The UCLA General Circulation Model (1961–present)

The UCLA General Circulation Model (UCLA GCM; also called the Mintz–Arakawa model in its early years and later the UCLA AGCM) is one of the three original atmospheric GCMs of the 1960s, alongside the GFDL model (Smagorinsky and Manabe, Washington DC / Princeton) and the NCAR model (Kasahara and Washington, Boulder). Unlike GFDL and NCAR, which were government-funded research laboratories, UCLA was an academic graduate programme, and its model has had arguably the most wide-spread indirect influence on the field: its numerical methods – the Arakawa Jacobian (1966), the A/B/C/D/E staggered grids (Arakawa & Lamb, 1977), and the Arakawa–Schubert convection scheme (1974) – are embedded in virtually every weather- and climate-model lineage alive in 2026.

Origins: Bjerknes, Mintz, and the SWAC (1940–1961)

The UCLA Department of Meteorology was founded in 1940 by Jacob Bjerknes (son of Vilhelm Bjerknes, the Bergen-school founder), who had a longstanding interest in the atmospheric general circulation. Yale Mintz, Bjerknes’s second-ever Ph.D. student (1949), inherited the descriptive UCLA General Circulation Project, which used the SWAC (Standards Western Automatic Computer – one of the earliest UCLA computers) to digest observational data by paper tape. By the late 1950s Mintz had concluded that the future lay not in data analysis but in numerical simulation, following Phillips (1956).

Phase 0: The Mintz–Arakawa Prototype (1961–1965)

Personnel

Yale Mintz (PI, visionary, non-programmer)
Akio Arakawa (visiting scientist from the Japan Meteorological Agency, 1961–1963; hired permanently 1965)

Mintz recruited Arakawa on the recommendation of a University of Tokyo meteorology professor; Arakawa arrived in Los Angeles in the spring of 1961, leaving his wife and two-year-old son in Tokyo for the first five months. He returned to Japan in 1963 to satisfy the two-year foreign- residency requirement of the J-visa; Mintz wrote him “every month” to persuade him to come back, which he did in 1965.

The first problem: computational instability

Mintz wanted to start coding a GCM immediately. Arakawa refused. He had read Phillips’s 1959 paper in Tokyo, knew that the 1956 experiment had blown up after about 30 simulated days from “nonlinear computational instability,” and insisted, to Mintz’s initial frustration, that “it would be futile to launch straight into programming” before the mathematical problem was solved. Mintz reluctantly agreed to delay.

The Arakawa Jacobian (1962–1966)

Arakawa’s solution was to rewrite the nonlinear advection terms of the vorticity equation in finite-difference form so that the total kinetic energy and total enstrophy (mean-square vorticity) were both exactly conserved at each time step. This simultaneously blocked the erroneous cascade of energy into the smallest resolvable scales (the aliasing mechanism Phillips had identified) and prevented any single wavenumber from growing without bound. The physical analogue is the two-dimensional turbulence cascade later formalized by Kraichnan (1967): in 2-D, energy cascades to large scales while enstrophy cascades to small. Arakawa’s scheme respected both budgets discretely.

Arakawa first presented the scheme at conferences in 1962. He was too busy coding to publish until 1966:

Arakawa, A. (1966). “Computational Design for Long-Term Numerical Integration of the Equations of Fluid Motion: Two-Dimensional Incompressible Flow. Part I.” Journal of Computational Physics 1(1), 119–143.

This paper is now regarded as a foundational document of computational fluid dynamics – cited far beyond meteorology, in engineering journals that no meteorologist ever reads. It was reprinted by J. Comp. Phys. in 1997 for the journal’s 50th anniversary with a new commentary by Arakawa. Initial reception was mixed: some colleagues found the approach “intuitively implausible” because it forced conservation of a quantity (enstrophy) that nature itself does not conserve. Arakawa’s answer was pragmatic: the conservation was a computational device, not a physical claim.

