«Medium-Range Weather Prediction Austin Woods Medium-Range Weather Prediction The European Approach The story of the European Centre for Medium-Range ...»
In September 1977, Michel Jarraud, then a young scientist from Météorologie Nationale, France, attended a weeklong Seminar prepared by Centre staff on Physical Processes in Models. By the end of the week, he thought that “this must be the best laboratory [for meteorology] on the planet!” Lennart Bengtsson was on the lookout for capable scientists. He visited Paris in early 1978. Jarraud, who like Frédéric Delsol had studied under Daniel Rousseau, was working on spectral techniques in the group led by Michel Rochas, a scientist who played a significant but perhaps sometimes unrecognised role over the years in propagating the advantages of spectral techniques for numerical modelling. There were not many in Europe, or indeed elsewhere, working in this area of research. The model resolution at the time in Météorologie Nationale was very low, constrained as it was by the computing power available. Although Jarraud had published nothing at that time in the open literature, Bengtsson recognised Jarraud’s talent and ability, and he wanted to have the best in spectral expertise at the Centre. He opened Jarraud’s eyes to the vision of the computing power planned at the Centre, and otherwise enthusiastically presented a prospect of the future in medium-range prediction. He encouraged Jarraud to apply for a post.
Jarraud remembered trying without success to find Bracknell, then the temporary site of the Centre, on his large-scale map of England; he had “no idea” where in the UK it was! He telephoned his colleague Jean-Francois Geleyn, who was already at the Centre working with Michael Tiedtke and others on physical aspects of the model. Geleyn, who was also keen for Jarraud to join the team at the Centre, picked him up at Heathrow for his interview. Jarraud joined the Centre in June 1978, and stayed until the end of 1985 as a scientist working on spectral methods in the Research Department. Although the Centre has its three “Working Languages” of English, French and German, Jarraud soon recognised the need to improve his English. It was frustrating for him not to be able to express himself fluently in what was the common language for most day-to-day communication with
102 Chapter 9his colleagues. He went on to play a major role in the development of the Centre’s model, together with Adrian Simmons, who brought to the Centre his experience of working on spectral modelling with Brian Hoskins at Reading University. After four years back in Paris as Director of the national forecasting division in France, Jarraud returned to the Centre in 1990 as Head of the Operations Department, and became Deputy Director in 1991.
In Chapter 1 we saw that Aksel Wiin-Nielsen became Secretary-General of WMO in 1980. Twenty-four years later, in 2004, Jarraud was appointed to the same post. In early 1994, he was approached to allow his name to be put forward as Deputy Secretary-General of WMO. He would have liked to have stayed longer at the Centre, but noted that “you cannot always choose the ideal time”. The post would become vacant in January 1995. While it was “a big gamble” it would give him the opportunity to “do more for many more countries”. The Centre had 18 Member States, WMO more than 190 Members. The challenge was attractive. After serving as Deputy Director of WMO from 1995, he was elected Secretary-General from 2004. Thus, two of the five Secretaries-General of this important specialised agency of the United Nations had significant ECMWF background.
On arriving at the Centre in 1978, Jarraud worked with Fons Baede; his first task was to make the spectral code work on the Centre’s CRAY-1. Later he worked on the model resolution, and on comparisons between the performance of the spectral and grid-point models.
By the end of 1977, work on the spectral numerical technique had progressed: preliminary ten-day forecasts were being run with the spectral model, and compared with the operational grid-point model. “It seems to me”, wrote Fons Baede of the Research Department, “that the spectral model is mathematically and numerically more elegant than the grid point model”. However, he noted that the spectral model “still requires a network of grid points on the globe”.
Other improvements to the numerical scheme were in hand. Clive Temperton wrote a highly efficient Fast Fourier Transform, substantially reducing the number of computations needed to make the forecast. The “semi-implicit” version of the model further reduced the computational time to 25% of that required for the explicit version, with almost identical results — a 15-minute time step, instead of a 21⁄2-minute step of the explicit model.
In early 1979, the decision was made to use the semi-implicit scheme for all forecast experiments. A limited-area version of the model was tested over a region of the Northern Hemisphere.
Improving the modelling of clouds and other physical aspects was a priority task for the Research Department. In Chapter 7 we mentioned the The Medium-Range Model 103 1978–79 “Spring Experiments”, which tested two different versions of the model. By the end of 1978, 14 global experimental forecasts had been run to ten days to compare the physics of two models, one with the physics developed at the Centre, the other with the GFDL physics. As we saw in Chapter 7, the results showed that “the performance of the two schemes is very similar, and the calculation time is also approximately the same”.
Bengtsson decided that since “the possibilities of improving the ECMWF scheme are much larger and it is very likely, when a more realistic treatment of topography and of clouds and of albedo will be introduced, the higher degree of physical realism with the scheme developed at the Centre will prove to be better”, the Centre’s scheme would be used.
Already studies of “systematic” model errors, errors that would normally be undetectable in a single forecast but were identified by diagnosing model behaviour over long periods, were actively pursued. The systematic errors of the two “Spring Experiment” models were similar.
Proper representation of mountains in the model — the orography — was clearly needed. As well as reducing errors from inadequate representation of steep slopes, the distribution of the orography as it affects the large-scale model flow somehow has to be taken into account. Aspects to be investigated included the barrier effect of mountain ranges, the low-level drag slowing the air as it flows over the rough ground and the influence of gravity waves as they propagate up from the mountain ranges to affect the flow in the stratosphere.
