GCM development; by 1955, Joseph Smagorinsky at Princeton adapted NWP barotropic models into the first three-dimensional GCM, running simulations lasting days on early computers like the IBM 701. This evolution reflected a shift from initial-value problems in NWP—highly sensitive to precise starting states due to chaos in nonlinear dynamics—to boundary-value problems in GCMs, where external forcings like solar radiation and greenhouse gases drive ensemble-averaged outcomes insensitive to exact initial conditions after weeks. Key operational differences include spatial and temporal resolution: NWP models typically employ grids of 1–10 km horizontally with frequent data assimilation from satellites, radars, and surface stations to correct errors, enabling skillful predictions up to 10 days for mid-latitudes, whereas GCMs use coarser 50–200 km grids optimized for computational efficiency over decades, relying less on real-time observations and more on prescribed boundary conditions. Despite these distinctions, hybrid applications bridge the gap, such as seasonal-to-subseasonal forecasting models that extend NWP frameworks with GCM-like coupling to oceans and land, as seen in systems like the European Centre for Medium-Range Weather Forecasts' Integrated Forecasting System, which supports predictions from days to months. Modern unified modeling frameworks, exemplified by Germany's ICON model introduced in 2015, further blur boundaries by configuring the same core equations for both NWP (high-resolution, short-term) and climate simulations (lower-resolution, long-term).

Climate Projections and Scenario Modeling

General circulation models (GCMs) generate climate projections by simulating future atmospheric, oceanic, and land surface responses to prescribed radiative forcing scenarios, primarily through coordinated experiments under the Coupled Model Intercomparison Project (CMIP). In CMIP6, contributing GCMs from approximately 49 modeling groups produced runs under Shared Socioeconomic Pathways (SSPs) that incorporate varying greenhouse gas concentrations, aerosol levels, and land use changes, such as SSP1-2.6 for sustainable development with low emissions and SSP5-8.5 for fossil-fueled development with high emissions. These scenarios enable assessment of equilibrium climate sensitivity (ECS), estimated in CMIP6 models to range from 1.8°C to 5.6°C for doubled CO2, wider than the 2.1–4.7°C in prior phases due to inclusion of models with higher sensitivity. Projections from CMIP6 GCM ensembles, as synthesized in IPCC AR6, indicate global mean surface temperature (GMST) increases of 1.5°C (likely range 1.0–1.8°C) under SSP1-2.6 and 4.4°C (3.3–5.7°C) under SSP5-8.5 by 2081–2100 relative to 1850–1900, with scenario uncertainty dominating long-term projections alongside model structural differences. Regional patterns show amplified warming over land and polar regions, with precipitation increases in high latitudes and decreases in subtropical zones, though GCMs exhibit substantial spread in monsoon circulation changes influenced by internal variability like the (AMV). Extreme event projections, such as intensified heavy precipitation, rely on multi-model means to mitigate individual biases, yet uncertainties persist from unresolved cloud feedbacks and ocean heat uptake. Historical evaluations of GCM projections demonstrate skill in capturing observed global warming trends since the 1970s, with 10 of 17 models from 1970–2007 closely matching subsequent observations after publication, though adjustments for volcanic aerosols and internal variability improve alignment. However, CMIP5 models simulated surface warming about 16% faster than observations from 1970 onward, partly attributable to overestimated tropical tropospheric warming and aerosol effects, highlighting systematic hot biases in some ensembles. These discrepancies underscore the need for ongoing validation against empirical data, as GCM projections inform policy but remain subject to epistemic uncertainties in parameterizations of sub-grid processes like convection and ice-cloud interactions.

Role in Paleoclimate and Regional Studies

General circulation models (GCMs) facilitate paleoclimate investigations by integrating physical equations to simulate equilibrium climates under altered boundary conditions, such as reduced greenhouse gas concentrations, expanded continental ice sheets, and modified orbital forcings. These simulations test hypotheses about causal mechanisms driving past climate shifts, including the amplification of cooling via ice-albedo feedbacks during glacial periods. For the Last Glacial Maximum (LGM), dated to approximately 21,000 years before present, multi-model ensembles from the (PMIP) indicate a global mean surface temperature anomaly of -4.5°C to -6.5°C relative to pre-industrial levels, with greater cooling over landmasses (up to 10°C in mid-latitudes) and polar amplification exceeding 15°C in some cases. Proxy validations against ice core δ¹⁸O records and pollen assemblages confirm broad patterns like equatorward shifts in westerly jets and expanded subtropical aridity, though tropical sea surface temperature discrepancies highlight ongoing model-proxy tensions. GCM paleosimulations extend to interglacial periods, such as the Eemian (ca. 130,000–115,000 years ago), where increased Northern Hemisphere insolation drives simulated warming of 1–2°C globally, modulated by vegetation and ocean circulation feedbacks that alter regional moisture transport. By isolating forcings, these models quantify sensitivities, such as CO₂'s radiative role in deglaciation, supporting estimates of equilibrium climate sensitivity around 3°C per CO₂ doubling derived from LGM cooling patterns. Limitations arise from idealized boundary conditions and coarse resolution, which underrepresent mesoscale dynamics, prompting integration with proxy data assimilation for refined reconstructions. In regional studies, GCMs provide boundary forcings for downscaling techniques that enhance spatial detail for impact assessments, as global grids (typically 100–250 km) inadequately resolve orographic precipitation, land-sea breezes, and convective extremes influenced by local geography. Dynamical downscaling via nested (RCMs) at 10–50 km resolution refines GCM outputs, reducing biases in present-day climatologies by 20–50% for variables like summer precipitation in complex terrains. Applied to paleoregions, such methods simulate hydroclimate variability, revealing enhanced aridity in the American Southwest due to strengthened subtropical highs and reduced winter storm tracks. Statistical downscaling empirically maps GCM large-scale fields to local observations using transfer functions, assuming relative stability in statistical relationships, and has been validated for regional paleoprecipitation proxies like speleothem δ¹⁸O, though non-stationarities under forcings like ice sheet melt challenge long-term applicability. These approaches enable sector-specific analyses, such as agricultural vulnerability in Mediterranean-like paleoregions during pluvial events, by linking global forcings to localized responses in evapotranspiration and runoff. Overall, GCM-driven regional frameworks bridge scales, informing evidence-based projections while exposing gaps in representing unresolved processes like aerosol-cloud interactions.

Criticisms and Limitations

Issues with Model Tuning and Overfitting

General circulation models (GCMs) incorporate numerous uncertain parameters in subgrid-scale parameterizations for processes such as clouds, convection, and aerosols, which are adjusted during tuning to align simulated outputs with observational targets like global mean surface temperature trends and top-of-atmosphere radiation balance. This process, often manual and iterative, involves selecting parameter values that minimize discrepancies with historical data, typically spanning the instrumental record from the late 19th or 20th century. However, tuning introduces risks of overfitting, where models achieve spurious agreement with training data by compensating for structural deficiencies in physics representations rather than resolving underlying errors, potentially inflating apparent skill on tuned metrics while impairing generalization to novel conditions like future forcings or paleoclimates. The degrees of freedom in tuning—often exceeding 10-20 adjustable parameters per model—exacerbate overfitting concerns, as limited and noisy observational datasets allow multiple parameter combinations to fit global aggregates, masking regional or process-level biases. For instance, U.S. modeling centers report tuning to targets with structural uncertainties, such as cloud radiative effects, which can lead to over-reliance on imperfect data and reduced out-of-sample performance, as evidenced by persistent errors in tropical precipitation or stratospheric dynamics post-tuning. Critics highlight that this practice lacks standardization and transparency, with documentation often inadequate to distinguish legitimate calibration from data assimilation-like fitting, fostering circular validation where models are deemed skillful primarily against the data used for adjustment. Empirical assessments show that tuned frequently exhibit degraded hindcast fidelity for pre-industrial or glacial periods, suggesting overfitting to anthropogenic-era signals rather than robust physical emulation. Efforts to mitigate overfitting include objective methods like perturbed parameter ensembles or machine learning-assisted calibration, but these remain nascent and do not eliminate the fundamental challenge of high-dimensional parameter spaces relative to sparse, heterogeneous observations. Over-tuning can also entrench compensatory errors, such as inflating cloud feedback parameters to offset convection scheme flaws, leading to unreliable equilibrium climate sensitivity estimates that diverge from independent constraints like those from volcanic eruptions or satellite radiances. Consequently, while tuning enhances mean-state realism for control simulations, it compromises causal inference in attribution studies, as adjusted parameters may implicitly encode unobserved forcings or feedbacks without first-principles justification.