The Prototype Model

Built between 1961 and 1963, using the IBM 709 belonging to the UCLA Business School – access was only available on weekends, and Arakawa (who had used an IBM 704 in Tokyo) operated the machine himself. Key characteristics:

2 vertical levels (sigma coordinates).
7 degree latitude x 9 degree longitude horizontal resolution.
Global domain (unlike Smagorinsky’s hemispheric model).
Realistic land–sea distribution and surface topography – a first.
Primitive equations (not quasi-geostrophic).
No water vapour in the prototype (dry dynamics only).
Programmed in machine language; only ~2,000 FORTRAN statements in later versions.
Written and operated by Arakawa; Mintz never learned to programme.

Mintz published the first results in 1964. The model was abandoned about 1965 when Arakawa returned to UCLA permanently.

Phase I: UCLA I – the “Production” Mintz–Arakawa Model (1965–early 1970s)

When Arakawa returned in 1965, he and Mintz built what Paul Edwards calls the first-generation “production” UCLA GCM:

2 levels, 4 degree x 5 degree horizontal resolution.
Introduced the Arakawa–Lamb “B” grid (Arakawa’s co-author Vivian R. Lamb was a computer scientist at UCLA). B-grid: scalars at cell centres, vector wind components together at the corners.
Added water-vapour physics, precipitation, and simple surface evaporation.
Ran on IBM mainframes (709 then 7094), with computational support from IBM’s Large Scale Scientific Computation Department in San Jose. IBM scientists W. E. Langlois and H. C. W. Kwok flew to UCLA twice a week (Langlois attending Arakawa’s NWP lectures) and wrote a detailed documentation: “Numerical Experiments with the Mintz–Arakawa General Circulation Model – Part 1: Description of the Model,” IBM Research Report RJ-501, 24 May 1968.

UCLA I was the model that “went out into the world.” Its lineage outward:

RAND Corporation: UCLA Ph.D. W. Lawrence Gates carried it to RAND in Santa Monica around 1970, where it was used in ARPA-sponsored studies. A documentation was published as “A Documentation of the Mintz–Arakawa Two-Level Atmospheric General Circulation Model,” RAND Report R-877, 1971.
Oregon State University: The RAND version migrated to Oregon State when Gates moved there; it became the foundation of the OSU GCM.

Phase II: UCLA II – Vertical Extension (late 1960s, early 1970s)

3- and 9-level versions were built, extending vertical resolution.
Retained the B-grid.
The 9-level UCLA II is the basis of “Description and preliminary results of the 9-level UCLA general circulation model” (Arakawa, Katayama, Mintz; Proc. WMO/IUGG Symposium on Numerical Weather Prediction, Tokyo, 1969).

Lineage outward (UCLA II):

NASA Goddard Institute for Space Studies (GISS), New York City (1972): Milton Halem and James Hansen adopted the UCLA II model in
1. Under Hansen’s direction the GISS ModelE and its ancestors (Model II, Model II’) were direct descendants – Hansen explicitly credited Arakawa’s scheme for letting GISS use a grid “as much as a thousand kilometers on a side” while still producing a realistic jet stream. The GISS model was central to the 1979–1988 climate-sensitivity debates and to Hansen’s 1988 Senate testimony.
Goddard Laboratory for Atmospheric Sciences and Goddard Laboratory for Atmospheres: later 1970s, same source.

Phase III: UCLA III – the C-grid (mid-1970s)

12 vertical levels.
Switched from the B-grid to the Arakawa–Lamb “C” grid: scalars at cell centres, u at east/west cell faces, v at north/south cell faces. The C-grid became the dominant choice for atmosphere and ocean models from the 1980s onward because it handles gravity waves and geostrophic adjustment correctly at the gridscale.
Two variants (different prognostic variables) were built.
The definitive technical description was:

Arakawa, A., and V. R. Lamb (1977). “Computational Design of the Basic Dynamical Processes of the UCLA General Circulation Model.” In General Circulation Models of the Atmosphere (J. Chang, ed.), Methods in Computational Physics, vol. 17. Academic Press, pp. 173–265.

