The Scientific Arc of Norman A. Phillips (1923-2019)

Research notes for Post 43. The 1956 GCM is the canonical centrepiece of Phillips’s career and is covered in detail in Post 7 (“The First Climate Model Had 5 KB of RAM”). This document maps the full intellectual arc – before, alongside, and especially after the 1956 GCM. The 1956 paper is treated here in one summary chapter; the surrounding work needs the bulk of the space.

Compiled 2026-05-17 from: Lewis 1998 BAMS retrospective (PDF, /sources/Phillips-sources), the Phillips 1995 NAS biographical memoir on Charney, the Phillips 1956 QJRMS paper itself, the existing Phillips.md people file, the existing NMC_LFM concept file, MIT News obituary, Franklin Institute citation, Peter Lynch’s tribute, the AMS award page, and primary papers verified via WebFetch and WebSearch (1957 sigma, 1959 NCI, 1960 initialisation, 1963 Reviews of Geophysics, 1979 NGM technical report, 1986 Tellus).

1. INTELLECTUAL ANCESTRY: Why Phillips ended up where he did

Phillips operates at the confluence of three lineages, all of which run through the University of Chicago Meteorology Department in the late 1940s.

1.1 The Bjerknes-Rossby line (synoptic dynamics)

The Bergen School of Vilhelm Bjerknes (1862-1951), Jacob Bjerknes, and Halvor Solberg gave meteorology its frontal-cyclone model in the years around 1920. Carl-Gustaf Rossby (1898-1957) imported this Norwegian tradition to the United States via MIT (1928-1938), where he hosted Erik Palmen and Jacob Bjerknes during the 1930s, and brought it to Chicago in 1941 as founding head of the Institute of Meteorology. From the Bergen line Phillips inherited the conviction that atmospheric motions are best understood as dynamical phenomena, not statistical regularities. The synoptic emphasis on cyclones, fronts, and air masses dominated his graduate education.

But Rossby’s own work was already moving beyond pure synoptics. Rossby (1939) discovered the planetary wave that bears his name, derived from the variation of the Coriolis parameter with latitude (the beta term). Rossby (1947) extended this to a theory of large-scale lateral diffusion of vorticity by turbulent eddies as the mechanism maintaining the zonal mean flow. Phillips inherited Rossby’s intuition that simple models could be quantitatively useful even when – maybe especially when – they did not look like the real atmosphere.

1.2 The Charney-Eady line (quasi-geostrophic theory)

Jule Charney’s 1947 paper “The dynamics of long waves in a baroclinic westerly current” (J. Meteorology 4, 135-162) and Eric Eady’s 1949 paper “Long waves and cyclone waves” (Tellus 1, 33-52) gave meteorology its first rigorous theory of baroclinic instability – the mechanism by which the mean north-south temperature gradient generates the storms that propagate through midlatitudes. Charney’s 1948 paper “On the scale of atmospheric motions” (Geofys. Publ. 17, 17 pp.) derived the quasi-geostrophic system by a systematic scale analysis, showing which terms in the full hydrodynamic equations could be safely dropped for large-scale midlatitude motions.

Phillips read Charney 1948 as a graduate student and later said it was this paper that “awakened” his interest in dynamic meteorology. He explicitly cited it as the methodological template for his PhD thesis: “Charney presented this recipe for how to predict large-scale motions” (Phillips 1989 oral history).

Reginald Sutcliffe’s 1947 development theory of cyclones (UK Met Office) belongs to the same intellectual moment. Sutcliffe and Charney converged on equivalent formulations – vorticity-tendency equations using the geostrophic approximation – and Phillips’s two-layer model is a direct discretisation of their framework.

1.3 The von Neumann-Charney computational line

John von Neumann formulated the Meteorology Project at the Institute for Advanced Study in 1946. Charney took over the project in 1948; the team produced the first ENIAC forecast in 1950 (Charney, Fjortoft, von Neumann, “Numerical integration of the barotropic vorticity equation,” Tellus 2, 237-254). Phillips joined the IAS in October 1951, eight months after completing his PhD, and was assigned to write the barotropic-model code for the new IAS machine. He inherited not only the equations but the computational toolkit – relaxation methods, hexadecimal machine-code, explicit-counter DO-loops – on which the entire numerical weather prediction enterprise then depended.

Phillips’s role inside the IAS group needs emphasis: he was not the theorist of the team. Charney was. Phillips was the programmer who could turn quasi-geostrophic theory into running code on the IAS machine. That role – the one who actually makes the equations compute – became his characteristic mode throughout his career.

2. THE 1951 PhD THESIS: THE TWO-LAYER MODEL

2.1 Exact title and dimensions

The PhD thesis is, in full:

Phillips, N. A. (1951). “A Simple Three-Dimensional Model for the Study of Large-Scale Extratropical Flow Patterns.” Ph.D. dissertation, University of Chicago. 25 pp.

Two-five pages. That is not a typo. The dissertation is short – the entire substantive treatment of the two-layer quasi-geostrophic model fits in 25 pages. Phillips was 28 years old.

Advisor: George W. Platzman, who would later write the definitive 1979 BAMS retrospective on the ENIAC forecasts. Note that some secondary sources are confused about this. The original Phillips.md notes the advisor as “sometimes listed as George Platzman, though the primary sources consulted for this note do not explicitly confirm this.” Lewis 1998 Appendix is decisive: Phillips himself in the 1989 oral history (p. 13) explicitly identifies Platzman as his thesis advisor and praises him as “accurate as well as being right.” Master’s degree (1948) was under Erik Palmen on a synoptic problem (subsidence of a cold dome).

2.2 The published version: Phillips 1951 in J. Meteorology

The thesis is not the canonical Phillips 1951 publication. The published form is:

Phillips, N. A. (1951). “A simple model for the study of motions in baroclinic atmospheric flow.” Article in conference proceedings. I have not located an exact J. Meteorology publication matching the prompt’s “Two-Layer Geostrophic Modeling” title. The first major published version of the two-layer model appears as a section of Charney and Phillips 1953 (see below).

FLAG AS UNCERTAIN: The prompt asks me to verify a J. Meteorology 1951 publication titled “Two-Layer Geostrophic Modeling of the Atmosphere.” I have not been able to locate this in WebSearch or the canonical reference list in Lewis 1998. The Lewis 1998 reference list gives only the 25-page Chicago dissertation for 1951 (no journal version of the same content), with the next Phillips publication being Charney and Phillips 1953 in J. Meteorology. The thesis itself is what Charney saw on the blackboard at Chicago in late 1950 or early 1951.