Discrepancies with Observations

General circulation models (GCMs) frequently exhibit systematic biases when compared to observational records, particularly in vertical temperature profiles, precipitation distributions, and cloud properties, which can amplify projected climate sensitivities beyond empirical evidence. For instance, in the tropical upper troposphere (200-300 hPa), GCMs predict amplified warming rates relative to the surface—termed the "hotspot"—driven by moist convective processes, yet radiosonde and satellite datasets such as those from the (UAH) show observed trends of approximately 0.09 K/decade from 1979-2014, compared to model ensemble means exceeding 0.15 K/decade over similar periods. This discrepancy persists across and ensembles, with high-equilibrium climate sensitivity (ECS > 4°C) models showing the largest inconsistencies against reanalyzed data like . Surface and tropospheric trends also reveal overestimation in many GCMs; CMIP6 simulations, evaluated against HadCRUT5 observations, display an warming of 0.24 K/decade from 1970-2020, surpassing the observed 0.18 K/decade, particularly in "hot" models with ECS above 4.5°C that contribute disproportionately to projections. These models often run warmer over land and mid-latitudes, with amplification ratios in CMIP6 averaging 3-4 times polar vs. since 1990, exceeding observed ratios closer to 2.5 in reanalysis products. patterns show analogous issues, including a "double intertropical convergence zone" (ITCZ) bias in the tropics, where models overestimate zonal rainfall by 1-2 mm/day compared to GPCP observations, linked to errors in ocean-atmosphere and convective parameterization. Extreme daily events exhibit wet biases of up to 14% in global land domains against ERA5 reanalysis, complicating regional projections. Cloud representation remains a core source of error, with GCMs underestimating low-cloud coverage over subtropical oceans by 10-20% relative to MODIS and observations, leading to overstated shortwave cloud radiative effects and positive feedbacks that inflate ECS estimates. Mixed-phase feedbacks in high latitudes show opposing errors—overly reflective clouds in some models reduce simulated warming, while others amplify it—resulting in net uncertainties of ±0.5 W/m²/K in feedback strength against satellite flux data. Ocean circulation discrepancies further compound these, as mid-tropospheric flows in models diverge from float and altimetry observations in intensity and direction, particularly in the where upwelling biases alter heat uptake by 0.1-0.2 PW. These persistent mismatches, evident across model generations despite tuning to historical data, underscore limitations in resolving sub-grid processes like and aerosols, with peer-reviewed evaluations indicating that no single GCM fully reconciles with all observational benchmarks.

Challenges from Natural Variability and Feedbacks

General circulation models (GCMs) face significant challenges in accurately simulating internal climate variability, such as oscillations driven by ocean-atmosphere interactions, which can obscure the detection and attribution of anthropogenic forcing. Modes like the El Niño-Southern Oscillation (ENSO), (PDO), and (AMO) exhibit amplitudes and teleconnections in observations that many GCMs fail to reproduce faithfully, often underestimating their strength or periodicity in control simulations without external forcing. For instance, Phase 5 (CMIP5) models largely lack robust internal multidecadal and bidecadal oscillations akin to the observed PDO and AMO, leading to inflated signal-to-noise ratios in projections where natural fluctuations are downplayed relative to forced trends. This underrepresentation contributes to biases in regional and patterns, as seen in persistent errors in North Atlantic winter variability across millennial simulations. Natural variability also exacerbates discrepancies between model ensembles and satellite observations, particularly in tropospheric warming rates, where multidecadal internal fluctuations can account for much of the observed-model over recent decades. GCMs' coarse and parameterized subgrid processes limit their ability to capture , nonlinear interactions underlying these modes, resulting in reduced ensemble that masks true uncertainty in decadal predictions. Consequently, projections may overestimate the emergence of forced signals in regions dominated by variability, such as the tropical Pacific, where ENSO modulation by decadal modes like the PDO remains poorly hindcast. Feedback mechanisms introduce further uncertainties, as GCMs rely on parameterizations for unresolved processes like cloud formation and phase transitions, leading to divergent estimates of (ECS) across models. Cloud feedbacks, in particular, dominate the intermodel spread, with low-altitude liquid clouds potentially providing stronger than simulated, while high-latitude mixed-phase clouds yield opposing signs in some configurations. and lapse-rate feedbacks amplify warming but are intertwined with schemes that exhibit systematic biases, such as overestimating moist static transport in the . Ice-albedo feedbacks are another source of asymmetric uncertainty, where models may miss nonlinear responses to sea ice retreat, including open-water formation and deposition effects not fully resolved at typical grid scales. In the , GCMs often produce unrealistic negative longwave feedbacks due to underestimated cloud and influences, contributing to excessive simulated cooling in polar amplification scenarios. These parameterization dependencies propagate into projections, where feedback uncertainties explain regional ECS variations, such as stronger ice-albedo effects at poles versus dominance in . Overall, the interplay between unresolved feedbacks and natural variability amplifies structural errors, underscoring the need for emergent constraints from observations to narrow parameter ranges without .

Comparisons with Alternative Models

Simplified Models like Radiative-Convective and EMICs

Radiative-convective models (RCMs) represent a class of one-dimensional simplified models that compute vertical profiles of , , and radiative fluxes in a single atmospheric column, assuming hydrostatic balance and moist convective adjustment to prevent superadiabatic . These models balance incoming solar radiation with , incorporating as a parameterization to redistribute heat vertically, as developed in the seminal work by Manabe and Wetherald in 1967, which demonstrated a global surface warming of approximately 2.3 K for a doubling of atmospheric CO2 concentration due to and feedbacks. RCMs serve as foundational tools for understanding equilibrium climate sensitivity (ECS), typically yielding estimates between 2.0 and 3.0 K per CO2 doubling, depending on assumptions about relative and effects, but they exclude horizontal , , and land-atmosphere interactions present in full general circulation models (GCMs). Compared to GCMs, RCMs offer significant computational efficiency, enabling rapid exploration of radiative-convective processes and sensitivity to parameters like aerosol optical depth or concentrations without the need for three-dimensional grid resolutions. Their simplicity allows isolation of key s, such as the positive amplifying warming by 50-100% over pure radiative models, providing a for validating GCM parameterizations of subgrid-scale . However, limitations include an inability to capture zonal asymmetries, tracks, or teleconnections, leading to overestimation of tropical uniformity and underrepresentation of dynamical cooling effects that moderate in GCMs. Earth system models of intermediate complexity (EMICs) extend beyond RCMs by incorporating reduced-form representations of multiple Earth system components, such as zonally or hemispherically averaged ocean circulation, simplified atmospheric dynamics, and biogeochemical cycles, while maintaining computational costs orders of magnitude lower than GCMs—for instance, the JUMP-LCM EMIC executes 63,000 times faster than the high-resolution GCM MIROC4h. EMICs often employ statistical-dynamical approaches, like diffusive closures for and transport, to emulate large-scale circulation without resolving eddies, enabling simulations over millennial timescales to study phenomena like evolution or feedbacks under paleoclimate forcings. In intercomparisons, EMICs reproduce GCM-like global mean and responses to CO2 forcing, with ensemble spreads comparable to those in comprehensive models, though regional patterns exhibit greater divergence due to simplified and . Relative to GCMs, EMICs facilitate through large ensembles and perturbed physics experiments, as demonstrated in studies using models like LOVECLIM to assess sensitivity to ocean parameters, revealing that EMIC ECS ranges (around 2-4 ) align with GCM multimodel means but with reduced from unresolved mesoscale processes. Their advantages include tractability for exploring long-term feedbacks, such as permafrost carbon release or heat uptake, which GCMs struggle with due to high resource demands, but they sacrifice fidelity in simulating transient variability, extreme events, and fine-scale features like El Niño-Southern Oscillation. EMICs thus complement GCMs by providing efficient scoping tools and theoretical insights, though their reliance on tuned can introduce systematic errors in meridional energy transport, underscoring the need for GCM validation of intermediate approximations.

Comprehensive Earth System Models

Comprehensive Earth System Models (ESMs) integrate simulations of the atmosphere, oceans, land surface, , and , incorporating physical, chemical, and biological processes to capture interactions across the system. These models extend beyond atmosphere-ocean general circulation models (AOGCMs) by including dynamic representations of biogeochemical cycles, such as carbon, , and aerosols, which enable the simulation of feedbacks like responses to and their effects on atmospheric . ESMs typically operate on global grids with resolutions ranging from 25 to 100 km horizontally and multiple vertical levels, relying on coupled component models linked via flux exchanges for , , , and tracers. Core components encompass atmospheric dynamics via modules like the Community Atmosphere Model (CAM), oceanic circulation through models such as the Parallel Ocean Program version 2 (POP2), land surface processes including hydrology and vegetation dynamics in the Community Land Model (CLM), sea ice evolution with the CICE model, and biogeochemical modules for terrestrial and marine ecosystems. For instance, ESMs simulate nutrient-limited in oceans and soils, aerosol-cloud interactions, and from wetlands, which influence and . Equilibrium in ESMs varies but has increased in recent generations; CESM2, for example, yields 5.1–5.3 °C, attributed to refinements in cloud microphysics and land carbon feedbacks. Prominent ESMs include the Community Earth System Model version 2 (CESM2), released in 2020 by the , which supports simulations from paleoclimate to future scenarios and has been used in CMIP6 intercomparisons. Other examples are NOAA's Laboratory (GFDL) ESMs, such as ESM4, which emphasize high-resolution ocean , and the European Centre for Medium-Range Weather Forecasts' Integrated Forecasting System extended to Earth system components. These models demand , with CESM2 runs requiring thousands of processor hours for century-scale simulations on platforms like supercomputers. In comparison to atmospheric-only GCMs or simplified energy balance models, ESMs provide more realistic projections of transient responses by accounting for slow feedbacks like thaw releasing gases, though their parameterizations for sub-grid processes introduce uncertainties that require empirical tuning against observations. Validation involves hindcasts against data, paleoproxies, and in-situ measurements, revealing strengths in large-scale circulation but gaps in regional extremes and biosphere-atmosphere . Despite computational costs, ESMs underpin IPCC projections, with advancements like CESM2 improving historical biases compared to predecessors.