Arakawa himself, in his 1997 AIP oral history, said some call this paper “the bible of modeling.” It is one of the most-cited papers in atmospheric science.

Lineage outward (UCLA III):

U.S. Naval Environment Prediction Research Facility (NEPRF) and Fleet Numerical Oceanographic Center (FNOC), Monterey, CA: UCLA III was transferred there and evolved into the operational NOGAPS (Navy Operational Global Atmospheric Prediction System), in use through the 1990s and 2000s before being replaced by NAVGEM.
Meteorological Research Institute (MRI), Tsukuba, Japan: a version was transferred there in the late 1970s and continued in use for forecasting and climate studies.

Phase IV: UCLA IV – PBL Vertical Coordinate (late 1970s–)

Begun in the late 1970s. The chief innovation was a new vertical coordinate system that used the top of the planetary boundary layer as a coordinate surface, rather than a fixed sigma level. This was the work of Akio Arakawa, Max Suarez, and David Randall.

Lineage outward (UCLA IV):

Colorado State University (CSU): David Randall, an Arakawa student who had finished his UCLA Ph.D. in 1976 under Arakawa, carried UCLA IV to CSU in 1988. Its CSU descendants include the CSU general circulation model and later the icosahedral/geodesic SAM/BUGS/Multiscale Modeling Framework line.
Lawrence Livermore National Laboratory: a version migrated there in the 1980s.
Central Weather Bureau of the Republic of China (Taiwan): also received a version.
Goddard Laboratory for Atmospheres (NASA): a reimplemented version of UCLA IV was used at Goddard into the 1980s.

The UCLA AGCM in 2026

A direct lineal descendant, the UCLA AGCM maintained by Carlos R. Mechoso and colleagues in the UCLA Department of Atmospheric and Oceanic Sciences, remained in active research use into the 2020s as the atmospheric component of the UCLA Earth System Model (ESM). It is a grid-point model extending from the surface to about 50 km, used in studies of coupled ocean–atmosphere-chemistry dynamics. Key versions include UCLA AGCM6.4 (Mechoso et al.). The model is no longer in operational forecasting use anywhere (NOGAPS has been replaced by NAVGEM; MRI has moved on; Hansen’s GISS ModelE has been substantially rewritten but still carries Arakawa heritage) – but Arakawa’s numerical methods are essentially ubiquitous:

Method	Where used today
Arakawa C-grid	MITgcm, NEMO (European ocean model), MOM (GFDL ocean), WRF (atmosphere), ROMS, COSMO, HARMONIE/AROME
Arakawa Jacobian / energy-enstrophy conservation	All serious atmospheric and oceanic dynamical cores descend from this logic
Cubed-sphere / icosahedral variants of C-grid	NOAA FV3 (GFDL finite-volume cubed sphere, the dynamical core of the US GFS from 2019 onward), MPAS (NCAR), NICAM (Japan), UKMO LFRic
Arakawa–Schubert convection	Historically NCEP GFS, ECMWF (briefly), JMA; now largely displaced by Tiedtke (1989) and Bechtold mass-flux schemes at ECMWF, and Zhang–McFarlane at NCAR. Scale-aware “Simplified Arakawa–Schubert” variants remain in WRF and some operational systems.
B-grid	Largely abandoned for modern atmospheres; persists in some ocean models (older MOM, HYCOM)

The Three First-Generation GCMs: a Comparison

	GFDL	NCAR	UCLA
Leaders	Smagorinsky, Manabe	Kasahara, Washington	Mintz, Arakawa
First results	1963–1965	1967	1964
Domain	Hemispheric → global	Hemispheric → global	Global from the start
Topography	Idealized initially	Idealized initially	Realistic from the start
Grid	Latitude-longitude	Latitude-longitude	B-grid, later C-grid (pioneered by UCLA)
Primary output	Climate sensitivity (Manabe–Wetherald 1967, 1975)	Diurnal and seasonal cycles	Monsoons, Mars, ozone
Institutional type	Federal research lab	Federal research centre	Academic graduate programme
Role in IPCC	Core contributor via GFDL	Core contributor via NCAR CESM	Indirect: via GISS, OSU, NASA

