Meteorologist, oceanographer, Python solvem probler, experimenter, Raspberry Pi tinkerer, geek dad, manager, space enthusiast, ham radio lover, science columnist, ADHDer, synthetic intellect curious, linux admin cosplayer.

NMC at Suitland, 1974-1981: The Late Shuman Era

Operational picture of the U.S. National Meteorological Center during Frederick G. Shuman’s last seven years as Director. Focus: the IBM 360/195 era, the LFM and PE 6-layer models, the 1980 transition to spectral, the 1981-1983 Cyber 205 procurement, and the data assimilation cycle.

1. Computer hardware at FOB-4

What was there in early 1974

When the period opens, NMC was running its operational suite on a CDC 6600. The 6600 had been installed in 1965-1966 and enabled the first global primitive-equation model run at NMC in June 1966 [WPC History; EMC History]. By 1974 it was nine years old.

The IBM 360/195 question

The single most-cited NMC hardware fact for this period is also one of the muddiest in the secondary sources. The WPC’s own institutional history is unambiguous on one point: “especially significant gains were noted with the introduction of the high resolution PE model on an IBM 360/195 in 1978” [WPC History, p. 4]. This is the canonical NMC-side statement.

Count. Post 31 of the blog mentioned “three 360/195s.” This appears to be an overstatement. Wikipedia records that “about 20 Model 195 systems were produced” over the 1969-1977 production run, and only NOAA, Argonne National Laboratory, and Oak Ridge are named as US customers [IBM 360/195 Wikipedia]. NMC was certainly an early-to-mid customer but not all of the production. The contemporaneous WPC text uses the singular “an IBM 360/195”; the 1986 NMC Product Suite document refers to “the IBM/195” in the singular [WPC History; Deaven 1984; Western Region Technical Attachment 86-05]. The most defensible claim is two machines (one for production, one for development/backup), with weaker evidence for three but no authoritative documentary support I have found. Flag this as an open question; if Post 31 says three, the post should be hedged.

Production status. IBM’s Model 195 was released on 20 August 1969 and withdrawn from marketing on 9 February 1977 [IBM 360/195 Wikipedia]. So when NMC installed its 360/195 in 1978 (per WPC), IBM had ceased manufacturing the line a year earlier. NMC’s 360/195(s) were therefore either among the last produced or were refurbished machines from elsewhere in the IBM service base. The Model 195’s introductory price was $7-12.5 million, with monthly rental of $165,000-$275,000 [IBM 360/195 Wikipedia, citing IBM brochures]. I have not located documentary evidence on whether NMC owned or leased. IBM continued field service into the early 1980s on the installed base; the British Rutherford Appleton Laboratory 360/195 ran until October 1982 and was decommissioned then, suggesting IBM maintained the line for roughly five years after withdrawal from sales [Chilton Computing history of the 360/195].

Successor. The 360/195 was a re-implementation of the older 360/91 in monolithic integrated circuits, with a 54-nanosecond CPU cycle and up to seven simultaneous operations [IBM 360/195 Wikipedia]. It was IBM’s last entry in the “supercomputer” class until the mid-1980s 3090 series; in performance terms it was comparable to the CDC 7600 (which NMC did not buy) but with a more conventional architecture.

Where physically. FOB-4 (Federal Office Building 4) at the Suitland Federal Center, Suitland, Maryland. NMC’s operational and forecast staff moved out of FOB-4 to the new World Weather Building at 5200 Auth Road, Camp Springs, Maryland, in January-February 1975 (the dedication was 22 October 1974; the WPC Forecast Division moved on Wednesday, 19 February 1975, the day after Presidents Day, on the testimony of employee Bill McReynolds) [WPC History; VLab World Weather Building]. But the central computers were too heavy to move; the NMC’s giant computers are too large to be moved. Their output will be transmitted two miles electronically to the World Weather Building with no loss in efficiency. [VLab World Weather Building]. The mainframes stayed in FOB-4 Suitland until 1999, when an IBM SP cluster was installed at a new site in Bowie, Maryland [WPC History].

