Numerical Weather Prediction and Supercomputers

Stormy Weather: An exhibit in partnership with the Met Office

By The Royal Society

Lewis Fry Richardson (c. 1920) by Meteorological OfficeThe Royal Society

Lewis Fry Richardson FRS (1881 - 1953)

Lewis Fry Richardson was the pioneer of weather forecasting using mathematical methods. As a Quaker and pacifist, Richardson left the Met Office to join the Friends' Ambulance Unit during WW1. He used his rest periods to continue developing the first ever numerical forecast based on scientific rules rather than past trends. Before the second Battle of Champagne his work was sent to the rear lines for safety and was lost under a coal heap for several months. Thankfully it was re-discovered and 'Weather Prediction by Numerical Process' was published in 1922. Soon after, Richardson left the Met Office because his beliefs did not fit with some of the work it was tasked to do for the Air Ministry. He would never work in meteorology again.

Weather Prediction by Numerical Process (1922) by Lewis Fry Richardson (1881-1953)The Royal Society

In Weather Prediction by Numerical Process, Richardson laid out the equations to calculate changes in temperature, pressure, evaporation and winds at several levels in the atmosphere.

Weather Prediction by Numerical Process (1922) by Lewis Fry Richardson (1881-1953)The Royal Society

He provided a range of computing or worksheets on which the complex calculations required for his forecasting process could be carried out.

Figure 1 from Weather prediction by numerical process (1922) by Lewis Fry Richardson (1881-1953)The Royal Society

To enable a forecast to be made, Richardson overlaid a geographical area with a grid. This is known as his chessboard or chequerboard pattern.

All of the required calculations were made for each square, then applied to the next in order to reflect the passing of time and the movement of air masses.

A good supply of observations is crucial to the success of numerical weather prediction.

Inevitably there were more observations available for some areas than for others. Richardson devised some methods to compensate for this but more effective smoothing techniques were a later development.

Weather Prediction by Numerical Process (1922) by Lewis Fry Richardson (1881-1953)The Royal Society

Richardson could not imagine the dawn of the electronic computing age. Instead he described how his forecasting might be carried out by humans in a giant 'Forecast Factory'. He estimated around 64,000 people would be required to complete the calculations in time to be useful but later analysis indicated this was an understatement. In essence his description of how the Forecast Factory would work describes the basic 'main/secondary' functionality of a modern supercomputer.

LIFE Photo Collection

The Dawn of Computers

The world's first programmable computer was designed at Bletchley Park in England during the Second World War. Bletchley was home to the Government Code and Cypher School, which successfully penetrated the secret communications of the Axis Powers. The computer was built to help break the Lorenz Cipher, which was used by the German High Command for encrypting and sending messages. 

The computer, known as 'Colossus', was built by Tommy Flowers, a General Post Office (GPO) research telephone engineer . Although Colossus was designed specifically to aid in cryptoanalysis, the concept of an electronic, programmable, digital computer marked the start of a revolution in data processing.

Met Office buildings, Dunstable (1945) by Meteorological OfficeThe Royal Society

Research became part of the standard ethos of the Meteorological Office during the post-war period and this was soon put to use in collaboration with the global community to improve short-range forecasting.

Work started to find a ‘meteorological electronic brain’ that could process significant amounts of data and to develop L.F. Richardson’s rules formulated some 30 years before.

ENIAC Programmers / United States Army (1946) by United States ArmyNational Women’s History Museum

John von Neumann, a mathematician from the Institute for Advanced Study in Princeton, had started working on a computer to carry out mathematical calculations. He began to collaborate with an international team of meteorologists to see if the computer could be used to make the mathematical calculations needed for Numerical Weather Prediction.

Ferranti Mark 1 - Manchester (c.1950) by Meteorological OfficeThe Royal Society

The first NWP forecast in the UK was completed by Fred H. Bushby and Mavis Kathleen Hinds in 1952 under the guidance of John Stanley Sawyer FRS. These experimental forecasts were generated using a 12 × 8 grid with a grid spacing of 260 km, a one hour time-step, and required four hours of computing time for a 24-hour forecast on the EDSAC computer at Cambridge and the LEO computer of Lyons & Company.

Following these initial experiments, work moved to the Ferranti Mark 1 computer at the Manchester University Department of Electrical Engineering.

Ferranti Mercury (Meteor) computer, Dunstable (c. 1959) by Meteorological OfficeThe Royal Society

Electronic computers and operational Numerical Weather Prediction

Work in Manchester proved that Numerical Weather Prediction could be a reality but the Meteorological Office needed its own computer to make more progress. 

Ferranti Mercury Computer at Dunstable known as Meteor (1959) by Meteorological OfficeThe Royal Society

In January 1959, a Ferranti Mercury computer, known as ‘Meteor’, was installed at the Met Office in Dunstable. It was the first computer dedicated to NWP research.

Ferranti Mercury computer - Meteor (1959) by Meteorological OfficeThe Royal Society

Meteor was used to trial an operational forecasting suite in early 1960, using the improved Bushby–Whitelam three-level model covering western Europe and the North Atlantic.

It consisted of a six-hour forecast from an 0000 UTC analysis, followed by a 0600 UTC re-analysis and a 24-hour forecast.

English Electric KDF9 (1965/1973) by Meteorological OfficeThe Royal Society

Although work progressed well, it was still taking too long to produce forecasts. A more powerful and reliable computer was required if NWP was ever to become operational.