Emerging Machine Learning and Neural GCMs

Neural general circulation models (Neural GCMs) represent a hybrid paradigm that integrates machine learning components, particularly neural networks, into the framework of traditional physics-based GCMs to enhance simulation efficiency and accuracy. These models retain a differentiable solver for large-scale atmospheric dynamics while replacing or augmenting subgrid-scale parameterizations—such as convection, clouds, and turbulence—with data-driven neural networks trained on reanalysis datasets like ERA5. This approach addresses longstanding challenges in traditional GCMs, where hand-tuned parameterizations often introduce biases and computational bottlenecks. A prominent example is NeuralGCM, developed by Research in collaboration with ECMWF and , which achieves state-of-the-art performance in medium-range up to 10 days while simulating realistic variability, including phenomena like El Niño-Southern (ENSO). NeuralGCM outperforms operational physics-based models like ECMWF's IFS in global forecast skill metrics and surpasses pure emulators, such as GraphCast, in physical consistency due to its hybrid structure. The model's differentiability enables end-to-end optimization, allowing joint training of dynamics and parameterizations to minimize errors against observations, which reduces computational costs by factors of 10-100 compared to fully numerical GCMs for equivalent resolutions. Advantages of Neural GCMs include accelerated simulations suitable for ensemble predictions and experiments, as demonstrated by NeuralGCM's ability to generate multi-decadal runs with equilibrium climate states aligning closely with observations in metrics like global energy balance and tropical precipitation patterns. For instance, in 2024 applications, variants of NeuralGCM improved forecasting accuracy for Indian agriculture by incorporating regional data biases. However, limitations persist: reliance on historical training data risks poor extrapolation to novel climates, such as high-emission scenarios beyond 2100, and neural components can amplify uncertainties in extreme events if not constrained by physics. Unlike traditional GCMs, which derive from first-principles equations, Neural GCMs exhibit reduced interpretability, necessitating rigorous validation against independent datasets to mitigate . Ongoing developments, including extensions for radiance as of December 2024, aim to enhance constraint and realism.

Historical Evolution

Pioneering Efforts (1950s-1960s)

The foundations of general circulation models (GCMs) emerged from early efforts in (NWP) during the 1950s, building on the vision of simulating atmospheric dynamics using computational methods. In 1950, Jule Charney led a team that performed the first successful numerical weather forecasts using the computer, applying barotropic equations to predict large-scale flow patterns over 24 hours, which demonstrated the feasibility of integrating the of motion for short-term predictions. These experiments, conducted at the Institute for Advanced Study under John von Neumann's influence, marked a shift from manual graphical methods to digital computation, though limited to simplified, two-dimensional models without full three-dimensional circulation. A pivotal advance came in 1956 with Norman Phillips' development of the first rudimentary GCM, a two-level quasi-geostrophic model simulating hemispheric on the IAS computer. Starting from an initial state of relative rest, Phillips' numerical experiment integrated the model over extended periods, producing realistic features such as mid-latitude , jet streams, and zonal-mean meridional circulations driven by differential solar heating, thereby validating the potential for computers to replicate observed general circulation patterns without external forcing beyond . This work, detailed in Phillips' paper "The General Circulation of the Atmosphere: A Numerical Experiment," highlighted the roles of synoptic eddies and mean flows in maintaining thermal balance, though the model's coarse resolution (about 1,000 km grid spacing) and lack of or constrained its realism. In parallel, Joseph Smagorinsky established the General Circulation Research Section at the U.S. Weather Bureau in 1955, initiating systematic development of three-dimensional primitive equation models aimed at global simulations. Smagorinsky's group advanced beyond quasi-geostrophic approximations by incorporating full hydrostatic and primitive equations, achieving early integrations of baroclinic global atmospheres by the late 1950s, which included realistic pressure gradients and vertical structure but struggled with computational instability and required manual adjustments for long-term stability. These efforts culminated in the first operational three-level global models by the early 1960s, setting the stage for coupled atmosphere-ocean representations, though initial runs revealed challenges in resolving small-scale processes like convection without excessive diffusion. Independent work, such as Cecil Leith's primitive equation model at Lawrence Livermore National Laboratory around 1960-1965, further explored energy-conserving formulations for sustained simulations, emphasizing spectral methods to mitigate grid-scale errors. Collectively, these pioneering models underscored the computational barriers of the era—limited to hours of integration on vacuum-tube machines—yet proved that first-principles fluid dynamics could yield emergent circulation resembling observations.

Expansion and Refinement (1970s-1990s)

During the 1970s, atmospheric general circulation models (AGCMs) advanced through increased vertical resolution and incorporation of physical processes such as the hydrologic cycle and , enabling more realistic simulations of global climate dynamics. and colleagues at the Geophysical Fluid Dynamics Laboratory (GFDL) published a seminal 1970 study using a nine-level AGCM that included moist convection, cloud formation, and interactions between , , and in radiation calculations, producing equilibrium climates comparable to observations. This model demonstrated seasonal variations in and temperature, highlighting the role of land-sea contrasts in driving circulation patterns. Concurrently, the UK developed its first AGCM in 1972, employing grid-point methods to simulate tropospheric flows with improved parameterizations. The 1979 Charney Report, commissioned by the , endorsed GCMs as reliable tools for predicting greenhouse gas-induced warming, estimating a 1.5–4.5°C global temperature rise for doubled CO₂ based on early model ensembles, though it noted uncertainties in cloud feedbacks. By the late , spectral transform methods emerged as a dominant numerical approach, allowing efficient handling of and reducing computational costs for higher resolutions, as implemented in models like the NCAR Community Climate Model (CCM). In the 1980s, efforts shifted toward coupled atmosphere-ocean GCMs (AOGCMs) to capture air-sea interactions without artificial flux corrections, though initial versions suffered from climate drift requiring adjustments. GFDL's R15-resolution AOGCM, coupled in 1985, simulated El Niño-Southern Oscillation (ENSO) variability but exhibited systematic errors in tropical mean states. James Hansen's GISS model II, updated in 1988, incorporated historical forcings and sulfate aerosols, projecting 0.7–1.3°C warming by 2019 under business-as-usual scenarios, with emphasis on stratospheric cooling and . The 1990s saw refinements in model resolution, parameterizations for and surface processes, and ensemble simulations for the inaugural IPCC assessment, where nine AGCMs and early AOGCMs provided equilibrium estimates averaging 2.5°C for doubled CO₂. HadCM3, released by the UK Met Office in 1998, featured a 2.75°×3.75° atmosphere and 1.25° , demonstrating stable coupled behavior without flux adjustments and improved ENSO hindcasts. These advances enabled projections of regional patterns, such as enhanced warming over and precipitation shifts, though discrepancies persisted in simulating observed cooling.

Contemporary Advances (2000s-2025)

The 2000s marked a shift toward fully coupled atmosphere-ocean general circulation models (AOGCMs) as the standard for long-term climate simulations, with improvements in representing ocean-atmosphere interactions and initial integrations of biogeochemical cycles in Earth system models (ESMs). The Climate Model Intercomparison Project Phase 3 (CMIP3), launched in 2005, enabled multi-model ensembles that quantified uncertainties in projections for the IPCC Fourth Assessment Report, demonstrating better skill in simulating historical temperature trends compared to prior phases. Physical parameterizations advanced, particularly for convection and large-scale dynamics, though persistent biases in cloud feedbacks remained evident across models. In the 2010s, CMIP5 models featured atmospheric resolutions typically around 100-250 km horizontally, incorporating refined representations of aerosols, land surface processes, and dynamics, which enhanced simulations of phenomena like El Niño-Southern Oscillation (ENSO) variability. These developments supported Representative Concentration Pathways (RCPs) for scenario-based forecasting, with ensembles showing reduced spread in global mean temperature projections relative to CMIP3. Computational advances, including on supercomputers, allowed for longer integrations and higher ensemble sizes, improving statistical robustness despite ongoing challenges in tropical precipitation biases. The 2020s brought CMIP6, with participating models exhibiting higher average spatial resolutions—often 50-100 km horizontally—and enhanced parameterizations for cloud-aerosol interactions and stratospheric processes, leading to better alignment with observed precipitation patterns in some regions. Shared Socioeconomic Pathways (SSPs) expanded scenario diversity, while incremental gains in simulating extreme events were noted, though equilibrium climate sensitivity ranges widened compared to CMIP5 due to diverse model physics. By 2025, experimental high-resolution simulations reached 9 km global grids, enabling finer depiction of regional variability and extreme weather, facilitated by exascale computing capabilities. These advances, while improving overall fidelity, have been characterized as evolutionary rather than transformative, with persistent structural uncertainties in unresolved subgrid processes.