The Cyber 205: the late-Shuman procurement decision

NMC took delivery of a CDC Cyber 205 vector supercomputer in summer 1983 [Deaven 1984]. This procurement was decided under Shuman (who retired in 1981) and executed under his successor William D. Bonner. The Cyber 205 was Control Data’s answer to the Cray-1; CDC announced the 205 in 1980 and the first customer, the UK Met Office, took delivery in 1981 [Cyber 205 IT History Society]. NMC was among the next batch of customers.

Operational dates on the Cyber 205, from the contemporary 1984 conference paper by Dennis Deaven (then at NWS) and from the December 1985 NMC Monthly Performance Summary [Deaven 1984; NMC Product Suite via Western Region TA 86-05]:

Component Cyber 205 operational
LFM (Limited-area Fine-mesh) 30 August 1983
FNL (Final analysis-forecast cycle) August 1982 (configuration date) - some confusion in dates; the Cyber 205 hosted the FNL by 1983-84
Global Spectral Model upgrade R30 -> R40 October 1983
Movable Fine Mesh (MFM) hurricane model 1983 (port from IBM/195)
RAFS (NGM-based Regional Analysis & Forecast System) 27 March 1985
MRF (40-wave, 18-layer global with full physics) 17 April 1985

The Cyber 205 LFM ran in 75 seconds (48-hour forecast), about 15 times faster than the IBM/195 LFM, which took about 1125 seconds (the Cyber 205 paper says “about a month and a half” of effort by a single meteorologist-programmer to port and vectorize) [Deaven 1984].

Backup machine. When the Cyber 205 was down (which appears to have been not infrequent), backup versions of the models ran on the NAS 9000 (a National Advanced Systems IBM-360 compatible) at considerably reduced speed [NMC Product Suite 1985]. The NAS 9000 was NMC’s successor to the 360/195 for the non-vector workload, taking over the I/O-heavy parts of the production suite from the 195. Exact NAS 9000 installation date is unclear from the sources I found.

2. The models 1974-1981

LFM (Limited-area Fine Mesh) - the U.S. workhorse

The LFM was NMC’s first regional model dedicated to North America. Implementation: 1971 [EMC regional models table]. The original specs:

  • Grid: 190.5 km on a polar-stereographic projection true at 60 degrees North, half-bedient
  • Domain: North America and adjacent waters
  • Vertical: 6 sigma layers (boundary layer + 3 tropospheric + 3 stratospheric)
  • Initial forecast length: 24 hours
  • Computer: CDC 6600 in 1971, then transitioning to IBM 360/195 in 1978

The model was built primarily by Joseph P. Gerrity and colleagues; the canonical reference is Gerrity (1977) The LFM Model – 1976: A Documentation, NOAA Technical Memorandum No. 60, NWS, NMC, 68 pp. The model was substantially Howcroft’s design at the origin (1969-1970) with Gerrity making it operational; subsequent versions are documented in Newell and Deaven (1981) The LFM-II – 1980, NOAA Tech Memo No. 66 [Western Region TA 86-05].

Upgrade chronology [EMC regional models table; Western Region TA 86-05]:

Year Configuration Forecast length
1971 190.5 km, 6 layers 24 hours
1975 190.5 km, 6 layers 36 hours
1976 190.5 km, 6 layers 48 hours
1979 127 km, 7 layers (LFM-II) 48 hours
1981 190.5 km, 7 layers, 4th-order numerics 48 hours

The cycle was run twice daily (00Z and 12Z) by the mid-1970s.

The LFM was operationally superseded by the Nested Grid Model (NGM) in 1985 [Hoke et al. 1985; EMC regional models]. The LFM itself “was frozen in 1986” (i.e., further development stopped, though it kept running for verification baseline) [Kalnay 2003 Fig. 1.1.1] and was retired in February 1996 [EMC regional models].