2.3 The content: a 25-page leap

The dissertation describes “the simplest model which contains some of the essential features of the baroclinic atmosphere” – two superimposed layers of incompressible fluid of different densities, constrained between two rigid horizontal plates, conserving potential vorticity. The streamfunction in each layer evolves under the quasi-geostrophic vorticity equation; the layers communicate through hydrostatic adjustment of the layer interface.

The dissertation had two stated goals (Phillips 1951, abstract):

Demonstrate that a two-layer model can represent the motions of a continuously stratified baroclinic troposphere.
Make a numerical forecast of atmospheric flow patterns by forecasting the two-layer flow.

Phillips succeeded on both counts. He derived the two-layer baroclinic instability criterion – giving the threshold vertical wind shear for instability of small perturbations – before reading Eady’s 1949 paper. He then learned about Eady’s continuous-stratification result, compared the two, and found his two-layer instability boundary agreed well with the continuous case (Lewis 1998, footnote 6; Phillips 1989, pp. 10-11). Phillips explicitly credited Platzman, Charney, and Rossby for help overcoming the mathematical difficulties in the derivation.

The second part of the thesis applied the two-layer model to a specific case: the Thanksgiving Day storm of November 1950 – the Great Appalachian Storm. Phillips computed tendencies for sea level pressure and vertical velocity using hand-iterated relaxation (Southwell’s method), without an electronic computer. This was, in Lewis’s words, “the first numerical weather forecast using a baroclinic model of the atmosphere” (Lewis 1998, p. 57).

2.4 The thesis as template for what followed

What Charney saw on the Chicago blackboard, in late 1950 or early 1951, was the same equations that vertical-differencing of the continuous quasi-geostrophic system reduces to in the two-layer limit. He recognised this immediately. Charney’s own recollection in Lindzen, Lorenz, Platzman 1990:

“Phillips just outlined very briefly on the blackboard what he had done. And what he had done, of course, was to develop a two-layer model. It was somewhat inspired by the geostrophic approximation [Charney 1948], but I think it would be historically of considerable interest to know what led him to do that… In any case, I recognized that right away as the next step. I hadn’t proposed to do it in that way, but it turned out that the finite difference formulation in the simplest way, by vertical differencing of the quasi-geostrophic equations, one arrives at the same equations as one gets for a two-layer model.” (Lindzen et al. 1990, p. 54)

Charney offered Phillips a job at the IAS on the spot. Phillips arrived in October 1951.

The first IAS publication came two years later:

Charney, J. and N. A. Phillips (1953). “Numerical integration of the quasi-geostrophic equations for barotropic and simple baroclinic flow.” Journal of Meteorology 10, 71-99.

Charney and Phillips 1953 is the operational two-layer model – the same equations Phillips developed in his thesis, now coded for the IAS computer, and integrated forward in time. This is the paper that closes the loop between Phillips’s solo Chicago work and the IAS computational programme.

The two-layer quasi-geostrophic model became, over the next two decades, the standard pedagogical tool for atmospheric dynamics. Every introductory dynamic-meteorology textbook (Holton’s An Introduction to Dynamic Meteorology; Vallis’s Atmospheric and Oceanic Fluid Dynamics) develops the two-layer system as the simplest closed model exhibiting baroclinic instability. Phillips 1954 (“Energy transformations and meridional circulations associated with simple baroclinic waves in a two-level, quasi-geostrophic model,” Tellus 6, 273-286) extended the analysis to nonlinear effects and meridional circulations, providing the theoretical foundation for the 1956 GCM.

3. THE 1956 GCM (one chapter only – the full story is Post 7)

3.1 The canonical citation

Phillips, N. A. (1956). “The general circulation of the atmosphere: a numerical experiment.” Quarterly Journal of the Royal Meteorological Society 82(352), 123-164. Manuscript received 17 October 1955; published April 1956.

This is the paper that won the first Napier Shaw Memorial Prize of the Royal Meteorological Society (announced 20 June 1956), and is in retrospect the unambiguous birth certificate of climate modelling. It is described by Smagorinsky as opening “a new era” and by Lewis as having “an almost irreverent disregard for the complexities of the real atmosphere.”

3.2 The setup in five numbers

Two layers (250 mb and 750 mb)
17 by 16 grid points (E-W extent 6000 km, periodic; N-S extent 10000 km, walls)
About 500 numbers specifying the atmospheric state at any instant
31 days of simulated time after a 130-day spin-up
11-12 hours of IAS machine time for the 31-day forecast

The model was integrated on the IAS computer at Princeton (sometimes called MANIAC I, though the IAS engineers preferred just “the IAS machine”). 1024 words of Williams-tube memory (40 bits per word; about 5 KB). 2048 words on a magnetic drum (about 10 KB). Programmed in hexadecimal machine code. The state arrays consumed half the internal memory; the other half held the program.

3.3 What the experiment showed

Starting from a uniform isothermal atmosphere at rest, the heating function (linear in latitude, with ±0.23 °C/day at the walls, vanishing at 45°N) drove a 130-day spin-up to a baroclinically unstable mean state. A small random perturbation was then added, and the model was integrated for 31 days. Within 10-15 days a single large baroclinic wave developed with wavelength about 6000 km, moving east at 21 m/s. By day 20:

Mid-latitude jet at 250 mb of order 60-80 m/s
Surface winds in the alternating easterly-westerly-easterly pattern (trades, midlatitude westerlies, polar easterlies)
Three-cell meridional circulation: direct (Hadley) cells equatorward and poleward, indirect (Ferrel) cell in midlatitudes
Surface frontogenesis in the developing baroclinic wave without water vapour, without latent heat, without orography – a pure dry-dynamic result
An energy cycle: heating generates mean APE, eddies convert mean APE to eddy APE then to eddy KE, eddy momentum flux converges into the jet to generate mean KE, friction dissipates

After about day 25 the model blew up. This was the discovery that fed Phillips 1959 (next section).

3.4 Why this and not Mintz?

Yale Mintz at UCLA had published earlier work on observational general-circulation statistics (Mintz 1951 Tellus 3, 195-200; Mintz 1955) – mean angular-momentum transports estimated from observational data. Mintz also began GCM development at UCLA in 1961 with Akio Arakawa, and the Mintz-Arakawa GCMs (1961 onward) are the canonical UCLA line.