KDF9 (Lyons Electronic Office) computer installation (1966) by English ElectricThe Royal Society

In 1961 the Met Office relocated to Bracknell where it had space for a larger and more powerful computer. In 1965 the new computer, an English Electric KDF9 called ‘Comet’, was installed. The old 'Meteor' computer was reassembled at the Chemical Defence Experimental Establishment at Porton Down.

Comet used transistors, had a speed of 60,000 arithmetic operations per second and a memory of 12,000 numbers.

KDF9 Super Computer (1965-11-02) by Meteorological OficeThe Royal Society

The first operational NWP forecast was issued on Monday 2 November 1965 with routine forecasts realised twice a day from then on.

The Meteorological Office was the first weather service in the world to launch operational numerical weather prediction.

1st operational Numerical Weather Prediction Output (1965-11-02) by Meteorological OfficeThe Royal Society

The Met Office held its first press conference to mark the event and souvenir forecast charts were given to everyone who attended.

The charts clearly show the use of Richardson's grid pattern - now much enlarged and developed from an experimental concept to an operational system.

IBM360/195 Console (1971/1982) by Meteorological OfficeThe Royal Society

The global exchange of weather observational data grew rapidly with satellite transmission. The Global Telecommunications System of the World Meteorological Organisation (WMO) also transformed exchanges between National Weather Services.

A more powerful computer than 'Comet' was now needed and an IBM 360/195 capable of running a ten-level model was installed in December 1971. It was nicknamed 'Cosmos'.

CYBER 205, Bracknell, (1981) by Meteorological OfficeThe Royal Society

Supercomputers and modelling advances

'Cosmos' was in turn replaced by the first supercomputer, a CDC Cyber 205, in early 1982. Capable of 400 million operations per second, the new machine was able to run the first global operational forecasting model, and just in time to assist in British operations surrounding the Falklands War. The supercomputer could run the global forecast model for 24-hours ahead including an extension to 15 levels in the atmosphere and the addition of a multi-level ocean model. 

Chernobyl - NAME model output (1987) by Meteorological OfficeThe Royal Society

Following the widespread dispersion of the radioactive cloud from the Chernobyl accident in 1986, the NAME (Numerical Atmospheric-dispersion Modelling Environment) modelling system was developed.

NAME has been continuously improved and used to model dispersion from a wide range of hazards including chemical fires, volcanic eruptions and disease outbreaks.

Early Unified Model Output (1991-03-21) by Meteorological OfficeThe Royal Society

The next step for Numerical Weather Predictions was the development of the Unified Model in 1990. The UM was a critical step for the Office, as it enabled different configurations of the same model to be run in order to produce both weather forecasts and longer term climate predictions.

CRAY Y-MP (1991) by CrayThe Royal Society

The UM was implemented operationally in June 1991 on a new supercomputer, a Cray Systems C90 Y-MP-8 which replaced the ageing Cyber 205.

Cray T3E (1997) by Meteorological OfficeThe Royal Society

When the Met Office moved from Bracknell to Exeter in 2003 it involved one of the largest IT relocations ever attempted, including the physical transport of the existing Cray T3E supercomputers.

The two computers were moved separately with one maintaining forecasts in Bracknell until the other took over in Exeter. The move was completed two weeks early and no forecasts were missed.

Ensemble Forecasting (2007-11-05) by Meteorological OfficeThe Royal Society

In 2005 the Met Office introduced short range ensemble forecasting (MOGREPS). MOGREPS is primarily designed to aid the forecasting of rapid storm development, wind, rain, snow and fog.

To start an ensemble forecast a set of small changes called perturbations are made to the analysis, which are consistent with the uncertainties in the starting conditions. MOGREPS then makes small random variations to the forecast model itself, as well as changes to the initial state.

Cray XC40 (2016) by Meteorological OfficeThe Royal Society

The Unified Model is in continuous development by the Met Office and its partners, adding state of the art understanding of atmospheric processes to new releases. For example, the operational global resolution run at the Met Office improved from 90 km in 1991 to 10 km by 2017.

The immense computing power required to achieve this is provided by the new Cray XC40 Met Office Super Computers.

Cray XC40 Met Office Super Computer (2017) by Meteorological OfficeThe Royal Society

The Met Office has three super computers, two located in the Exeter HQ which run the operational models and one which has more than double the power of the other two combined, located in the new Richardson Building in Exeter Science Park. This is dedicated to weather and climate research.

The combined computing power is roughly equivalent to every person on the planet performing 2,000,000 calculations per second.

This increased power enables even higher resolution forecasts and experimental models are now running at 1km scale.

Elements of Cray Super Computer (2016) by CrayThe Royal Society

Supecomputer Components

Top left: Computer daughter board - the main processing engine of the supercomputers. There are 6,466 of these in total.

Top right: Aries Power and Distribution units - these provide power to the daughter boards and and allow compute nodes to transfer data between each other at very high speed and very low latency. There are about 3,300 of these.

Bottom left: Central Processing Unit - these carry out the calculations. There are 25,864 in the systems.

Bottom right: Active Optical Cables. Cables like those at the top connect storage systems to the computer, and can transfer data at 56 gigabits per second. Those at the bottom connect power units more than 2m apart.

Stormy Weather- from lore to science (2020) by The Royal SocietyThe Royal Society

To discover more on Lewis Fry Richardson, the pioneer of numerical weather predictions, browse to our People of Science story dedicated to Richardson.

Credits: Story

All rights reserved © The Royal Society 2020

The digital and physical exhibits were curated by Dr Catherine Ross (National Meteorological Archive, Met Office) and Dr Louisiane Ferlier (The Royal Society).

Credits: All media
The story featured may in some cases have been created by an independent third party and may not always represent the views of the institutions, listed below, who have supplied the content.
Google apps