References

  1. [1]
    What is a GCM? - IPCC Data Distribution Centre
    General Circulation Models or GCMs, representing physical processes in the atmosphere, ocean, cryosphere and land surface, are the most advanced tools ...
  2. [2]
    General Circulation Models | METEO 469 - Dutton Institute
    GCMs attempt to describe the full three-dimensional geometry of the atmosphere and other components of Earth's climate system.
  3. [3]
    Stable climate simulations using a realistic general circulation model ...
    May 16, 2022 · Stable climate simulations using a realistic general circulation model with neural network parameterizations for atmospheric moist physics and ...
  4. [4]
    Clarifying the Dynamics of the General Circulation: Phillips's 1956 ...
    Phillips adopted a set of dynamical constraints not unlike those used in short-range forecasting of the large-scale tropospheric flow—a two-level quasi-.
  5. [5]
    General Circulation Models of the Atmosphere
    Smagorinsky's goal was the one first envisaged by von Neumann and Charney: a general circulation model ... ocean-atmosphere GCM that had a roughly Earth-like ...
  6. [6]
    (PDF) Coupled Atmosphere-Ocean Models in Climate Science
    Nov 19, 2019 · The atmospheric general circulation model (AGCM) was the most commonly used climate model up until around mid-1980s when coupled atmosphere- ...
  7. [7]
    Are general circulation models obsolete? - PNAS
    Nov 14, 2022 · The general circulation model, or GCM, is a mainstay of research into the evolving state of the Earth system over a range of timescales. The ...
  8. [8]
    On the limitations of general circulation climate models - AGU Journals
    We also demonstrate that GCM simulations of the large scale transports of heat are sensitive to the (uncertain) subgrid scale parameterizations. This leads ...
  9. [9]
    [PDF] Climate Models An Assessment of Strengths and Limitations
    cific Ocean general circulation model. J. Climate, 14, 1377–1398. Li, Z.X., 1999: Ensemble atmospheric GCM simulation climate inter- annual variability from ...
  10. [10]
    Global climate models performance: A comprehensive review of ...
    GCMs are typically used to assess climate change, but their simulations come with significant uncertainties due to factors like model structure, ...
  11. [11]
    [PDF] General circulation models cannot predict climate
    Apr 19, 2024 · Abstract. This study draws on Chaos Theory to investigate the ability of a General Circulation Model to predict climate. The.<|separator|>
  12. [12]
    Glossary of Climate-Related Terms - Physical Sciences Laboratory
    GCM. General Circulation Model. A numerical climate model that tries to incorporate as much of the atmosphere and ocean processes as possible. Used for both ...
  13. [13]
    Climate Modeling - Geophysical Fluid Dynamics Laboratory - NOAA
    Models investigate the degree to which observed climate changes may be due to natural variability, human activity, or a combination of both.
  14. [14]
    Neural general circulation models for weather and climate - PMC
    Jul 22, 2024 · General circulation models (GCMs) are the foundation of weather and climate prediction. GCMs are physics-based simulators that combine a ...<|separator|>
  15. [15]
    Atmospheric Modeling at NYU
    The primitive equations are a simplification of the Navier-Stokes equations of fluid motion, which are appropriate for the large scale circulation of the ...<|separator|>
  16. [16]
    [PDF] 2 The equations governing atmospheric flow. - Staff
    2.5 Summary of the primitive equations. The above equations that we have derived for atmospheric flow are often known as the primitive equations. The ...
  17. [17]
    A New Class of Dynamical Cores - AMS Journals
    Spectral models often solve the vorticity–divergence form of the primitive equations, as opposed to solving for the vector components of the momentum equation.
  18. [18]
    [PDF] Numerically Converged Solutions of the Global Primitive Equations ...
    To the best of our knowledge, this is the first set of published solutions of the time-dependent, atmospheric, primitive equations in spherical coordinates ...
  19. [19]
    [PDF] Ocean Modelling - the NOAA Institutional Repository
    A primitive-equations model in an idealised double-gyre configuration at eddy-resolving horizontal resolution is used to diagnose the relationship between the ...
  20. [20]
  21. [21]
    [PDF] Grids in Numerical Weather and Climate Models - IntechOpen
    May 16, 2013 · Although latitude-longitude grid is the most common, it suffers from the polar problem. This problem can be overcome by using a triangular or ...
  22. [22]
    [PDF] Latitude-Longitude Grid Suitable for Numerical Time Integration of a ...
    coordinates at two widely separated longitudes can pro- duce serious error near the poles because of the large difference in the directions of the unit ...
  23. [23]
    [PDF] Numerical methods for Earth system modelling - ICTP
    calculated exactly using the quadrature indicated. Page 40. Spectral transform method: Gaussian grids. ▻ The Gaussian grids are defined by the quadrature ...
  24. [24]
    [PDF] comparison of hexagonal-icosahedral and cubed sphere grids - GMD
    May 20, 2014 · The algo- rithm is implemented and tested on two families of grids: hexagonal–icosahedral Voronoi grids, and modified equian- gular cubed- ...
  25. [25]
    Application of Local Discretization Methods in the NASA Finite ...
    Jul 17, 2002 · We present the basic ideas of the dynamics system of the finite-volume General Circulation Model developed at NASA Goddard Space Flight ...
  26. [26]
    [PDF] non-standard grids - ECMWF
    2.1 Early applications. Both the icosahedral and cubic grids have been used, from time to time, in global numerical mod- els of the atmosphere. Sadourny et ...
  27. [27]
    Stable machine-learning parameterization of subgrid processes for ...
    Jul 3, 2020 · Smaller-scale processes, such as convection, are represented by subgrid parameterization schemes that typically rely on heuristic arguments.
  28. [28]
    [PDF] The mass-flux approach to the parametrization of deep convection ...
    The mass-flux approach, formulated by Ooyama, is used to parametrize convection in large-scale models, providing a stronger physical basis than simpler schemes.
  29. [29]
    [PDF] Parameterization of convection: The basics
    Nowadays, most GCMs use a parameterization of convection based on the mass-flux approach which tends to explicitly represent the underlying physical processes, ...
  30. [30]
    A practical approach to scale-adaptive deep convection in a GCM by ...
    Jun 11, 2018 · The cumulus base mass flux is used as a control parameter to decrease the sub-grid scale vertical transport in the GCM. In the mass flux type ...
  31. [31]
    Higher-Order Turbulence Closure and Its Impact on Climate ...
    This paper describes climate simulations of the Community Atmosphere Model, version 5 (CAM5), coupled with a higher-order turbulence closure known as Cloud ...
  32. [32]
    An Updated CLUBB PDF Closure Scheme to Improve Low Cloud ...
    Dec 2, 2022 · The updated closure scheme is found to improve the low cloud simulation by simulating more symmetric turbulence in stratocumulus and stronger ...
  33. [33]
    Cloud Simulations in Response to Turbulence Parameterizations in ...
    The response of cloud simulations to turbulence parameterizations is studied systematically using the GISS general circulation model (GCM) E2.
  34. [34]
    [PDF] Modeling the General Circulation of the Atmosphere. Topic 1
    1966: US uses 3-level primitive equation model. — Global coverage since 1973. Page 46. Numerical Methods. — Gridpoint methods: ○ Fields ...
  35. [35]
    [PDF] A general circulation model of the atmosphere using the full ...
    The Galerkin method is another way to represent the spatial structures of atmospheric variables and has become more popularly adopted for GCMs. This method uses ...<|separator|>
  36. [36]
    Enforcing conservation of axial angular momentum in the ... - GMD
    Feb 21, 2020 · Numerical general circulation models of the atmosphere are generally required to conserve mass and energy for their application to climate ...
  37. [37]
    A New Mass Flux Correction Procedure for Vertically Integrated ...
    Nov 10, 2020 · The present study proposes a new method of mass flux correction by constraining the mass, energy, and water budgets that can be applied to meridional energy ...
  38. [38]
    (PDF) Emerging Numerical Methods for Atmospheric Modeling
    Aug 6, 2025 · ... methods. It employs discontinuous elements and flux integrals along their boundaries, ensuring local flux conservation. However, as opposed ...
  39. [39]
    GEOS Systems - NASA GMAO
    Aug 25, 2025 · The Atmospheric General Circulation Model (AGCM): this configuration uses the predictive model components for the atmosphere and land, and is ...
  40. [40]
    [PDF] The GEOS-5 Atmospheric General Circulation Model: Mean Climate ...
    Apr 28, 2012 · The GEOS-5 AGCM presented here is the model used for the GMAO's atmosphere-only and coupled CMIP-5 simulations. The seasonal mean climate of the ...
  41. [41]
    CGCM and AGCM Seasonal Climate Predictions -- A study in CCSM4
    ... (AGCM) ensemble predictions that are forced by time‐varying sea surface temperatures (SSTs). CGCMs and AGCMs have differences in the way surface energy ...