What makes Phillips 1956 the “first GCM” and not Mintz:

Three-dimensional time-evolving simulation, not statistics: Phillips integrated forward in time; Mintz 1951 was diagnostic from observations.
Hemispheric domain, not zonal-mean: Phillips’s channel was 17x16 in latitude and longitude; Mintz’s earlier observational work computed zonal-mean fluxes.
A coherent demonstration of an integrated energy cycle: Phillips simulated all the major large-scale features – jet, surface winds, three cells, fronts – emerging dynamically from a near-trivial initial state with simple forcing.

Mintz’s contribution to GCM history is real, but it comes later (1961 with Arakawa) and acknowledges the Phillips precedent. Arakawa explicitly: “I myself was also extremely inspired by Phillips’ work.” (Lewis 1998, quoting Arakawa’s 1997 letter.)

3.5 The “first” qualifier and the 1955 Rossby paper at IAS

The framing of Phillips 1956 as “the first GCM” should be qualified: it is the first published computer simulation of the time-evolving three-dimensional general circulation. Earlier computational work (Charney, Fjortoft, von Neumann 1950) was barotropic and 24-hour forecast, not climate. Earlier observational work (Bjerknes, Palmen, Starr, Mintz, Priestley) was statistics-from-data, not simulation. So Phillips occupies a unique slot.

One important contextual note: Rossby presented work at the IAS in 1956 on the general-circulation problem – not a GCM, but a theoretical synthesis. Rossby visited Princeton from Stockholm and engaged with Phillips’s results. Lewis 1998 reconstructs (via Wiin-Nielsen 1993) the discussion in which Rossby asked Phillips whether the fronts in his simulation were “really there” and accepted Phillips’s claim that they emerged purely from dynamics, not from latent heat or clouds. Rossby died in August 1957, about 16 months after Phillips’s QJRMS paper.

4. THE 1957 SIGMA-COORDINATE PAPER

4.1 The reference

Phillips, N. A. (1957). “A coordinate system having some special advantages for numerical forecasting.” Journal of Meteorology 14(2), 184-185. (Two pages. Journal was renamed Journal of the Atmospheric Sciences in 1962; the 1957 paper is now indexed under JAS.)

4.2 The definition

Phillips defined the new vertical coordinate as:

σ = p / p_S

where p is pressure and p_S is the surface pressure. σ ranges from 0 at the top of the atmosphere to 1 at the surface, with σ = 1 everywhere the surface is, regardless of orography. In modern notation, sometimes the convention has σ = (p − p_T) / (p_S − p_T) where p_T is a fixed model-top pressure; the original Phillips form has p_T = 0.

4.3 Why it matters: orography without coordinate intersection

The conventional vertical coordinate used in early NWP was pressure – isobaric surfaces. Phillips and others had been using p directly in the 1950s. The problem: pressure surfaces are not parallel to the ground over mountains. The 1000-mb surface intersects the Rockies; the 700-mb surface intersects the Tibetan Plateau. To carry a pressure-coordinate model around an orographic obstacle, you need to interpolate “below ground” or apply complicated boundary conditions.

Phillips’s sigma coordinate eliminates this problem definitionally: σ = 1 is always the surface. Topographic boundaries become coordinate surfaces. The price is that the vertical advection operator now involves the time-evolution of surface pressure, so the equations gain extra terms; but those terms preserve mass and energy conservation, and the boundary-condition simplification more than compensates.

The 1957 paper is a brief technical communication – two pages – but its operational consequence is enormous. Within a decade, sigma coordinates became the default for global atmospheric models. The NMC LFM, the NMC Nested Grid Model (1985), the early ECMWF model (1979), the UK Met Office model, the JMA models, the GFDL GCMs – all used sigma. The 2.5° 9-layer sigma coordinate global primitive equation model that became operational at NMC in 1974 (Shuman, 1989) is a direct application of Phillips 1957.

4.4 The drawbacks and the hybrid-coordinate succession

Sigma has a fundamental geometric drawback: the coordinate surfaces tilt over mountains. This means horizontal pressure gradients require computing differences of two large terms (the pressure-gradient component along the sigma surface and a correction from the tilt), and the cancellation between them suffers from numerical error over steep orography. For very large mountains – the Himalayas, the Andes, the Rockies – pressure-gradient errors in pure sigma models can dominate the dynamics.

The standard remedy emerged in 1981:

Simmons, A. J. and D. M. Burridge (1981). “An energy and angular-momentum conserving vertical finite-difference scheme and hybrid vertical coordinates.” Monthly Weather Review 109(4), 758-766.

Simmons-Burridge introduced the hybrid sigma-pressure coordinate at ECMWF. The hybrid scheme blends pure sigma in the lower troposphere with pure pressure in the upper atmosphere; in modern ECMWF practice, the coordinate transitions smoothly from terrain-following to isobaric somewhere above 200 hPa. This combines the orographic convenience of Phillips’s sigma with the cleaner pressure-gradient behaviour of an isobaric upper coordinate.

(The prompt asks about “Simmons-Strufing 1981” – this is actually Simmons-Burridge 1981. Strufing is not the author. The 1981 paper is one of the most-cited papers in numerical-modelling history.)

The eta coordinate, introduced by Fedor Mesinger in 1984 and operationalised in NMC’s Eta model (June 1993), takes a different approach: it replaces the smooth-terrain σ with a discrete step-mountain representation, keeping coordinate surfaces strictly horizontal except across the steps. Eta eliminates the pressure-gradient problem at the cost of vertical resolution near complex topography.

Modern operational systems use a variety:

ECMWF IFS, Met Office UM, NCEP GFS: hybrid sigma-pressure (Simmons-Burridge descent)
NCEP Eta/NAM: step-mountain eta (Mesinger)
GFDL FV3 (current): a Lagrangian hybrid sigma-pressure
Ocean models (POM, ROMS): pure sigma is still standard for coastal/regional work

Every one of these traces back to Phillips’s two-page 1957 communication.

5. NONLINEAR COMPUTATIONAL INSTABILITY (Phillips 1959)

5.1 The reference

Phillips, N. A. (1959). “An example of non-linear computational instability.” In B. Bolin, ed., The Atmosphere and the Sea in Motion (Rossby Memorial Volume). Rockefeller Press, pp. 501-504.

This is the paper the prompt may have confused with a 1974 paper. There is no 1974 Phillips paper on this topic; the 1959 Rossby Memorial chapter is the canonical reference. The Rossby Memorial Volume was compiled to honour Rossby on his 60th birthday and published shortly after his death; Phillips’s chapter is among the most-cited contributions.