PE 6-layer (Shuman-Hovermale): the hemispheric workhorse

The hemispheric primitive-equation model, the canonical Shuman-Hovermale model, became operational in mid-1966 on the CDC 6600 [Shuman & Hovermale 1968 Journal of Applied Meteorology 7(4):525-547; cited via WPC History and EMC history]. Original specs:

  • Grid: NMC octagonal grid, 381 km true at 60 degrees N
  • Vertical: 6 sigma layers
  • Domain: most of the Northern Hemisphere
  • Equations: primitive (hydrostatic) – a clean break from the prior quasi-geostrophic/filtered models

This was the first NMC operational primitive-equation model and the workhorse for hemispheric forecasts through the 1970s. The “high resolution PE model on an IBM 360/195 in 1978” mentioned by the WPC institutional history is the upgraded version of the Shuman-Hovermale 6-layer model running on the new hardware [WPC History]. EMC’s global models table shows successive versions:

Year Resolution Layers Forecast length
1974 2.5 degrees 8 12-h (FNL)
1976 2.5 degrees 9 12-h (FNL), discontinued 8/1980
1977 3.75 degrees 3 (! - this may be a reduced version for global) Global to 48-h; NH to 252-h

These were operational gridpoint PE models; the 1980 transition (below) replaced them with the spectral model. Some uncertainty: the EMC table is terse; the 3-layer 1977 figure may be an extended-range special configuration rather than the operational primary.

The 1980 spectral transition

In August 1980 the NMC Global Spectral Model went operational under the leadership of Joseph G. Sela [Sela 1980 Monthly Weather Review 108:1279-1292; NMC TPB #282]. Configuration:

  • Truncation: rhomboidal 30 (R30, equivalent to about 375 km at the equator)
  • Vertical: 12 sigma layers
  • Forecast: 48 hours
  • Hardware: IBM 360/195
  • Lineage: descended from Australian (BMRC) and NCAR spectral work; Sela had been developing it at NMC since 1975

The motivation was the chronic problem of pole-convergence in the gridpoint hemispheric models: as the polar-stereographic grid is bent onto a sphere, the time-step must shrink (CFL condition). Spectral methods avoid this entirely by expanding the prognostic variables in spherical harmonics. By the late 1970s, Bourke at BMRC, Hoskins at Reading, and others had shown spectral methods were competitive with gridpoint at moderate truncations. NMC’s gamble was to commit to spectral as the primary global formulation.

The model was a single global forecast, retiring the older hemispheric octagonal grids. From August 1980, NMC’s global forecasts were spectral. The R30L12 configuration ran on the IBM 360/195 until 1983, when the Cyber 205 enabled the upgrade to R40L12 in October 1983 [NMC Product Suite 1985; Deaven 1984]. Joseph Sela is remembered as “the father of NOAA’s Spectral Model” [MetForum.org Sela seminar abstract].

MRF (Medium-Range Forecast) and AVN (Aviation)

The MRF system as a distinct medium-range run was a 1985 development (operational 17 April 1985, on the Cyber 205, 40-wave 18-layer global with full radiation and improved physics) [NMC Product Suite 1985]. Prior to 1985, “medium range” was handled by extending the standard global forecast. The AVN run (Aviation, the morning global forecast to 72 hours) and the MRF run (Medium-Range, to 240 hours) became distinct elements of the production suite in the 1983-1985 reorganization.

NGM (Nested Grid Model) – brief mention

Norman Phillips developed the NGM at NMC, documented as NOAA Technical Report NWS 22, 1979 [Phillips 1979]. It was operational from 1985 at 80 km resolution, 16 layers, replacing the LFM for U.S. operational regional forecasts. The “RAFS” (Regional Analysis and Forecast System) wrapped the NGM with a regional optimum interpolation analysis and went operational on 27 March 1985 [NMC Product Suite 1985]. The NGM was nicknamed Norm’s Great Model [Phillips Wikipedia and obit]. It was retired in March 2009.