  42. [42]
    [PDF] Reproducible Forced Modes in AGCM Ensemble Integrations and ...
    AGCM's capability to forecast seasonal variations. ... Kirk,. 1990: Simulation of ENSO related surface wind anomalies with an atmospheric GCM forced by observed ...
  43. [43]
    The GEOS-5 Atmospheric General Circulation Model: Mean Climate ...
    Apr 30, 2012 · This report is a documentation of the Fortuna version of the GEOS-5 Atmospheric General Circulation Model (AGCM).
  44. [44]
    Atmosphere-only GCM (ACCESS1.0) simulations with prescribed ...
    One such example is to prescribe sea surface temperatures (SSTs) such that 70 % of the Earth's surface temperature field is observationally constrained (known ...
  45. [45]
    [PDF] Experimental ENSO predictions by the UCLA atmospheric GCM ...
    Experimental ENSO predictions by the UCLA atmospheric GCM coupled to the MIT and ... SST distribution for December by the UCLA AGCM coupled to POP (left) and ...
  46. [46]
    Comparison of cloud fields from atmospheric general circulation ...
    This paper focuses on the comparison of cloud amounts derived from an atmospheric general circulation model (AGCM), Satellite-observed clouds, ...
  47. [47]
    [PDF] General Circulation Models - Florida State University
    A general circulation model (GCM) of the ocean is nothing more than that—a numerical model that represents the movement of water in the ocean.
  48. [48]
    [PDF] Chapter 1 OGCMs and MRI.COM
    ... equation of state of sea water, and the mass conservation equation, collectively called primitive equations (Chapter 2). If necessary, equations for ...
  49. [49]
    Primitive Equation - an overview | ScienceDirect Topics
    The primitive equations are based on the so-called hydrostatic approximation, in which the conservation of momentum in the vertical direction is replaced by the ...<|control11|><|separator|>
  50. [50]
    [PDF] OPA 8.1 Ocean General Circulation Model Reference Manual
    multitasked general circulation model of the ocean. In. Science and engineering on Cray Supercomputers,. Proceedings of the 4th International Symposium.
  51. [51]
    Modular Ocean Model (MOM)
    The Modular Ocean Model (MOM) is a numerical representation of the ocean fluid with applications from the process scale to the planetary circulation scale.
  52. [52]
    [PDF] Elements of the Modular Ocean Model (MOM)
    The Modular Ocean Model (MOM) is a hydrostatic generalized level coordinate numer- ical ocean code with mass conserving non-Boussinesq or volume conserving ...
  53. [53]
    General Circulation Model (GCM) Modeling Overview – US KESS
    The Parallel Ocean Program (POP) is a three-dimensional, z-level, primitive equation general circulation ocean model with an implicit free surface.
  54. [54]
    [PDF] NEMO ocean engine - UCSD
    The ocean engine of NEMO (Nucleus for European Modelling of the Ocean) is a pri- mitive equation model adapted to regional and global ocean circulation ...
  55. [55]
    History and Methodology of Modelling the Circulation of the World ...
    This paper recounts the historical development of numerical models of the ... -J., 1984: A numerical world ocean general circulation model. Dyn. Atmos ...<|separator|>
  56. [56]
    15.6: Coupled Ocean and Atmosphere Models
    Nov 11, 2024 · Coupled numerical models of the atmosphere and ocean are used to study the climate, its variability, and its response to external forcing.
  57. [57]
    1.5.3 Coupled Models: Evolution, Use, Assessment
    The first attempts at coupling atmospheric and oceanic models were carried out during the late 1960s and early 1970s (Manabe and Bryan, 1969; Bryan et al., 1975 ...
  58. [58]
    Development of global coupled ocean-atmosphere general ...
    Beginning in the late 1960s and continuing through the early 1980s, general circulation models (GCMs) of the atmosphere, ocean, and sea ice were coupled and run ...
  59. [59]
    A Caveat Note on Tuning in the Development of Coupled Climate ...
    Nov 24, 2017 · Flux correction represents a useful alternative: Flux correction does not alter the model physics and therefore no optimization is required.Missing: challenges | Show results with:challenges
  60. [60]
    [PDF] The BRIDGE HadCM3 family of climate models - GMD
    This model is a three-dimensional, fully dynamic, coupled atmosphere–ocean global climate model without flux adjust- ment. Our version of the model is very ...
  61. [61]
    [PDF] GFDL's CM3 Coupled Climate Model: Characteristics of the Ocean ...
    The purpose of this paper is to document elements of the ocean and sea ice simulations in a new coupled climate model developed at the Geophysical Fluid ...Missing: GCMs HadCM3 CMIP
  62. [62]
    CMIP - History - PCMDI - Lawrence Livermore National Laboratory
    In recent years, coupled GCMs have also been used to separate natural variability from anthropogenic effects in the climate record of the 20th century, and to ...
  63. [63]
    Coupled Ocean‐Atmosphere Modeling Over the Maritime Continent ...
    May 12, 2020 · The Princeton Ocean Model (POM) is a community general numerical model for ocean circulation that can be used to simulate and predict oceanic ...Introduction · Two-Way Coupled... · Uncertainty in Models · Conclusions and...
  64. [64]
    Advances in atmospheric, oceanic, and coupled models for ...
    Oct 10, 2025 · Ocean models such as Nucleus of European Modeling of the Ocean (NEMO) and Hybrid Coordinate Ocean Model (HYCOM) are assessed for their ...
  65. [65]
    Coupled Atmosphere–Ocean Reconstruction of the Last Millennium ...
    Mar 18, 2021 · We present an illustrative investigation of the climate periods known as the Medieval Climate Anomaly (MCA) and Little Ice Age (LIA), and ...
  66. [66]
    Why are Climate models written in programming languages from ...
    Feb 21, 2021 · Climate models try to simulate the world. The simulations usually start at some preindustrial control state (1850~1900) and iteratively step forward in time.
  67. [67]
  68. [68]
    What coding languages do you use for writing climate models?
    Jul 5, 2018 · I would suggest any of the following, MATLAB, C++, Python. You might even be able to get away with FORTRAN or even something as mundane as Java.Which programming language is more suitable for large scale ...What software and programming languages do atmospheric ... - QuoraMore results from www.quora.com
  69. [69]
    Every weather prediction model, and every climate model, is in ...
    Nov 22, 2023 · There are a few other active projects porting weather, ocean, and climate models to Julia, C++, or Python.
  70. [70]
    CSM Flux Coupler (CPL) - Community Earth System Model
    CSM software is based on a framework which divides the complete climate system into component models connected by a flux coupler.
  71. [71]
    The Community Earth System Model: A Framework for Collaborative ...
    The Community Earth System Model (CESM) is a flexible and extensible community tool used to investigate a diverse set of Earth system interactions across ...
  72. [72]
    Software - Climate Dynamics Group
    GCMs: General Circulation Models. We use a variety of GCMs for our research, from dry atmosphere models to relatively complex coupled atmosphere-ocean models.
  73. [73]
    [PDF] The OASIS3 coupler: a European climate modelling community ...
    Abstract. This paper presents the OASIS3 coupling soft- ware used in five of the seven European Earth System Mod- els (ESMs) participating to the Fifth ...
  74. [74]
    Time Step Sensitivity and Accelerated Spinup of an Ocean GCM ...
    Abstract. The distorted physics (DP) technique of Bryan and Lewis is applied to an ocean GCM to allow longer time steps to be taken.
  75. [75]
    Fast, accurate climate modeling with NeuralGCM - Google Research
    Jul 22, 2024 · NeuralGCM presents a new approach to building climate models that could be faster, less computationally costly, and more accurate than existing models.
  76. [76]
    Neural general circulation models for weather and climate - Nature
    Jul 22, 2024 · GCMs are physics-based simulators that combine a numerical solver for large-scale dynamics with tuned representations for small-scale processes ...
  77. [77]
    The computational and energy cost of simulation and storage ... - GMD
    Apr 19, 2024 · This paper presents the main findings obtained from the CPMIP (the Computational Performance Model Intercomparison Project), a collection of a common set of ...Missing: GCM | Show results with:GCM
  78. [78]
    Supercomputers can take months to simulate the climate – but my ...
    May 1, 2024 · Not surprisingly, these models are expensive. The simulations take time, frequently several months, and the supercomputers on which the models ...
  79. [79]
    The Parallel Scalability of the Spectral Transform Method in
    The spectral transform method is a natural candidate for general circulation models (GCMs) designed to run on large-scale parallel computers due to the large ...
  80. [80]
    A scalable parallel algorithm for atmospheric general circulation ...
    We propose a scalable parallel algorithm with two-dimension domain decomposition for atmospheric general circulation models.
  81. [81]
    Is it possible to run a general circulation model (GCM) of the ... - Quora
    Jul 21, 2020 · So, if you are executing a certain model on your CPU, the same can be simulated with a minimum of 2x speedup GPUs.
  82. [82]