5.2 The problem

Phillips’s 1956 GCM run blew up after about 25 simulated days, even though the model satisfied the von Neumann linear stability criterion (CFL: time step less than grid spacing divided by wave speed). The blow-up was real – the energy increased without bound – and the cause was not obvious. Phillips initially suspected truncation error from finite differencing and tried adding lateral eddy viscosity to damp the smallest scales; this delayed but did not prevent the catastrophe.

Three years of investigation produced the diagnosis: nonlinear computational instability (NCI), caused by aliasing of unresolvable short waves into the resolvable spectrum.

5.3 The mechanism

The quasi-geostrophic vorticity equation contains nonlinear advection terms – products of velocity gradients with vorticity. When two waves of wavenumber k₁ and k₂ interact through such a product, they generate waves at the sum and difference wavenumbers k₁+k₂ and

k₁-k₂

. On a discrete grid of spacing Δx, the maximum wavenumber that can be represented is k_max = π/Δx – the Nyquist wavenumber.

If k₁ + k₂ > k_max, the resulting wave cannot be resolved by the grid. Instead, it is aliased – mapped onto a resolvable wavenumber 2k_max - (k₁ + k₂). The aliased wave is fictitious. It does not exist in the continuous physical system. But the discrete equations propagate it forward, where it interacts again, generating still more aliased waves. Energy accumulates at the shortest resolvable scales. The simulation explodes.

This is not the linear instability that von Neumann analysed. The von Neumann criterion governs the time-stepping of single waves; aliasing is a structural problem of the spatial discretisation in nonlinear systems.

5.4 Phillips’s solution and Arakawa’s successor

Phillips’s fix: transform the grid-space solution into a Fourier series, truncate everything above k_max/2, transform back. This dealiased the grid representation – since quadratic interaction now only generated wavenumbers up to k_max, no aliasing was possible. With this filter, Phillips’s 1956-style GCM could be integrated indefinitely.

This is the spectral approach’s first foothold in atmospheric science. Phillips did not yet have a fully spectral model – he was using grid-point computation with periodic Fourier filtering – but the logic of spectral truncation as a dealiasing procedure was established. Bourke 1972 (Australian Numerical Meteorological Research Centre) and Orszag 1970 developed full spectral methods that exploit the same insight.

The grid-point alternative emerged in 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, 119-143.

Arakawa introduced an energy- and enstrophy-conserving finite-difference scheme for the Jacobian operator. This guaranteed that the discrete operator inherits the conservation properties of the continuous system, which means aliasing cannot drive energy upward in scale without a corresponding cascade downward. Arakawa’s scheme does not eliminate aliasing – it remains present – but it controls its consequences.

The Phillips-1959 vs. Arakawa-1966 divide is the structural choice for every later atmospheric modelling group:

Spectral models (ECMWF, JMA, NCEP global, Australian BMRC): pursue Phillips’s dealiasing logic through spectral truncation
Grid-point models (NCAR CCM/CESM lineage, GFDL, UCLA Mintz-Arakawa): use Arakawa-conserving schemes

Both lineages descend from Phillips 1959. The discovery of NCI is arguably as important as the 1956 GCM itself for the practical development of numerical modelling. Steve Easterbrook (2019) calls it “the foundational discovery for all subsequent atmospheric modelling.”

6. THE PRIMITIVE-EQUATION INITIALISATION PROBLEM (Phillips 1960)

6.1 The reference

Phillips, N. A. (1960). “On the problem of initial data for the primitive equations.” Tellus 12(2), 121-126. Manuscript received 18 September 1959; written while Phillips was at MIT (he had moved from IAS in 1956).

6.2 The problem

The primitive equations (the full hydrostatic Euler equations, with no quasi-geostrophic filtering) admit fast gravity-inertia waves alongside the slow meteorologically significant Rossby modes. If you initialise the model from observed wind and pressure fields that are not in geostrophic balance, the model immediately generates large-amplitude gravity-inertia oscillations that obscure the meteorological signal.

Hinkelmann (1951) had shown that simply using geostrophic initial winds reduces noise, but not enough. Charney (1955) used the balance equation – a higher-order constraint relating the streamfunction to the geopotential – and obtained noise-free forecasts. Phillips’s 1960 paper analyses why.

6.3 The result

For a linear barotropic shallow-water system, geostrophic initial conditions still admit gravity-inertia waves with amplitude of order ε (the Rossby number; about 0.1 in midlatitudes). The amplitude can be driven to ε² by also specifying that the divergence is non-zero and matches the divergence implied by the quasi-geostrophic ω-equation. Phillips writes (1960, p. 123):

“The initial wind field must contain some divergence if the gravity-inertia waves are to be satisfactorily eliminated from the forecast.”

This was the theoretical reason balance-equation initialisation worked. Phillips extended the linear analysis to nonlinear forecasts using a quasi-geostrophic expansion, defining initial conditions for two cases: (a) only geopotential observed; (b) only the non-divergent part of the wind observed. In both cases, the implied initial tendencies preserve the assumed geostrophic structure.

6.4 The descendants

Phillips’s 1960 paper is the foundational paper for atmospheric initialisation theory. Three lineages descend from it:

Normal-mode initialisation (Machenhauer 1977, Baer 1977, Daley 1979): project the initial state onto the eigenmodes of the linearised primitive equations and zero-out the fast modes. This was the standard initialisation method at ECMWF and Met Office from the late 1970s through the 1990s.
Digital filter initialisation (Lynch and Huang 1992): integrate the model forward and backward through the initial time, applying a low-pass filter to remove gravity-inertia oscillations. Lynch became the standard for both ECMWF (early 1990s) and later operational systems.
Variational data assimilation (3DVar then 4DVar): the modern approach embeds Phillips’s balance constraints into the cost function, treating balance not as a post-hoc correction but as a prior on the analysis.

Peter Lynch’s 2006 textbook The Emergence of Numerical Weather Prediction devotes a chapter to Phillips 1960 as the origin point of all initialisation work.

7. PHILLIPS 1963: THE COMPREHENSIVE REVIEW

Phillips, N. A. (1963). “Geostrophic motion.” Reviews of Geophysics 1(2), 123-176.

This is a 54-page synthesis of the entire quasi-geostrophic theoretical apparatus – scale analysis, derivation of the QG system, energetics, geostrophic adjustment, Rossby-wave theory, applications to numerical forecasting and to Gulf Stream meanders. It became, alongside Holton’s 1972 textbook, the standard reference for graduate students learning QG dynamics.