MFM (Movable Fine Mesh) hurricane model

The MFM was NMC’s operational hurricane track model: 10 sigma layers, 60 km grid, the unique feature that the grid moves to keep the hurricane at the center, with boundary conditions refreshed each step from the contemporary global forecast [NMC Product Suite 1985]. It first became operationally available in 1975 [WPC summary]; the GFDL/Princeton hurricane research line stretches back to Yoshio Kurihara’s work in the early 1970s, but the NMC operational version was a separate development by John B. Hovermale (Shuman’s PE co-author), D. G. Marks, and S. H. Scolnik [Hovermale et al. 1977]. The 1977 paper is the formal documentation; the model ran operationally through the 1970s-1980s and was ported to the Cyber 205 in 1983.

The model “first became operationally available (on an earlier machine) in 1975 and has undergone a number of evolutionary changes since then” [NMC Product Suite 1985].

3. Data assimilation 1974-1981

From four-times-daily to the current GDAS architecture

The four-times-daily analysis cycle (00Z, 06Z, 12Z, 18Z) at NMC predates the period in question; NMC had been running four cycles per day since the late 1960s to support both the U.S. domestic forecast cycle and the global cycle. By the mid-1970s the configuration was: two principal cycles at 00Z and 12Z driving the full forecast suite, plus 06Z and 18Z analyses primarily as updates to the first-guess for the next principal cycle. This system formalized into the Global Data Assimilation System (GDAS) by 1982 [NMC Product Suite 1985]: four six-hour cycles per day, each consisting of an optimum-interpolation analysis followed by a six-hour forecast, the latter providing the first guess for the next analysis. The current FNL run configuration “dates from August 1982” [NMC Product Suite 1985].

Cressman analysis -> Hough analysis -> optimum interpolation

The Cressman successive correction method (Cressman 1959) remained NMC’s primary analysis method into the early 1970s. The Cressman analysis applies successive corrections from observations to a first-guess field with decreasing scan radii.

Hough analysis (Flattery 1971) replaced Cressman as the operational analysis for some products in the early 1970s. Hough functions are the solutions to the linearized tidal equations on a rotating sphere; they form an orthogonal basis adapted to atmospheric motions. Flattery’s scheme used Hough functions as the interpolating basis on the sphere, which had the practical advantage of treating mass and motion fields self-consistently. By 1985 the Hough Analysis System remained NMC’s backup analysis for the Early Run when the optimum-interpolation system was unavailable; “NMC plans to drop this analysis and forecast from operations shortly” [NMC Product Suite January 1986].

Optimum interpolation at NMC: The seminal NMC paper is McPherson, Bergman, Kistler, Rasch, and Gordon, “The NMC operational global data assimilation scheme,” Monthly Weather Review (1979). This established the NMC operational OI analysis just in time for FGGE. The British Met Office line (Lorenc 1981 in MWR) was a parallel development. NMC’s full Global Optimum Interpolation (GOI) for the operational global system ran from 1979-1980 onwards [McPherson et al. 1979; Dey and Morone 1985]. The 1985 paper by Dey and Morone (MWR 113:304-318) documents the evolution from January 1982 through December 1983 – which is essentially the period when the GOI matured into the form used through the 1980s.

FGGE 1979

The First GARP Global Experiment (FGGE, in 1978-1979) was the single biggest event for global data assimilation in the period. FGGE deployed dropwindsondes, drifting buoys, geostationary satellite imagery, and the first concentrated effort to assimilate satellite temperature retrievals (TOVS/HIRS). NMC’s operational global OI was rushed into production specifically to handle the FGGE data stream. The legacy is twofold: the OI architecture (in place by 1979-1980), and the body of FGGE reanalyses that the international community produced from the 1979 special observing periods (SOP-1 January-March 1979 and SOP-2 May-June 1979). NMC was the U.S. analysis center for the FGGE; ECMWF was its European counterpart. FGGE pushed NMC harder than any other observational program in the period; the OI was operational essentially because FGGE demanded it.