    Evaluation of CMIP6 HighResMIP models in simulating precipitation ...
    The HighResMIP simulations incorporate the model resolution (grid spacing) that range from typical CMIP6 resolutions (∼250 km in the atmosphere and 100 km in ...
  83. [83]
    CMIP6 projections at high-resolution - the climate data factory
    Sep 16, 2021 · Typical horizontal resolutions range from 1° to 2.5° with some high-resolution models achieving finer grids of 25 to 50 km in the HighResMIP ...
  84. [84]
    Optimizing climate models with process knowledge, resolution, and ...
    Jun 19, 2024 · We propose a balanced approach that leverages the strengths of traditional process-based parameterizations and contemporary artificial intelligence (AI)-based ...
  85. [85]
    Q&A: How do climate models work? - Carbon Brief
    Jan 15, 2018 · In general, increasing the spatial resolution of a model by a factor of two will require around 10 times the computing power to run in the same ...Missing: runtime | Show results with:runtime<|separator|>
  86. [86]
    What Are the Limitations of High-Resolution Climate Models?
    Apr 1, 2025 · High-resolution models enhance detail but face limits in computation, parameterization, data, and inherent climate system uncertainty.Missing: general | Show results with:general
  87. [87]
    Comparison of Global Mesoscale Convective System Simulations in ...
    Coarser resolution (>100 km) general circulation models (GCMs) struggle to simulate these processes due to scale-separation assumptions in convection ...<|separator|>
  88. [88]
    A seamless approach for evaluating climate models across spatial ...
    Oct 5, 2023 · We developed a new framework to quantify the added value of high spatial resolution in simulating precipitation over the contiguous United States (CONUS).
  89. [89]
    Atmosphere-Ocean General Circulation Model - NASA GMAO
    Aug 20, 2025 · The GEOS-5 Atmosphere-Ocean General Circulation Model (AOGCM) has been developed to simulate climate variability on a wide range of time scales ...
  90. [90]
    Regional validation of means, variances, and spatial patterns in ...
    Jan 20, 1990 · The focus of this study is the control run performance of four general circulation models (GCMs): the Oregon State University (OSU) ...
  91. [91]
    On the discrepancy between observed and CMIP5 multi-model ...
    Apr 12, 2017 · This study aims to understand the relative roles of external forcing versus internal climate variability in causing the observed Barents Sea ...Missing: CMIP | Show results with:CMIP
  92. [92]
    The Double‐ITCZ Bias in CMIP3, CMIP5, and CMIP6 Models Based ...
    Jun 16, 2022 · We find that the double‐ITCZ bias with a big inter‐model spread persists in all CMIP models and still remains a serious problem in the latest ...Missing: discrepancies | Show results with:discrepancies
  93. [93]
    Evaluation of global climate models for the simulation of ...
    May 9, 2023 · The performance of GCMs in daily minimum and maximum temperatures shows better than that in daily precipitation, where GCMs show a difference in ...
  94. [94]
    Comparison of General Circulation Model and Observed Regional ...
    Here we discuss two approaches to compare output of individual GCM grid boxes with local station observations near the surface and in the free troposphere. The ...Missing: observational | Show results with:observational
  95. [95]
    On the discrepancy between observed and CMIP5 multi-model ...
    Apr 12, 2017 · The work presented here aims to provide insights on these questions by comparing observations with the CMIP5 multi-model ensemble simulations ...
  96. [96]
    Comparisons of Coupled Model Intercomparison Project Phase 5 ...
    Jul 8, 2022 · In this report, we compare sea-ice projections by CMIP6 models to those of CMIP5 models in each of four polar bear ecoregions over the 21st century.
  97. [97]
    [PDF] Evaluation of Climate Models
    Regional climate models, driving general circulation models and interannual ... dynamic global vegetation models under climate change. New Phytologist, 187 ...
  98. [98]
    Evaluating the Performance of Past Climate Model Projections
    Dec 4, 2019 · While models are commonly evaluated by comparing “hindcasts” of prior climate variables to historical observations, the development of hindcast ...1 Introduction · 2 Methods · 3 ResultsMissing: CMIP | Show results with:CMIP
  99. [99]
    Robustness of CMIP6 Historical Global Mean Temperature ...
    Sep 10, 2020 · The post-1998 warming is overestimated in 90% of the simulations. Only six out of 29 models reproduce the observed long-term persistence. All ...Introduction · Data and Methods · Results · Conclusions
  100. [100]
    Global surface air temperatures in CMIP6: historical performance ...
    Oct 1, 2020 · The CMIP6 mean can capture the trends of global surface temperatures shown by the observational data during 1901–1940 (warming), 1941–1970 (cooling) and 1971– ...
  101. [101]
    Constrained CMIP6 projections indicate less warming and a slower ...
    Jul 15, 2022 · However, most of the CMIP6 and CMIP5 models overestimate historical precipitation when compared with the observations (Fig. 1c, Section 1 in ...
  102. [102]
    Comparison of CMIP6 historical climate simulations and future ...
    May 10, 2021 · (2020b) and CONSTRAIN (2020) that some of the CMIP6 models overestimate recent warming trends. Tokarska et al. (2020b) examine the trend in ...
  103. [103]
    [PDF] Why Does the Ensemble Mean of CMIP6 Models Simulate Arctic ...
    May 3, 2024 · while the CMIP6 simulation shows a large global warming overestimate since the 2000s [30]. The average underestimate of global warming in ...
  104. [104]
    The Double‐ITCZ Bias in CMIP3, CMIP5, and CMIP6 Models Based ...
    Mar 28, 2020 · The double-intertropical convergence zone (ITCZ) bias is one of the most outstanding errors in all previous generations of climate models.Introduction · Data and Methodology · Results · Summary
  105. [105]
    Linking large-scale double-ITCZ bias to local-scale drizzling bias in ...
    Sep 2, 2022 · This study shows that local-scale drizzling bias in atmospheric models can lead to large-scale double-ITCZ bias in coupled models.
  106. [106]
    Climate models without an East Pacific Double ITCZ better simulate ...
    Our findings suggest that reducing the double-ITCZ bias in climate models has the potential to substantially improve climate projections.Tian, B. & Dong, X. The ...
  107. [107]
    Local and Remote Causes of the Equatorial Pacific Cold Sea ...
    A long-standing problem in CGCMs is the cold equatorial Pacific (EP) SST bias (hereafter cold bias), which manifests as an excessive and overly narrow cold ...
  108. [108]
    Impacts of Wind Profile Shear and Curvature on the Parameterized ...
    May 9, 2025 · The cold pole and westerly wind biases associated with an overly strong polar vortex are typical systematic biases in climate models, ...
  109. [109]
    The Cold Biases in the Soil and Surface Air Temperature ...
    Apr 25, 2025 · The CMIP5 global climate models (GCMs) produce consistent cold biases of the surface air temperature, ground temperature, and soil temperature ...
  110. [110]
    Linking Large-Scale Double-ITCZ Bias to Local-Scale Drizzling Bias ...
    This study shows that local-scale drizzling bias in atmospheric models can lead to large-scale double-ITCZ bias in coupled models.
  111. [111]
    Diurnal cloud cycle biases in climate models | Nature Communications
    Dec 22, 2017 · Clouds' efficiency at reflecting solar radiation and trapping the terrestrial radiation is strongly modulated by the diurnal cycle of clouds ...Results · Errors Of Dcc · Methods
  112. [112]
    [PDF] A Multimodel Study on Warm Precipitation Biases in Global Models ...
    The biases identified in vertical microphysical structure of GCMs corroborate the too fast productions of precipitation on one hand, and hint at how cloud ...
  113. [113]
    [PDF] Global Climate Models and Bias Correction | GLISA
    Spatial downscaling is often used to tailor model simulations to localities. Statistical downscaling aims to tailor GCM output by generating statistically ...
  114. [114]
    Understanding the Cascade: Removing GCM Biases Improves ...
    May 4, 2024 · We find that native GCMs tend to exhibit surprisingly common mean biases that, when downscaled, effectuate an overly wet, cold, and snowy ...Introduction · Results · Using Physics to Tie GCM... · Conclusions
  115. [115]
    Modeling general circulation model bias via a combination ... - ASCMO
    Feb 2, 2023 · In this article, we propose a novel extension to standard quantile mapping that allows for a continuous seasonal change in bias magnitude using localized ...
  116. [116]
    ICON: Toward Vertically Integrated Model Configurations for ...
    On a physical level, one notes a great similarity between the governing equations underlying the models for numerical weather prediction (NWP) and for climate ...
  117. [117]
    Toward Data‐Driven Weather and Climate Forecasting ...
    Nov 12, 2018 · It is shown that it is possible to emulate the dynamics of a simple general circulation model with a deep neural network.
  118. [118]
    The Beginnings of Numerical Weather Prediction and General ...
    This chapter presents early recollections of Joseph Smagorinsky on the beginning of numerical weather prediction and general circulation modeling.
  119. [119]
    Coupled Model Intercomparison Project: CMIP