Phillips distinguished two types of geostrophic motion:

Type 1: horizontal scale L much smaller than Earth’s radius. Classical QG, beta-plane derivable. Standard midlatitude dynamics.
Type 2: horizontal scale L comparable to Earth’s radius. Planetary geostrophic motion. The Coriolis parameter varies significantly and the standard beta-plane breaks down; one has to retain the spherical geometry.

Type 2 PG dynamics is the framework later used for planetary-scale climate theory and for oceanic large-scale circulation work (Pedlosky’s 1979 textbook, Vallis’s 2006 textbook). It is the closest Phillips comes to a global-dynamics framework – and it is consistent with the limitations he had to live with in 1956 (channel model, beta-plane).

The 1963 review also contains the clearest summary in the open literature of the 1956 GCM’s theoretical foundations by their author. Phillips 1963, Section 9 (“Applications”) covers the GCM, the initialisation problem, the nonlinear instability, and the sigma coordinate – effectively a state-of-the-art survey of his own work to that date.

8. TROPICAL AND EQUATORIAL DYNAMICS (1960s)

Phillips’s contributions to tropical wave theory are less prominent than his midlatitude work, and a careful search yields fewer canonical papers. The major equatorial wave theory belongs to Matsuno (1966) (“Quasi-geostrophic motions in the equatorial area,” Journal of the Meteorological Society of Japan 44, 25-43), who derived the full spectrum of equatorial trapped modes (Kelvin waves, mixed Rossby-gravity waves, equatorial Rossby waves, equatorial inertio-gravity waves) on the equatorial beta-plane.

Where Phillips fits in:

Phillips 1963 Reviews of Geophysics Section 8 covers planetary-geostrophic motion (Type 2) and applies the framework to the equatorial region. Phillips notes that the equatorial Kelvin wave – a purely zonal mode trapped near the equator by Coriolis – is a member of the slow geostrophic family in Type 2 sense. Matsuno (1966) refers to Phillips 1960 in the context of methods for filtering gravity oscillations from primitive-equation initial data, but Matsuno’s derivation is his own.
No major Phillips paper on equatorial waves comparable to Matsuno 1966 has been identified in the WebSearch results. Phillips’s principal continuing interest was the midlatitude general circulation and operational NWP, not tropical dynamics.
The 1985 paper by Charney’s student Stevens and others (“Quasi-balanced dynamics in the tropics”) – a system that uniquely includes equatorially trapped Kelvin waves with consistent budgets for KE and PV – belongs to the MIT lineage that grew out of Phillips and Charney’s department, but Phillips himself does not appear as author.

FLAG AS UNCERTAIN: The prompt asks about “Phillips’s work on tropical waves and equatorial inertio-gravity oscillations in the 1960s” and “long-wave studies.” I have not located substantial Phillips publications in tropical wave dynamics in 1960s. His major output in that period was on the general-circulation problem (the 1963 review, mid-1960s GFDL collaborations), not on equatorial dynamics specifically. If Post 43 needs to discuss tropical dynamics, the canonical figures are Matsuno (1966), Holton (1970s onward at U. Washington), and later Wallace, Yanai, Webster – not Phillips.

9. THE MIT YEARS 1956-1974

9.1 The “package deal” and arrival at MIT

Henry Houghton, head of MIT’s Department of Meteorology, recruited both Jule Charney and Norman Phillips in the summer of 1956. The two arrived together. Charney had been at IAS leading the Meteorology Project; Phillips had been on his staff for five years. They were the package deal Houghton thought MIT needed.

The MIT department in the late 1950s included:

Henry Houghton (department head until 1958)
Jule Charney (1956-1981)
Norman Phillips (1956-1974)
Edward Lorenz (joined 1948, full faculty by 1955)
Victor Starr (faculty since 1947; the general-circulation observational programme)
Hurd Willett (long-range forecasting)
Frederick Sanders (synoptic dynamics)
Reginald Newell (joined late 1950s)

By the 1960s this was, without serious challenger, the leading academic department of atmospheric science in the world.

9.2 Department leadership

The sequence of department heads tells the story:

1970-1974: Norman Phillips
1974-1977: Jule Charney
1977-1981: Edward Lorenz
1981-1983: Carl Wunsch (oceanography; the department merged into PAOC)

Phillips was the transitional head through the period when the Department of Meteorology was being reorganised into the larger Department of Earth, Atmospheric and Planetary Sciences (the renaming and merger came later but the institutional groundwork was 1970-1974).

9.3 Phillips’s MIT students

The exact roster of Phillips’s doctoral students is not easily reconstructed from secondary sources. WebSearch and the obituaries do not provide a complete list. Known associations:

Phillips supervised several of the early data-assimilation theorists who later went to NMC and NCAR
Kerry Emanuel (who would later become MIT’s most prominent post-Charney/Lorenz dynamicist) was a student of Charney, not Phillips, but the broader MIT lineage descends from both
The MIT students who went on to lead the operational centres (NMC, ECMWF) generally trained with Charney or Lorenz; Phillips’s strongest line of descent is through his NMC years rather than his MIT years (Section 10)

9.4 The 1958 Appalachian Storm paper

Phillips returned to the 1950 Thanksgiving Day Storm – the case study from his PhD thesis – in a 1958 paper:

Phillips, N. A. (1958). “Geostrophic errors in predicting the Appalachian storm of November 1950.” Geophysica 6, 389-405. Published in a special issue commemorating Erik Palmen’s 60th birthday.

Phillips revisited Charney’s three-level model results and showed that “Charney’s three-level computations did not have the magic he thought they did, but that the essence of the cyclogenesis was already present in the two-level model, hidden by error due to the artificially high geostrophic wind in the initial upper-level trough” (Phillips 1997 personal communication, quoted in Lewis 1998). This was Phillips correcting his former boss in print. The exchange was civil; Charney and Phillips remained close professionally and personally until Charney’s death in 1981.

9.5 Quasi-geostrophic theory: the synthesis years

Through the 1960s, Phillips’s work at MIT consolidated quasi-geostrophic theory. The 1963 Reviews of Geophysics article is the canonical statement. Phillips also taught the QG dynamics course at MIT for years, training a generation of theoretical meteorologists. His teaching style was distinctive: working directly from the equations, demanding mathematical precision, and – in keeping with Platzman’s influence on him – valuing accuracy as much as correctness.