4. Forecast products

Standard products

By the mid-1970s NMC was producing the daily suite that the U.S. weather enterprise depended on:

  • 500 hPa height analyses and prognoses (the primary mid-troposphere forecast diagnostic)
  • Mean sea level pressure analyses and forecasts (the surface forecaster’s primary input)
  • Precipitation forecasts from the LFM and PE models
  • Vertical profiles at standard mandatory levels (1000, 850, 700, 500, 400, 300, 250, 200, 150, 100 hPa)

The verification metric was the S1 score (Teweles and Wobus 1954), which measured the relative error in horizontal pressure gradient over North America. NMC’s S1 score for 36-hour 500-hPa forecasts dropped from about 60% in 1955 to about 35-40% by 1980, with continuing improvement [Kalnay 2003 Fig 1.1.1].

MOS (Model Output Statistics)

MOS originated at NWS’s Techniques Development Laboratory (TDL, the precursor of MDL) under H. Robert Glahn, with planning beginning in 1965 and operational forecasts from the technique first issued in 1968 [Glahn and Lowry 1972 Journal of Applied Meteorology 11:1203-1211; Wikipedia MOS]. MOS uses screening regression on a long sample of model forecasts paired with observed weather to derive station-specific probability forecasts (probability of precipitation, max/min temperature, cloud cover, etc.). MOS was operational throughout the 1974-1981 period and ran as a post-processing step on the LFM and PE model outputs. The 1972 Glahn-Lowry paper is the canonical reference. MOS is not a Klein paper – the question prompt named Klein/Glahn but the operational MOS at NMC was Glahn-Lowry. Klein refers to W. H. Klein’s earlier work on perfect-prognosis statistical forecasting (a precursor approach to MOS), not the operational MOS itself.

Hurricane track forecasts

Operational track forecasts from a numerical model began with the MFM in 1975 (per WPC summary and Hovermale et al. 1977). Prior to 1975, NMC supplied general guidance from the hemispheric PE model, and the National Hurricane Center forecasters used CLIPER (climatology + persistence statistical) and HURRAN-class techniques alongside the model. Sometimes 1978 is given as the date when the “first full-physics” model went operational; this likely refers to a substantially upgraded MFM, with the 1975 date being the first operational version with simpler physics.

5. The 1981 architectural pivot

Shuman retired in 1981. His successor, William D. Bonner, took over a center that had committed to two large bets that were now coming due:

  1. The Cyber 205 procurement (which Shuman had pursued – the Class VI supercomputer competition was a major NOAA-wide procurement in the late 1970s and early 1980s).
  2. The spectral model as primary global formulation (Sela’s R30L12 had been operational for about a year when Shuman retired).

Both bets paid off. The Cyber 205 came online in summer 1983; the spectral model scaled to R40 in October 1983 on the new hardware; and the RAFS/NGM-based regional system replaced the LFM in 1985. In retrospect, the 1980-1985 period at NMC is a single architectural sweep: spectral global + nested-grid regional + optimum-interpolation analysis on vector hardware.

Shuman’s role. Shuman was the in-house technical lead; the spectral model and the Cyber 205 procurement were both pushed by him through the late 1970s. He was famously prudent about operational risk; the spectral model went operational in August 1980 only after Sela had demonstrated parity with the gridpoint PE in parallel runs.

Cressman’s role. George Cressman, NMC’s first director (1954-1964), had become Director of the Weather Bureau / National Weather Service in 1965 (after the ESSA reorganization) and held that position until 1979. He was Shuman’s boss across the period and a strong supporter of NMC’s modernization. After 1979, his successors at NWS (initially Richard Hallgren) continued the support.