    CMIP is a project of the World Climate Research Programme (WCRP) providing climate projections to understand past, present and future climate changes.CMIP Modelling Centres and... · CMIP Brand and Visual Identity · CMIP Overview
  120. [120]
    CMIP6: the next generation of climate models explained - Carbon Brief
    Dec 2, 2019 · One way to evaluate hindcast performance is to compare the rate of warming in models and observations over a period of time. The figure below ...
  121. [121]
    [PDF] Future Global Climate: Scenario-based Projections and Near-term ...
    ... (IPCC, 2014), and these uncertainties matter a great deal when assessing remaining carbon budgets consistent with limiting global warming to 1.5°C above pre ...
  122. [122]
    Chapter 4 | Climate Change 2021: The Physical Science Basis
    Projected changes in the global monsoon circulation are also uncertain, because they are influenced by internal variability such as AMV and PDV (see Section ...
  123. [123]
    Quantifying CMIP6 model uncertainties in extreme precipitation ...
    Projected changes in precipitation extremes and their uncertainties are evaluated using an ensemble of global climate models from phase 6 of the Coupled Model ...
  124. [124]
    Study Confirms Climate Models are Getting Future Warming ...
    Jan 9, 2020 · A study found that most climate models have been quite accurate in projecting future warming, with 10 of 17 projections closely matching ...
  125. [125]
    Analysis: How well have climate models projected global warming?
    Oct 5, 2017 · Global surface air temperatures in CMIP5 models have warmed about 16% faster than observations since 1970. About 40% of this difference is due ...
  126. [126]
    A Short Guide to the Climatic Variables of the Last Glacial Maximum ...
    Here, we analyzed the climatic predictions for the LGM of 9 different GCMs in order to help biogeographers to select their GCMs and climatic layers for mapping ...
  127. [127]
    [PDF] The PMIP4 Last Glacial Maximum experiments: preliminary results ...
    May 20, 2021 · Most of the models that have run the PMIP4-CMIP6 LGM simulations are general circulation models (GCMs) but iLOVECLIM is an Earth system model of ...
  128. [128]
    Modeling the climate of the Last Glacial Maximum from PMIP1 to ...
    The Last Glacial Maximum is an example of an extreme climate, and has thus been a target for climate models for many years.
  129. [129]
    General circulation models, palaeoclimatic data and last interglacial ...
    GCM simulations of last interglacial climates suggest that the relatively simple patterns of change due to direct insolation effects were radically altered by ...
  130. [130]
    [PDF] The climate of the Last Glacial Maximum: Results from a coupled ...
    This paper presents the results of a decadal simulation of a coupled atmosphere-ocean GCM which has been configured for the LGM. A decadal simulation is ...
  131. [131]
    2 Paleoclimatic Modeling--A Review with Reference to Problems ...
    The purpose of this chapter is to review and evaluate climate modeling as a tool for paleoclimatic research.The Climate System And... · Characteristic Model... · Paleoclimatic Data Assembly...
  132. [132]
    Regional climate models downscaling analysis of general ...
    Apr 22, 2008 · The RCMs' downscaling reduces significantly driving GCMs' present-climate biases and narrows inter-model differences in representing climate sensitivity.
  133. [133]
    Regional climate models: 30 years of dynamical downscaling
    RCMs have been used to derive climatologies, analyze climate variability, provide input for climate change impact assessment, and downscale the hydrologic cycle ...
  134. [134]
    [PDF] High-resolution regional simulation of last glacial maximum climate ...
    ABSTRACT. A fully coupled atmosphere–ocean general circulation model is used to simulate climate conditions during the last glacial maximum (LGM).
  135. [135]
    Downscaling Climate Models: Sharpening the Focus on Local-Level ...
    Jan 1, 2012 · “Downscaling” climate models are an attempt to bridge the gap between global and local effects by layering local-level data over larger-scale climate models.
  136. [136]
    Reassessing the Value of Regional Climate Modeling Using ...
    Nov 9, 2019 · Regional climate models (RCMs) are often assumed to be more skillful compared to lower-resolution general circulation models (GCM).
  137. [137]
    The Art and Science of Climate Model Tuning in - AMS Journals
    We survey the rationale and diversity of approaches for tuning, a fundamental aspect of climate modeling, which should be more systematically documented.DEFINITION OF CLIMATE... · APPLYING OBJECTIVE... · TUNING AND MODEL...
  138. [138]
    Practice and philosophy of climate model tuning across six U.S. ...
    Some observational targets have important (and sometimes unrecognized) structural uncertainties and therefore any tuning to those targets risks over-fitting the ...
  139. [139]
    A Test of the Tropical 200-300 mb Warming Rate in Climate Models
    Sep 17, 2018 · John and I have published a new paper in Earth and Space Science using tropical mid-troposphere trend comparisons (models versus observations).
  140. [140]
    INCONSISTENCY OF MODELED AND OBSERVED TROPICAL ...
    modeled and observed tropical upper-troposphere temperature trends has bee challenged because it strikes at the heart of the evidence used to back claims.
  141. [141]
    Advanced Testing of Low, Medium, and High ECS CMIP6 GCM ...
    Mar 14, 2022 · The fact that the high and medium-ECS GCMs do not appear to be consistent with the observations over the past 40+ years imply that their ...
  142. [142]
    Use of 'too hot' climate models exaggerates impacts of global warming
    May 4, 2022 · But for the 2019 CMIP6 round, 10 out of 55 of the models had sensitivities higher than 5°C—a stark departure. The results were also at odds with ...
  143. [143]
    Global Warming: Observations vs. Climate Models
    Jan 24, 2024 · The observed rate of global warming over the past 50 years has been weaker than that predicted by almost all computerized climate models.
  144. [144]
    High Values of the Arctic Amplification in the Early Decades of the ...
    Nov 29, 2023 · The reason is that the ensemble mean of the CMIP6 models overestimates the rate of global warming since about 1990. Since the AA is defined as a ...
  145. [145]
    Understanding the Origin of Tropical Precipitation Bias in Climate ...
    Dec 15, 2022 · Despite generations of development, the double-ITCZ bias has persisted in climate models. This impairs the simulation of tropical climate and ...
  146. [146]
    Diagnosing drivers of tropical precipitation biases in coupled climate ...
    Aug 1, 2024 · In this study, we examine the main sources of the large-scale tropical (30°S-30°N) precipitation mean biases in coupled climate model ...
  147. [147]
    Understanding the Biases in Daily Extreme Precipitation ...
    Jun 17, 2025 · Compared with ERA5, models demonstrate a significant wet bias in GLM domain for the annual maximum daily precipitation (14.14%) and the extreme ...
  148. [148]
    Contextualizing GCM Biases in Low‐Cloud Coverage: The Role of ...
    Jul 24, 2025 · Large biases in low clouds exist over the oceans in climate simulations. These biases are usually considered a result of cloud properties being ...
  149. [149]
    Moderate climate sensitivity due to opposing mixed-phase cloud ...
    Mar 4, 2025 · Many studies have since focused on identifying sources of errors in the representation of clouds in GCMs that impact climate sensitivity in ...
  150. [150]
    Widespread global disparities between modelled and observed mid ...
    Apr 12, 2023 · The study highlights discrepancies between modelled and observed ocean circulation in terms of spatial structure, intensity, and direction that ...
  151. [151]
    Do General Circulation Models Underestimate the Natural Variability ...
    The results indicate that unarguable improvements in the treatment of sea ice in the GFDL climate model should amplify significantly the natural variability in ...Missing: challenges | Show results with:challenges
  152. [152]
    Absence of internal multidecadal and interdecadal oscillations in ...
    Jan 3, 2020 · There are no other robust spectral features found in the control simulations, including neither bidecadal PDO (15–20 year) nor multidecadal AMO ...Missing: GCM errors
  153. [153]
    Persistent Model Biases in the Spatial Variability of Winter North ...
    Nov 28, 2023 · In this study, we compare the winter weather patterns over the North Atlantic region from the last millennium to future scenarios in climate models.<|separator|>
  154. [154]
    Natural variability contributes to model–satellite differences in ...
    Mar 22, 2021 · Our results indicate that multidecadal variability can explain current model–observational differences in the rate of tropical tropospheric warming.
  155. [155]
    The contribution of natural variability to GCM bias - NASA ADS
    Investigating the impacts of climate change often entails using projections from inherently imperfect general circulation models (GCMs) to drive models that
  156. [156]
    Modulation of the Relationship between ENSO and Its Combination ...
    Jun 1, 2020 · We compute the anomalies by subtracting the long-term monthly mean from the simulation itself. 3. Decadal variations of ENSO combination mode ...