10. THE NMC YEARS 1974-1988

This is the bridge to Post 44 (Optimal Interpolation) and arguably the period that defined the modern operational forecasting enterprise. Phillips was 51 when he left MIT for NMC in July 1974, took the title Principal Scientist of the Development Division of the National Weather Service’s National Meteorological Center in Suitland, Maryland, and remained until his retirement in 1988 at age 65.

10.1 The NMC he joined: Fred Shuman’s centre

NMC under Fred Shuman (director 1964-1981) was running:

The Six-Layer Primitive Equation (6LPE) model for global guidance
The LFM (Limited-area Fine Mesh) for regional US guidance, operational from September 1971
A Cressman (1959) successive-correction objective analysis for data preprocessing
Twice-daily forecast cycles

The transition from quasi-geostrophic to primitive-equation models was complete; the transition from Cressman’s local-influence successive correction to statistical objective analysis (optimal interpolation) was just beginning.

10.2 Phillips’s role

Phillips was hired as the theoretical anchor for the Development Division. He did not run NMC – Shuman did, then Eugenia Kalnay later (1987-1997). Phillips’s job was to provide scientific direction for the major model and analysis developments, to publish technical reports establishing the theoretical basis, and to train the next generation of NCEP scientists.

His most direct technical contribution was the Nested Grid Model.

10.3 The Nested Grid Model (NGM)

Phillips, N. A. (1979). “The nested grid model.” NOAA Technical Report NWS 22. US Department of Commerce, NOAA, National Weather Service, Silver Spring, Maryland. April 1979.

Phillips’s NGM was a hemispheric primitive-equation model with one or two interior rectangular nested grids of successively finer resolution. The technical specification:

Outer (A) grid: polar stereographic, hemispheric, coarse mesh
Inner (B) grid: rectangular nest of finer mesh inside A, fully interactive (two-way nesting)
Innermost (C) grid: even finer, optionally
Vertical coordinate: σ = 1 − p/H (a variant of Phillips’s own 1957 sigma)
Numerical scheme: second-order accurate finite-difference, two-step Lax-Wendroff
Physics: gravity-wave damping at the Himalayas (the model’s first major numerical challenge); equatorial-symmetry boundary at the southern edge

NGM is two-way nested – not a static one-way embedding. After each outer-grid time step, the inner grid’s boundary points are interpolated from A; after each inner-grid sub-step, A’s interior points near the nesting boundary are updated from B. This was algorithmically novel in 1979 and required careful treatment of the interpolation operators to avoid generating spurious waves at the nest boundary.

Operational implementation came in March 1985, as the regional forecast component of the Regional Analysis and Forecast System (RAFS). The development team was J. E. Hoke, N. A. Phillips, G. J. DiMego, J. J. Tuccillo, and J. G. Sela (Hoke et al. 1989). Note the team membership: this is the cohort of NCEP scientists who would dominate operational NWP through the 1990s.

The popular nickname at NMC for the NGM was “Norm’s Great Model” – a play on the NGM initials and Phillips’s first name. The nickname appears in the MIT News obituary and several other retrospectives. It is affectionate; the operational forecasters liked the NGM, and they associated it with Phillips personally.

NGM operational lifetime: March 1985 to 3 March 2009 (when its Model Output Statistics products were finally discontinued, post-retirement of the model itself in the late 1990s). The Eta model superseded it in 1993 for primary regional guidance.

10.4 Data assimilation: from Cressman to OI

NMC’s analysis system in 1974 was Cressman’s (1959) successive-correction method – a sequence of grid-point corrections each weighted by a radius-of-influence function. Cressman is local, ad-hoc in its weighting, and does not propagate observation information across long correlation lengths.

The transition to Optimal Interpolation (OI) – which Lev Gandin had developed in the USSR in 1963 – happened slowly at NMC. The operational implementation paper is:

McPherson, R. D., K. H. Bergman, R. E. Kistler, G. E. Rasch, and D. S. Gordon (1979). “The NMC operational global data assimilation system.” Monthly Weather Review 107(11), 1445-1461.

This was the first operational global OI scheme at NMC, updating a 9-level primitive equation model at 6-hour intervals through a local, multivariate, three-dimensional statistical interpolation directly in the model’s vertical coordinate. Bergman 1979 (“Multivariate analysis of temperatures and winds using optimum interpolation,” MWR 107, 1423-1444) is the companion theoretical paper.

Phillips’s role: he was not co-author on the 1979 McPherson-Bergman paper, but he was the senior scientist whose office signed off on the development direction. The Development Division’s strategic decision to move from Cressman to OI was made under Phillips’s scientific leadership.

His most cited contribution to the OI theory came later:

Phillips, N. A. (1986). “The spatial statistics of random geostrophic modes and first-guess errors.” Tellus A 38(4), 314-332.

This paper analysed the statistical structure of forecast errors in operational OI – a critical theoretical input because OI requires a model for the error covariance of the first-guess (background) field. Phillips hypothesised that “a data assimilation system with excellent forecast and analysis components and frequent access to good data should have first-guess errors similar to an ensemble of random slow modes with equipartition of energy.” He then computed the covariance statistics implied by this hypothesis on a constant-f plane.

Key findings (from the abstract):

First-guess errors have wind and geopotential variance increasing with elevation
The horizontal velocity scale of the error field increases with elevation
The vertical correlation is sharper for wind than for geopotential
The “separable mathematical models used to represent correlations in operational practice misrepresent some important features”

This last point is critical. Operational OI in the 1980s used separable correlation models – the horizontal and vertical correlations were factored into independent functions. Phillips 1986 showed that the slow-mode equipartition assumption produces fundamentally non-separable correlations, and that real first-guess errors observed over North America fit the non-separable model better than the separable one.

This is Phillips’s signature theoretical contribution to operational data assimilation. It is the closest he came to publishing an “operational” result – a theoretical paper that directly informed how NMC’s OI system was tuned through the 1980s and 1990s.

Re. “the Norman Phillips correction”: The prompt asks me to verify this term. WebSearch returned mixed results – some references to “Phillips correction” in the context of forecast-error covariance modelling (sometimes attributed to the 1986 Tellus paper), others to economic Phillips curves (a different person). The 1986 Tellus paper is the canonical reference; the “correction” language refers to how vertical correlations should be adjusted away from the simple separable assumption to match the slow-mode equipartition prediction. The Bannister 2008 review in QJRMS on forecast-error covariance statistics treats Phillips 1986 as a foundational reference. So yes, there is a meaningful “Phillips correction” to vertical correlations, and it lives in this 1986 paper. The prompt’s framing is accurate but the formal name is rarely used.