Procurement context. The Cyber 205 was a Class VI supercomputer in the U.S. federal procurement vocabulary of the early 1980s; alternatives were the Cray-1 (which most other federal centers chose: NCAR, ECMWF, GFDL) and the Cyber 203 (the 205’s predecessor). NMC’s choice of the Cyber 205 over the Cray-1 was driven partly by price/performance arguments and partly by the fact that CDC offered a more developed vector compiler at the time. The decision was contested at the procurement stage; the OMB process required justification. The exact procurement history would require digging into NOAA records I have not found online.

6. People

William D. Bonner (NMC Director 1981-1990)

Bonner came to NMC as Director in 1981, succeeding Shuman. He was an atmospheric scientist from UCLA’s Department of Meteorology (UCLA’s atmospheric sciences program); his most-cited research is Bonner (1968) “Climatology of the Low Level Jet,” Monthly Weather Review 96(12):833, the foundational paper on Great Plains nocturnal LLJ. He was very much a science-trained director rather than an operations director. Some sources list him only as “Director from 1981 to 1990” without his earlier UCLA affiliation; one search result attributes him to “Department of Meteorology at the University of California at Los Angeles.”

Ronald McPherson

McPherson is the lead author of the 1979 NMC operational global data assimilation paper that established the OI system; he later became NCEP Director (1990-1998) after Bonner. In the 1974-1981 period he was NMC’s lead data-assimilation scientist.

Norman A. Phillips

Phillips moved from MIT to NMC in 1974 (he had been MIT Department Head 1970-1974) and served as Principal Scientist of the NMC Development Division from 1974 to 1988 [Phillips Wikipedia; existing research file]. He developed the Nested Grid Model (NGM), NOAA Technical Report NWS 22, 1979, which became the basis for the 1985 RAFS. Phillips was the senior scientific authority at NMC throughout the late Shuman / early Bonner era. His move to NMC in 1974 is exactly aligned with the start of the period under study; he played a major role through the entire late-Shuman era.

John B. Hovermale (1939-1994)

Hovermale was the operational hurricane modeler at NMC. He co-authored the 1968 PE 6-layer paper with Shuman, then in the 1970s built the MFM (Movable Fine Mesh) hurricane model that became NMC’s first numerical hurricane track forecaster. He worked at NMC from 1965, briefly returned to Penn State faculty in the late 1970s, then back to NMC. He died 1994, age 55 [Washington Post obit summary]. He won the Commerce Department Gold Medal for the MFM.

George Cressman

Director of the National Weather Service 1965-1979 (i.e., the entire pre-1979 portion of our period). His objective analysis paper (Cressman 1959) was the foundation NMC analysis method into the mid-1970s. After retiring from NWS in 1979 he consulted for foreign weather services. He died in 2008.

Joseph G. Sela

The spectral modeler. Began spectral development at NMC in 1975; brought the model to operations in August 1980. He remained at NMC/NCEP through to retirement and is celebrated as “the father of NOAA’s Spectral Model.”

Eugenia Kalnay

Kalnay came to NMC later, as Director of EMC from 1987 to 1997. She is not a 1974-1981 figure. Her Atmospheric Modeling, Data Assimilation and Predictability (Cambridge, 2003) is nevertheless the standard textbook reference for NMC operational history through 1989.

7. Buildings

  • Federal Office Building 4 (FOB-4), Suitland Federal Center, Suitland, Maryland. Built by 1942 as part of the Census Bureau complex. Housed JNWPU from 1955, NAWAC from 1955, NMC from January 1958. Held NMC’s central computers continuously from 1955 to 1999.
  • World Weather Building (WWB), 5200 Auth Road, Camp Springs, Maryland. Dedicated 22 October 1974. NMC forecast operations moved here in January-February 1975 (the bulk move was Wednesday 19 February 1975). Computers stayed in FOB-4; output transmitted electronically two miles to WWB.
  • Bowie, Maryland: NMC computers moved to a new IBM SP cluster here in 1999.
  • Gaithersburg, Maryland: Replaced Bowie in 2002.
  • NOAA Center for Weather and Climate Prediction (NCWCP), College Park, Maryland: NCEP centers (including WPC, formerly HPC) moved here in August 2012.