  157. [157]
    Missing feedbacks, asymmetric uncertainties, and ... - AGU Journals
    May 26, 2006 · [2] The uncertainty reported for GCM projections of climate change stems largely from the treatment of ice-albedo, water vapor, and cloud ...
  158. [158]
    Current GCMs' Unrealistic Negative Feedback in the Arctic in
    In the Arctic, the negative longwave feedback is particularly strong because of weak positive longwave cloud cover and water vapor feedbacks. Generally, given ...
  159. [159]
    A less cloudy picture of the inter-model spread in future global ...
    Sep 8, 2020 · We show that the ice-albedo feedback spread explains uncertainties in polar regions while the water vapor feedback spread explains uncertainties ...<|separator|>
  160. [160]
    Re-Examining the First Climate Models - AMS Journals
    Not taking clouds into account is a deliberate choice, as their inclusion introduces a considerable number of additional degrees of freedom, which are best ...A. Radiation (rrtmg) · 4. Radiative Feedback · B. Climate Radiative...<|control11|><|separator|>
  161. [161]
    Climate sensitivity from radiative-convective equilibrium
    Sep 1, 2023 · C. A radiative-convective equilibrium model. Convection brings water vapor and heat from the surface into the interior of the atmosphere, where ...INTRODUCTION · Single-layer radiative equilibrium · A radiative-convective...
  162. [162]
    1.7: Models - Geosciences LibreTexts
    Jul 22, 2021 · Therefore, radiative-convective models include convection, mostly by limiting the lapse rate to the observed or moist adiabatic rate.Missing: explanation | Show results with:explanation
  163. [163]
    4.4.2. Radiative-Convective Models - Climate Change
    Aug 25, 2018 · RCMs simulate in detail the transfer of energy through the depth of the atmosphere, including: a) the radiative transformations that occur as energy is ...
  164. [164]
    EMIC intercomparison project (EMIP-CO2) - MIT CS3
    In particular EMICs show similar temperature and precipitation changes with comparable magnitudes and scatter across the models as found in the GCMs. The ...
  165. [165]
    The utility of Earth system Models of Intermediate Complexity (EMICs)
    Intermediate-complexity models are models which describe the dynamics of the atmosphere and/or ocean in less detail than conventional General Circulation ...
  166. [166]
    How parameter specification of an Earth system model of ...
    Jun 26, 2019 · In this study, we used an Earth system model of intermediate complexity (EMIC)—LOVECLIM as an example to illustrate how SA methods can be used ...
  167. [167]
    8.8.3 Earth System Models of Intermediate Complexity - AR4 WGI ...
    Actually, there is a broad range of EMICs, reflecting the differences in scope. In some EMICs, the number of processes and the detail of description are reduced ...
  168. [168]
    Earth System Modeling, a definition - Climateurope
    Earth system models (ESM) seek to simulate all relevant aspects of the Earth system. They include physical, chemical and biological processes.
  169. [169]
    Earth System Model - Geophysical Fluid Dynamics Laboratory - NOAA
    The climate system consists of five interacting components: the atmosphere, the hydrosphere, the cryosphere, the land surface and the biosphere. Scenarios of ...Missing: definition | Show results with:definition
  170. [170]
    DOE Explains...Earth System and Climate Models
    Earth system models and climate models are a complex integration of environmental variables used for understanding our planet.
  171. [171]
    The Community Earth System Model Version 2 (CESM2)
    Jan 16, 2020 · The Community Earth System Model Version 2 (CESM2) is the latest generation of the coupled climate/Earth system models developed as a ...<|separator|>
  172. [172]
    Community Earth System Model 2 (CESM2)
    CESM is a fully-coupled, community, global Earth system model that provides state-of-the-art computer simulations of the Earth's past, present, and future ...Downloading Instructions · Release Series Information · Atmosphere Component
  173. [173]
    Community Earth System Model: Home
    CESM is a fully-coupled, community, global Earth system model that provides state-of-the-art computer simulations of the Earth's past, present, and future ...Models · About · Experiments · CESM2
  174. [174]
    Fact sheet: Earth system modelling at ECMWF
    Mar 8, 2021 · ECMWF's Earth system model aims to represent interactions between as many Earth system components as required, at the necessary level of complexity.
  175. [175]
    Differentiable programming for Earth system modeling - GMD
    Jun 2, 2023 · Comprehensive Earth system models (ESMs) are the key tools to model the dynamics of the Earth system and its climate and in particular to ...
  176. [176]
    Neural General Circulation Models for Weather and Climate - arXiv
    Nov 13, 2023 · General circulation models (GCMs) are the foundation of weather and climate prediction. GCMs are physics-based simulators which combine a ...
  177. [177]
    How AI is helping 38 million farmers with advance weather predictions
    Sep 15, 2025 · The University of Chicago uses NeuralGCM, a Google Research model, to more accurately predict the monsoon season in India and support farmer ...
  178. [178]
    Simpler models can outperform deep learning at climate prediction
    Aug 26, 2025 · New research shows the natural variability in climate data can cause AI models to struggle at predicting local temperature and rainfall. Adam ...Missing: 2023-2025 | Show results with:2023-2025
  179. [179]
    Neural general circulation models optimized to predict satellite ...
    Dec 16, 2024 · Our model runs at 2.8^\circ resolution and is built on the differentiable NeuralGCM framework. The model demonstrates significant improvements ...
  180. [180]
    The History of Numerical Weather Prediction - NOAA
    Oct 31, 2023 · The group was headed by Jule Charney, who had done extensive work on ... The first one-day, nonlinear weather prediction was made in April, 1950.
  181. [181]
    The ENIAC Computations of 1950—Gateway to Numerical Weather ...
    The first numerical weather prediction was made on the ENIAC computer in 1950. This lecture gives some of the historical background of that event.
  182. [182]
    The general circulation of the atmosphere: A numerical experiment
    A long-period numerical forecast is made with a two-level quasi-geostrophic model, starting with an atmosphere in relative rest.
  183. [183]
    Clarifying the Dynamics of the General Circulation: Phillips's 1956 ...
    Analysis of the model results clarified the respective roles of the synoptic-scale eddies (cyclones-anticyclones) and mean meridional circulation in the ...
  184. [184]
    1955-65: Establishment of Atmospheric General Circulation Modeling
    In the mid-1950s, Norman Phillips created a 2-layer model. The General Circulation Research Section started in 1955, and UCLA and other groups began building 3 ...
  185. [185]
    See One of the First Climate Models - Earth Matters
    May 21, 2015 · One of the first general circulation models was developed at Lawrence Livermore National Laboratory by Cecil “Chuck” Leith in the early 1960s.Missing: 1950s | Show results with:1950s
  186. [186]
    Bibliography - Syukuro Manabe
    Manabe, Syukuro, Joseph Smagorinsky, J L Holloway, Jr, and H Stone, 1970: Simulated climatology of a general circulation model with a hydrologic cycle. III ...
  187. [187]
    Timeline: The history of climate modelling - Carbon Brief
    Jan 16, 2018 · Norman Phillips' first general circulation model in 1956. The establishment of a modelling group at the National Center for Atmospheric ...
  188. [188]
    Extended ENSO Predictions Using a Fully Coupled Ocean ...
    Sophisticated ocean–atmosphere coupled general circulation models (GCMs) should resolve both ENSO and its dynamically linked global climate variations in an ...
  189. [189]
    CMIP6 – description and comparison to CMIP5
    May 19, 2025 · The IPCC Sixth Assessment found that CMIP6 shows several incremental improvements and enhancements over CMIP5, but many of the same features and ...
  190. [190]
    Comparisons Between CMIP5 and CMIP6 Models: Simulations of ...
    Apr 20, 2021 · This study compares the behavior of climate models participating in the Climate Model Intercomparison Project Phase 6 (CMIP6) with their CMIP5 predecessors.
  191. [191]
    Applicability of CMIP5 and CMIP6 Models in China - AMS Journals
    Compared to CMIP5, CMIP6 models feature increases in spatial resolution, improvements in physical parameterizations (e.g., the representation of clouds), and ...
  192. [192]
    Comparison of CMIP5 and CMIP6 GCM performance for flood ...
    The results indicated that precipitation from CMIP6 had a higher correlation and a lower error coefficient than CMIP5. Similarly, the simulation of GCM ...
  193. [193]
    Earth's future climate and its variability simulated at 9 km global ...
    Jul 17, 2025 · The study simulated Earth's future climate at 9 km resolution, showing a 6.5°C warming by 2100, and increased regional information on ...
  194. [194]
    The Fall and Rise of the Global Climate Model - AGU Journals - Wiley
    Aug 29, 2021 · A recent study describes a model that appears to be much better at aerosol–cloud interactions than previous global models, potentially solving half the problem.