10.5 The spectral GFS transition (August 1980)

The transition from grid-point to spectral global modelling at NMC happened in August 1980 with the implementation of Joseph Sela’s spectral model. This is the model that became the GFS (Global Forecast System). The first operational version was:

Hydrostatic spectral dynamic core
R30 triangular truncation (about 375 km equivalent grid spacing)
12 vertical layers
Sigma coordinate (Phillips 1957)
Limited physics package

Phillips’s role: This is sometimes attributed to Phillips because he was Principal Scientist of Development during the period. The actual technical work was Joseph Sela’s; Phillips’s role was sponsorship and theoretical oversight. Sela 1980 (Monthly Weather Review) is the canonical implementation paper; Phillips is not co-author. But the transition happened under his scientific direction, and the spectral approach – of which Sela’s GFS is the operational descendant – is conceptually a descendant of Phillips’s 1959 NCI dealiasing argument.

10.6 Phillips’s NMC students and trainees

The cohort that Phillips trained or worked with at NMC went on to dominate American operational NWP:

Eugenia Kalnay (NMC Director 1987-1997; ECMWF, NASA, U. Maryland)
Stephen Lord (NMC/EMC Director through 2010s)
Geoff DiMego (NMC, then EMC director; NGM team member)
John Derber (variational data assimilation at NMC/EMC)
Jordan Alpert (operational NMC scientist; spectral model development)
Joseph Sela (spectral GFS; NMC scientist 1975-2010)
Ronald McPherson (NMC director after Shuman; the 1979 OI paper’s lead author)

This is the NCEP lineage. Through it, Phillips’s influence runs into every operational US numerical forecast made today. The NMC scientists trained or supervised by Phillips are now responsible for the GFS, NAM, HRRR, GEFS, and every other NCEP operational model.

10.7 The Norman Phillips Award

The American Meteorological Society’s Norman Phillips Award was established in 2018-2019, around the time of Phillips’s death. It recognises “outstanding contributions to the advancement of operational numerical weather prediction.” This is distinct from the AMS Award for Outstanding Contribution to the Advance of Applied Meteorology, which Phillips himself received during his career.

FLAG AS UNCERTAIN: I have not been able to locate the AMS Norman Phillips Award’s establishment date or detailed citation language in WebSearch. The award’s existence is referenced in the MIT obituary and AMS pages, but the exact founding year and the recipient list have not been directly verified. If Post 43 needs this detail, a direct query to AMS may be required.

11. LATER WORK AND THE POST-NMC PERIOD (1988-2019)

11.1 The Charney memoir (1995)

Phillips, N. A. (1995). “Jule Gregory Charney.” Biographical Memoirs (National Academy of Sciences) 66, 80-113.

When Charney died in June 1981, the NAS commissioned Phillips to write his biographical memoir. Phillips delivered a substantial 33-page essay that is, in its way, as much a self-portrait of Phillips’s scientific worldview as it is of Charney’s. The memoir traces Charney’s path through UCLA, the IAS, MIT; details the ENIAC forecast and its consequences; and covers Charney’s later work on monsoon dynamics, ocean circulation, and climate sensitivity.

Read against Lewis 1998 (which Phillips also helped write through extended correspondence), the Charney memoir reveals Phillips’s lifelong conviction that the mathematical physics of the atmosphere – not the empirical or descriptive synoptics – is the central explanatory framework. He returns again and again to Rossby waves, quasi-geostrophic theory, baroclinic instability. He is appreciative of synoptic meteorology but he is unsentimental about what it can and cannot do. Charney’s contribution, in Phillips’s reading, is the demonstration that mathematical physics can predict atmospheric behaviour.

11.2 Continued AMS service

Phillips remained active in the AMS through the 1990s and 2000s:

Editor, Journal of the Atmospheric Sciences (co-editor for periods)
Honorary Member of AMS
AMS Awards committee service (specifics not documented in available sources)

11.3 The Foucault pendulum paper

In 2005 – not the prompt’s claimed 2014 – Phillips published:

Phillips, N. A. (2005). “What Makes the Foucault Pendulum Move among the Stars?” In M. R. Matthews, C. F. Gauld, and A. Stinner, eds., The Pendulum: Scientific, Historical, Philosophical and Educational Perspectives. Springer, Dordrecht.

Phillips was 82 years old at publication. The paper is a careful pedagogical treatment of the Foucault pendulum problem: why the pendulum’s plane of oscillation appears to rotate, what reference frame is operative, and the connection to atmospheric Coriolis dynamics. This is the same physics that underlies geostrophic motion – the inertial-frame versus rotating-frame distinction that motivates the Coriolis force. Phillips’s lifelong interest in the foundations of rotating-fluid dynamics surfaces one final time.

The 2014 paper the prompt asks about, “at age 90,” may actually be this 2005 chapter or a similar later writing. The MIT obituary states Phillips “published his final academic paper at age 90, on the Foucault pendulum, demonstrating lifelong intellectual curiosity.” Age 90 would be 2013 or 2014. I have not located the exact 2013-2014 paper. FLAG AS UNCERTAIN.

11.4 The 1990 Cocke memorial

The prompt asks about “the 1990 Cocke memorial.” I have not been able to identify this. WebSearch did not return relevant results for any “Cocke memorial” associated with Phillips in 1990. The Phillips colleague Steven Cocke – a research scientist at Florida State University working on tropical dynamics in the 1990s and 2000s – is alive and not a memorial subject. There is Norman Cocke, an actor who died in 1990, but no scientific connection. FLAG AS UNCERTAIN: this prompt query may be in error.

11.5 Textbook contributions

Phillips never wrote a stand-alone textbook. His 1973 chapter “Principles of Large Scale Numerical Weather Prediction” in P. Morel, ed., Dynamic Meteorology (Springer) is the closest he came; it is a chapter-length treatment of the theoretical and computational foundations of NWP, drawing on his 1963 Reviews of Geophysics synthesis but emphasising the operational rather than theoretical side.

The 1963 Reviews of Geophysics article is, in practice, the Phillips textbook – the synthesis that every later graduate student read.