By 1980, the physical parametrization had been improved. Convective heating was more realistically modelled, leading to reduction in an erroneous drift of the jet stream. The model’s boundary layer — the lower level that feels the effects of the surface below — had been improved. Better exchange coefficients for heat, moisture and momentum reduced erroneous creation of intense low-pressure systems. Investigation of the creation and dissipation of kinetic and available potential energy, energy conversions and transfers of heat, moisture and momentum globally and in defined geographical areas were important for identifying model errors.
The systematic errors in the forecasts became well organised and persistent after day five, with two maxima, one over northwest Europe, the other over Alaska. Substantial research went into understanding the causes of these, and minimising them. It became evident that tropical systems were not active enough in the model. Too much energy was transferred from equatorial regions into the cyclone belt, leading to a westerly circulation more intense than the observed.
104 Chapter 9Tests in 1980 showed that model orography strongly influenced prediction of blocking weather patterns over Europe, when a depression to the south and an anticyclone to the north block the westerly flow. The Alps in particular played a significant role in development of low-pressure systems in the Mediterranean. In April 1981, a more realistic representation of orography was introduced in the model.
In the year to September 1980, the spectral version of the model was run weekly, to give 53 model integrations that could be compared with the operational grid-point forecasts. The year-long trial ensured that the seasonal variability was taken into account. Claude Girard and Michel Jarraud summarised the results in a paper “Short and medium range forecast differences
between a spectral and grid point model”:
• The spectral model gave better forecasts.
• The differences, although small numerically, were synoptically significant.
• The systematic errors of the two were similar.
Overall, the spectral model gave an impressive six-hour improvement in forecast performance.
Jarraud later recalled the methodological approach at the Centre, unique in the meteorological world, to careful and exhaustive testing of research results. The Centre had the talented staff and the necessary tools to do its job of improving medium-range forecasting.
In co-operation with scientists at Météo France, Jarraud and Ulrich Cubasch, who later went on to the Max Planck Institute in Germany, ran the spectral model for a single six-year “forecast” starting from 15 November
1979. The purpose was not to attempt a forecast, and not even to see how the model “climatology” would compare to that of the real atmosphere.
Rather the experiment was designed to study the time variability of the model atmosphere in its most important aspects. In the event, it did pretty well; the annual cycle was a major feature, and even though the sea surface temperature was the same from year to year, the model proved its ability to simulate anomalous years.
There was a certain amount of “creative tension” at this time, leading to some heated discussions in the Research Department between the grid point and spectral teams. Jarraud remembered the friendly competition between the grid point supporters Burridge and Gibson on the one hand, and the spectral team on the other including himself and Simmons. Finally, in November 1980, “the only rational choice was made”, in the words of one of the spectral modellers. It was decided in principle to develop a new operational code based on the spectral method.
The independent scientists of the Scientific Advisory Committee (SAC) The Medium-Range Model 105 expressed concerns about how the spectral model would deal with steep mountains. Testing of higher horizontal and vertical resolution began in
1981. It was this work that lead to development of the envelope orography outlined below.
On 21 April 1983, the “new operational model” was introduced, the first operational forecast of the Centre based on the spectral code. As SecretaryGeneral of WMO, Jarraud displayed prominently in his office the charts of this first spectral forecast. More than 20 years later, he was still using the punch cards from the model as notepaper and bookmarks! It was a “T63 resolution” spectral model, that is, with “triangular truncation” at total wave number 63, meaning that it could resolve 63 waves in the atmosphere around a great circle on the globe. Thus, weather systems with wavelengths down to about 700 km were computed. Sub-grid-scale processes were computed at the grid points of what was now referred to as the “Gaussian” grid.
The Gaussian grid was a latitude/longitude grid in which the spacing of the latitudes was (almost!) regular. It had a “hybrid” vertical coordinate with 16 levels and a revised time-stepping scheme.
Simmons later recalled how this was “an exciting period of really productive research” when he, an Italian scientist Stefano Tibaldi, visiting scientists Ed Lorenz from MIT and Mike Wallace from the University of Washington, Michel Jarraud and others, were running an intensive programme of experiments on many aspects of research such as predictability, model performance, and representation of orography.
By 1985, atmospheric models based on spectral techniques had taken over from their finite difference predecessors in many operational and research institutes.
The mean orography of the earth was used at the beginning. Studies showed that effects of mountain barriers were being systematically underestimated. More generally, a marked sensitivity to the orography was found in experiments. For example, formation of cyclones in the lee of the Alps was improved if the model orography was artificially raised. An “envelope orography” was developed, and used operationally from April 1983. The mean orography was raised by adding √2 times the standard deviation of the very small-scale orography as measured by satellites to the grid-square mean orography. Objective comparisons of forecasts made with and without envelope orography of important winter situations had shown that the model had been improved and systematic errors reduced.
The end of 1984 saw completion of a major programme of experiments, developing a model that would take full advantage of the multi-tasking capability of the CRAY X-MP computer. At the same time, modelling of the
106 Chapter 9boundary layer, radiation and convection were all being intensively investigated. The high-resolution, now T106, model, with improved modelling of shallow convection and of radiation, including better representation of the effects of clouds and aerosols, was ready. Waves down to 400 km were modelled at this resolution. The comprehensive physical parametrization schemes included shallow and deep convection, a radiation scheme that allowed interaction with model-generated clouds, and the diurnal radiative cycle.
At last, Lennart Bengtsson was ready to propose introducing the new model as the operational model. The SAC was shown the results of experiments comparing the new model with that currently operational. The results were not very spectacular. The SAC Chairman, Fred Bushby of the UK, noted informally that “the real secret when you bring in a new scheme or model is not to make the forecasts worse! The main benefit of the new system is its potential for further development.” The new T106 model became operational in May 1985.