The “Camp Springs” address persisted as NCEP’s headquarters from 1975 to 2012. The “Suitland” address persisted only for the computer site within FOB-4 from 1975 to 1999.

8. Open questions and uncertainties

  1. Exact number of IBM 360/195s at NMC. Primary sources I have found use the singular. Post 31’s claim of three should be verified or hedged. Best evidence is for one (and possibly two, with one as backup); evidence for three would require finding the original procurement documentation in NOAA records, which is not online.

  2. Exact installation date of the IBM 360/195(s) at NMC. The WPC summary says “high resolution PE model on an IBM 360/195 in 1978” – this anchors 1978 as the date the new hardware became operational for production. Hardware delivery and acceptance testing likely preceded by 6-12 months. The WPC summary also notes “each new system about 6 times more powerful than the one before” implying the 360/195 was a successor to the 6600 (which dates to 1965-1966). The decision and contract may date to 1976-1977 given IBM ended production in February 1977.

  3. Exact date the CDC 6600 was retired at NMC. Not located. The 6600 was likely run in parallel with the 360/195 during the transition (a standard NMC practice).

  4. NAS 9000 installation date. Not located. By the 1985 NMC Product Suite document the NAS 9000 was the secondary/backup machine. Probable installation: 1982-1983, contemporary with the Cyber 205.

  5. The 1977 EMC table entry of “PE 3-layer”. This may be an extended-range global configuration with reduced vertical resolution; not a primary operational model. Worth flagging in the post.

  6. Exact MFM operational date. Sources give both 1975 (WPC summary) and 1978 (some tropical cyclone sources, calling it “first full physics”). The likely interpretation: the original 1975 MFM had simpler physics, and 1978 marks a major upgrade.

  7. OI implementation date at NMC. McPherson et al. 1979 is the publication; the operational implementation was 1979 or 1980 in advance of FGGE Special Observing Period 1 (January-March 1979). Best evidence: operational implementation late 1978 or 1979, consistent with the McPherson et al. paper.

  8. Cyber 205 procurement timeline. The Cyber 205 was announced in 1980; UK Met Office took delivery 1981; NMC took delivery summer 1983. The procurement decision was likely 1981-1982 under Bonner, but the technical groundwork was laid under Shuman.

9. Summary timeline (the operational picture 1974-1981)

Year Event
1974 NMC has CDC 6600. Norman Phillips moves from MIT to NMC. World Weather Building dedicated 22 October 1974.
1975 NMC forecast staff move from FOB-4 Suitland to WWB Camp Springs (Feb 19). Computers stay at FOB-4. MFM hurricane model first operational. LFM extended to 36-hour forecasts.
1976 LFM extended to 48-hour forecasts. PE model upgraded to 9 layers.
1977 IBM ends production of 360/195 (9 Feb 1977). LFM Documentation Tech Memo 60 (Gerrity) published. Hovermale-Marks-Scolnik MFM paper published.
1978 IBM 360/195 installed at NMC. High-resolution PE model operational on the 360/195. MFM upgrade with full physics.
1979 Cressman retires as NWS Director. McPherson et al. operational global OI paper. FGGE begins (Special Observing Periods Jan-Mar and May-Jun). LFM-II upgrade to 127 km, 7 layers. Phillips’s NGM technical report NWS 22 published.
1980 NMC Global Spectral Model R30L12 operational (August), Joseph Sela lead. CDC announces Cyber 205.
1981 Shuman retires after 17 years as NMC Director. William D. Bonner becomes Director. LFM-II 4th-order numerics. Lorenc multivariate OI paper.

Sources