12. INTELLECTUAL DESCENT

12.1 The MIT lineage

Through the MIT department, Phillips’s intellectual descent runs through:

Edward Lorenz (department head 1977-1981; “discoverer” of chaos with the 1963 Lorenz attractor). Phillips and Lorenz were close colleagues for 18 years at MIT (1956-1974). Lorenz’s 1963 paper used a low-order spectral truncation of a 2D convection problem; the truncation methodology descends conceptually from Phillips’s 1959 spectral filtering of nonlinear instability. The post-Phillips MIT students who picked up the Lorenz chaos programme – predictability theory, attractor analysis, ensemble forecasting – inherit Phillips through Lorenz.
Kerry Emanuel (PhD 1978 under Charney; MIT faculty 1981-present). Emanuel’s work on tropical-cyclone potential intensity, convective parameterisation, and the maximum potential intensity (MPI) theory is the direct continuation of the MIT theoretical-dynamics line. Emanuel was not Phillips’s student directly, but he is a direct lineal descendant of the MIT department Phillips helped build.
Carl Wunsch (MIT physical oceanography). Wunsch’s ocean-inverse-problem and adjoint-method work is a different intellectual path, but the rigorous mathematical-physics tradition is unmistakably the MIT line.

12.2 The NMC/NCEP lineage

The NMC trainees who would dominate NCEP operations from 1990 onward all worked with Phillips in his Development Division years:

Eugenia Kalnay: NMC Director 1987-1997; later one of the foundational figures in ensemble Kalman filter data assimilation. The MERRA, NCEP/NCAR reanalyses and the modern data-assimilation infrastructure descend from Kalnay’s NMC tenure.
Stephen Lord: NMC/EMC Director 2003-2010; oversaw the modern GFS development.
Geoff DiMego: NMC scientist; the operational engineer behind the regional modelling line (NGM through Eta to NAM).
John Derber: 3DVar implementation at NMC; the operational data-assimilation system through the 1990s.
Joseph Sela: spectral GFS development from 1975 until his death in 2010.

The current GFS, NAM, GEFS, HRRR, RAP – every NOAA operational model – runs sigma coordinates (Phillips 1957), uses dealiasing or energy-conserving schemes (Phillips 1959 / Arakawa 1966), employs balance-equation initialisation theory (Phillips 1960), and uses statistical objective analysis with first-guess covariances structured along Phillips 1986 lines.

12.3 Citations in modern textbooks

James R. Holton, An Introduction to Dynamic Meteorology (4th ed., 2004): cites Phillips 1956 in the general-circulation chapter; develops the two-layer quasi-geostrophic system in the basic dynamics chapters; uses sigma coordinates as the standard pedagogical vertical coordinate.

Dennis L. Hartmann, Global Physical Climatology (1st ed., 1994; 2nd ed., 2016): cites Phillips 1956 as the first GCM in the climate-model chapter; reproduces the Phillips energy cycle in the energetics chapter.

Geoffrey K. Vallis, Atmospheric and Oceanic Fluid Dynamics (2nd ed., 2017): cites Phillips’s 1963 review as the canonical synthesis of geostrophic motion theory; develops the QG system using Phillips’s formulation.

David J. Stensrud, Parameterization Schemes (2007): cites Phillips 1959 on nonlinear computational instability as the foundational reference for energy-conserving discretisations.

Eugenia Kalnay, Atmospheric Modeling, Data Assimilation and Predictability (2003): cites Phillips 1956, 1957, 1959, 1960, and 1986; treats Phillips as the unifying figure of mid-twentieth century NWP theory. Kalnay’s textbook is, in its way, an extended tribute by a former NMC trainee.

13. THE KEY FIGURE FOR POST 43

If Post 43 needs a single sentence summarising Phillips beyond the 1956 GCM:

Phillips invented the two-layer quasi-geostrophic model (1951), wrote the first GCM (1956), invented sigma coordinates (1957), diagnosed nonlinear computational instability and dealiasing (1959), founded primitive-equation initialisation theory (1960), wrote the canonical synthesis of quasi-geostrophic theory (1963), built the Nested Grid Model (NGM, 1979/1985), provided the theoretical basis for operational Optimal Interpolation (1986), and trained the next generation of NCEP scientists who run American operational NWP today.

Or, more compactly:

Phillips is the figure through whom theory becomes operational practice in numerical weather prediction. He is the bridge from Charney’s IAS programme to NCEP’s current operations.

14. UNCERTAINTIES FLAGGED

The exact title and J. Meteorology publication status of Phillips 1951. The 25-page Chicago dissertation is canonical; whether it was also published as a separate journal article remains unverified.
The 1990 “Cocke memorial” reference – not located.
The “Norman Phillips Award” at AMS – exists but exact founding date not verified.
The exact 2013-2014 final paper “at age 90” – the 2005 Foucault pendulum chapter is canonical but a later 2013-2014 paper may exist.
The full roster of Phillips’s MIT PhD students – not systematically documented in available secondary sources.
The specific extent of Phillips’s authorship vs. supervisory role in NMC operational documents (1974-1988) – partially documented but not exhaustively.

15. PRIMARY SOURCES TO CONSULT FOR POST 43

These are the references Post 43 should cite directly:

Phillips 1951 (PhD thesis, U Chicago, 25 pp.)
Charney and Phillips 1953 (J. Meteor. 10, 71-99)
Phillips 1954 (Tellus 6, 273-286) – the theoretical predecessor of the 1956 GCM
Phillips 1956 (QJRMS 82, 123-164) – the GCM paper (covered in Post 7)
Phillips 1957 (J. Meteor. 14, 184-185) – sigma coordinates
Phillips 1958 (Geophysica 6, 389-405) – Appalachian storm revisited
Phillips 1959 (Rossby Memorial Volume, pp. 501-504) – NCI
Phillips 1960 (Tellus 12, 121-126) – primitive-equation initialisation
Phillips 1963 (Rev. Geophys. 1, 123-176) – comprehensive review
Phillips 1973 (in Morel ed., Dynamic Meteorology) – principles of NWP
Phillips 1979 (NOAA Tech Rep. NWS 22) – the Nested Grid Model
Phillips 1986 (Tellus A 38, 314-332) – first-guess error statistics
Phillips 1995 (NAS Biographical Memoirs 66, 80-113) – Charney memoir
Phillips 2005 (Springer chapter) – Foucault pendulum

Secondary sources:

Lewis 1998 (BAMS 79, 39-60) – the definitive Phillips retrospective
Smagorinsky 1983 (Advances in Geophysics 25, 3-37) – early recollections of NWP
MIT News obituary, 16 May 2019
Franklin Institute Benjamin Franklin Medal citation (2003)
Easterbrook 2019 memorial
Lynch 2019 ThatsMaths tribute
Phillips’s 1989 oral history (NCAR Archives, 85 pp.)

Michał Brennek