Direct Descendants (a Family Tree)

Mintz-Arakawa prototype (1961-1965)
         |
     UCLA I (B-grid, 1965-early 1970s)
       /       \
      |         \--> RAND (1970) --> OSU GCM (Gates)
     UCLA II (3- and 9-level, late 1960s)
         \--> GISS (1972) --> Hansen's Model II/II'/ModelE
         \--> NASA Goddard Lab for Atmos. Sciences
     UCLA III (12-level, C-grid, mid-1970s)
         \--> NOGAPS (Navy, Monterey) --> replaced by NAVGEM
         \--> MRI Tsukuba (Japan)
     UCLA IV (PBL vertical coord, late 1970s)
         \--> CSU (Randall, 1988) --> CSU GCM --> SAM/BUGS/MMF
         \--> LLNL
         \--> Goddard Lab for Atmospheres
     UCLA AGCM (Mechoso, present day)
         |--> part of UCLA Earth System Model

Scientific Results of Note

Because UCLA operated as an academic graduate programme, “production” simulations were often run elsewhere; the UCLA group’s own headline results include:

First global GCM with realistic topography and land–sea distribution (1964).
First Mars general-circulation simulation (Leovy & Mintz, mid-1960s) – a full five years before Mariner 9 arrived.
First simulations of the Asian monsoon with realistic Tibetan Plateau orography (1970s).
Three-dimensional stratospheric ozone transport (Schlesinger & Mintz).
Early coupled ocean–atmosphere experiments (proposed by Mintz in the late 1960s; implemented by the UCLA group and collaborators in the 1980s).
Arakawa–Schubert convection scheme (1974) – first cumulus parameterization built on the idea of convective quasi-equilibrium, still one of the conceptual foundations of the field.

The Arakawa–Schubert Convection Scheme (1974)

With Wayne Schubert – Arakawa’s Ph.D. student, who moved to Colorado State in 1973 – Arakawa formulated what became the archetype of mass-flux cumulus parameterization:

Arakawa, A., and W. H. Schubert (1974). “Interaction of a Cumulus Cloud Ensemble with the Large-Scale Environment, Part I.” Journal of the Atmospheric Sciences 31(3), 674–701.

The key concept, convective quasi-equilibrium: the rate at which large-scale dynamics destabilize the atmosphere (by lifting moist air upward in the tropics) is, to good approximation, balanced by the rate at which cumulus convection stabilizes it (by vertical mixing of heat and moisture). The ensemble of cumulus clouds adjusts to the large-scale forcing on a timescale short compared to the large-scale dynamics – so one can parameterize the clouds as an ensemble in quasi-equilibrium.

Arakawa noted in his 1997 oral history that the paper received two kinds of criticism: modellers said it was “too complicated” and too expensive to run operationally; observationalists said real clouds were not simple enough to fit its assumptions. Over the following decades the scheme was progressively adopted and then gradually displaced by simpler mass-flux schemes (Tiedtke 1989 at ECMWF, Zhang–McFarlane 1995 at NCAR, and many “simplified Arakawa–Schubert” variants in operational practice). The concept of quasi-equilibrium, however, remains foundational.

Machines the UCLA GCM Has Run On

IBM 709 (UCLA Business School), weekends only, 32K memory – first prototype, 1961–1965.
IBM 7094, IBM 360, later IBM 3090 – through the 1960s–1980s.
Cray-1, Cray X-MP, Cray Y-MP – 1980s.
Massively parallel machines (Intel Paragon, Cray T3E, IBM SP) – 1990s–2000s, via NASA and DOE allocations.
Modern HPC clusters – present day, through DOE/NCAR time.

Primary Sources and References

Accessed: 2026-04-19

Michał Brennek