«Medium-Range Weather Prediction Austin Woods Medium-Range Weather Prediction The European Approach The story of the European Centre for Medium-Range ...»
In retrospect, it is worthwhile recognising that the Centre was able to avoid becoming either another international bureaucracy, or an ivory tower research institute. Above all, it was planned as an operational scientific and technical institute. Under Wiin-Nielsen’s admirable leadership, it developed into just that. Simmons remembered that “in working atmosphere it was more like a university research department than a national weather service”.
Partly this was because all the staff were newcomers. Everyone was learning to work with others in a complex mixture of nationalities. The Centre was fortunate to be able to recruit many good young scientists, the best in Europe in their fields: meteorology, numerical methods, modelling and data assimilation, computer science and technology, telecommunications, international administration, and related disciplines. This provided all the virtues of vigour, passion and pluralism that such a blend of nationalities and skills can bring. In the first years, there was a great deal of social intermingling between nationalities, both informally and — after the restaurant in the new building became available in late 1978 — semi-formally. For example, there were the “International Social” evenings, when staff of different nationalities brought national food dishes to be shared, with sometimes music and dancing to follow. One scientist arriving from France noted the “beautiful melting-pot effect, the first in European science”.
Undoubtedly the size of an organisation matters. An organisation is made up of individuals, all with different kinds of responsibility. For an organisation of up to perhaps 150 or so, everyone in principle can quickly recognise everyone else, and have an idea of what they do and their role. This was especially important in the early days, when all were learning and experimenting in one way or another. Many were for the first time working with people from other countries, with different work and social habits and expectations. Friendly and productive lateral discussions between staff in the three departments of Administration, Operations and Research were normal, often informally over coffee or lunch in the Centre’s restaurant after the move from Bracknell to the new building at Shinfield Park.
74 Chapter 7It is worth trying to cover the story of these years with a broad brush, and to exemplify the intellectual effort and dedicated hard work with some details. In the chapter “The first Director”, we covered some of these matters from Wiin-Nielsen’s point of view. We will avoid unnecessary repetition.
Staff were recruited. The Convention came into force in November 1975, and staff then in post were given new contracts from January 1976.
Number of staff in post on 31 December of the years 1976 to 1979.
The table of staff requirements for 1980 showed 145 posts; however, of these, four were suppressed following a review of the Administration Department in that year. The number of staff stabilised at about 140 for many years to come.
The Centre was set up with a clear focused objective: to produce the best
medium-range weather forecasts in the world. The timing was fortunate:
there was a massive storehouse of research and development from the 1960s from the United States as well as Europe that could be tapped — sources of data, analysis techniques, different numerical schemes, and more.
A forecast begins from an analysis of the current state of the atmosphere, which has assimilated many different kinds of observational data, from ships, land stations, balloons, satellites and other sources.
Selection of a good data assimilation system was crucial. Lennart Bengtsson, the Head of Research, was well connected to the best sources of scientific advice. He had highly relevant experience from his work preparing for the First GARP Global Experiment (FGGE), which we have considered in Chapter 3. We have noted the explicit reference to the Global Atmospheric Research Programme (GARP) in the introduction to the Convention. By the time of the first session of the Council in November 1975, GARP was underway. Wiin-Nielsen noted in his Report to the Council that “FGGE as part of GARP is essential for the Centre. FGGE happens to take place at the time the Centre will be ready to start operational forecasting. We hope to have an opportunity to participate in FGGE — it will give the best initial state to start operational forecasting.” 1974 to 1980: the Formative Years 75 Looking back, we can see that Bengtsson, with his GARP experience, exercised good judgement in making important strategic decisions that put the Centre on the right path immediately. He decided to use a global model, not a hemispheric or regional system. The analysis would be based on “three-dimensional Optimum Interpolation”. This used the statistics of past observations of temperature, wind, humidity and so on to ensure best use the current observations, and to ensure that neighbouring observations were in conformity with one another; we consider this further in Chapter 8. In addition, he was able to recruit the right people to do the work.
An early suggestion was made that the Centre, being a medium-range forecasting centre, should use global analyses already being produced at other major European short-range forecast centres. The Centre could then devote its research and development efforts, and computing resources, to developing the numerical scheme and physics of its forecasting model.
This approach was firmly rejected. Although with some appeal at first glance, it was argued that the Centre would need to do its own analyses if it was to make best use of the data expected to be available from satellites.
This turned out to be a vitally important decision. It meant that the Centre had from the beginning complete control over its entire system. In the event, a significant proportion of the improvement that the Centre achieved in its medium-range forecasts has been due to its sophisticated analysis system. Over the years, of the computing resources used for the operational forecasts, about 40% has been devoted to producing the most accurate starting point for the forecasts, and 60% to producing the medium-range and ensemble forecasts. In 2005, about 30% of the computing resources used in daily operations were for assimilating the data, 20% for two high-resolution “deterministic” forecasts, and 50% for two runs of the Ensemble Prediction System.
A computer model was required. We have seen in the first Chapter that Wiin-Nielsen contacted two groups in the USA, who were well advanced in terms of model building: Dr Joseph Smagorinsky, the Director of the Geophysical Fluid Dynamic Laboratory (GFDL), and Professors Yale Mintz and Akio Arakawa of University of California at Los Angeles (UCLA).
Smagorinsky was a visionary who played an important leadership role during FGGE. His paper at the joint American Meteorological Society/Royal Meteorological Society meeting in London in 1969 set out GFDL’s agenda for 20 years. He was also chairman of the Joint Steering Committee, which was leading GARP and planning FGGE. Smagorinsky knew Wiin-Nielsen well, and had worked with Bengtsson on the Working Group for Numerical Experimentation. Smagorinsky and Arakawa agreed to provide the Centre with copies of their model codes.
76 Chapter 7Tony Hollingsworth had joined the Centre on 1 March 1975, as the second member of the new Research Department, under Lennart Bengtsson, its Head of Research. There were just eight names in the telephone list at that time — Aksel Wiin-Nielsen, Jean Labrousse, Lennart Bengtsson, Ernest Knighting, and Jim Clarke, with secretarial staff, Jane Khoury, Martine Russell and Jill Llewellyn. Hollingsworth had had a long interview with Wiin-Nielsen in December 1974. For the first hour or more, he felt that the interview was going badly. Hollingsworth then realised that they had been discussing his work at MIT for the three years to 1970, followed by his year at Oregon State University, and then his period as a founding fellow of the UK Universities Atmospheric Modelling Group at the University of Reading. This was all good stuff, but was not in line with Wiin-Nielsen’s determination that the Centre would not become an ivory-tower research centre. At last reading Wiin-Nielsen’s body language, Hollingsworth stressed his three-year stint as a bench forecaster at Shannon Airport in his native Ireland, making operational forecasts for the public as well as the special forecasts needed for aviation. Suddenly the tone of the interview lightened. Wiin-Nielsen offered him a post.
Further scientific staff joined on 1 May including David Burridge, Roger Newson, Robert Sadourny — on six month’s leave from CNRS in France — and Zavisa Janjic. The Centre at this time was housed on the top two floors of the Social Security Office at Fitzwilliam House in Bracknell.
Robert Sadourny had spent two periods at UCLA, at the department led by Prof Mintz, first as a student in 1965-66, later as a visitor in 1969. Mintz was making major contributions to the science, combining theory, diagnostic analysis, and modelling across a broad range of interests. Mintz’s career lasted more than four decades, and included work on analysis and modelling of the planet’s general circulation, planetary atmospheres, stratospheric ozone transport and ocean circulation. Much of his scientific work involved collaborations with an unusually talented array of younger scientists.
As we saw in Chapter 1, Sadourny spent four weeks at UCLA in 1975 investigating the UCLA model code for its suitability for the Centre’s work.
Hollingsworth was sent to GFDL to pick up Dr Kikuro Miyakoda’s forecasting version of the GFDL code. Miyakoda supplied Hollingsworth with an informal documentation of his model in April, which Hollingsworth studied, together with all available information about the GFDL model. He visited GFDL for six weeks from early June. The weather in Princeton was sweltering. Some consolation came from a telephone call from Burridge back in Bracknell where he mentioned that it was snowing. A few weeks earlier, temperatures of 30°C in Bracknell had melted the insulation on the 1974 to 1980: the Formative Years 77 card-reader that provided access to a CDC 6600 at a computer centre in Rijswijk, Netherlands!
On arrival at GFDL, Smagorinsky handed Hollingsworth over to the care of Miyakoda, whose successful experimental medium-range forecasts published a few years earlier — and shown on page 33 — were influential in the decision to set up the Centre. In his first conversation, Miyakoda noted that the ECMWF initiative was extremely important for the future of numerical weather prediction. In his opinion, if ECMWF succeeded, that success would open many doors for the future development of meteorology. On the other hand, if ECMWF failed, those doors would be closed for meteorology for decades to come. Hollingsworth’s mission was vitally important to GFDL.
In September 1975, Miyakoda would be one of the principal lecturers at the Centre’s seminar on “Scientific Foundations of Medium-Range Weather Forecasts”. He would review the existing methods of modelling physical processes of the atmosphere in mathematical terms, and the numerical procedures for making the forecasts.
Miyakoda and his associates showed Hollingsworth several cabinets full of listings of their codes — model code, GATE data assimilation code, field interpolation codes, diagnostic and verification codes, graphics codes and more. This rich library of meteorological knowledge, the result of some tens of skilled man-years of intellectual effort by some of the masters of the science, was offered to Hollingsworth. He was overwhelmed.
This openness was characteristic of most US federally funded science then and since. Software and data developed or collected with federal funds were essentially in the public domain. The Centre received the software free and essentially without conditions. Over a cup of coffee, Smagorinsky explained why there were no conditions on the software, and why he was so free with the model: his policy was to distribute the model to any scientific institute that would have available the computer power required to run it at sufficiently high resolution. Smagorinsky’s only requirement was that the Centre should acknowledge GFDL in any work done with their software, and should not pass it on to third parties without GFDL’s consent.
Hollingsworth gratefully accepted copies of the model, interpolation, diagnostic, and verification codes, but decided to concentrate most of his effort on the model. Miyakoda, together with his associates Lou Umscheid and Joe Sirutis, helped Hollingsworth formulate a work-plan for his visit.
His objective was to bring back the GFDL model source code, and modify it so that it could run on the CDC 6600 being installed at John Scott House, Bracknell. At the time GFDL was using the Texas Instruments Advanced
78 Chapter 7Scientific Computer (ASC). Hollingsworth wanted an initial dataset which, although at low resolution, would be sufficient for test purposes, and two ten-day forecast runs, at differing resolutions. Jim Walsh was GFDL’s main computer expert. Hollingsworth needed his help to get the forecasts with all their special write-ups through the ASC. Walsh’s disposition was sunny and positive, but when Hollingsworth outlined his work-plan, he shook his head.
There was no way he or Hollingsworth could get through the work in less than six months, much less in six weeks.
Hollingsworth got started, relying heavily on Sirutis, Umscheid and Walsh to get the computing done. He was living in an ancient army base close to downtown Princeton, sharing a wooden apartment with Carlos Mechoso. There was no air-conditioning in his room, so he was happy to work 14-hour days in the luxury of the air-conditioned GFDL offices. His work proceeded apace.
Smagorinsky, Miyakoda, and others at GFDL thus provided a major impetus to getting the Centre operational. Their practical help and generosity in providing their model in 1975 was of great importance to the Centre in its planning, software design, and scientific development.