Science Week 2016: The super-power behind weather forecasting
Since the earliest attempt at numerical weather prediction (NWP) by Lewis Fry Richardson in 1911, it has been clear that forecasting the weather is a computationally expensive process. It took Richardson several months to calculate just a 6 hour pressure change at a single location by hand.
Following the development of the earliest computers, meteorologists and computer scientists collaborated to develop an NWP model which could run on the machines. The first computer forecast was performed on a machine called ENIAC (Electronic Numerical Integrator And Computer) located at the University of Pennsylvania. It began at midday on Sunday 5th March 1950 and continued 24 hours a day for 33 days and nights. Four reasonably accurate 24 hour forecasts were produced as a result, each of which took about 24 hours of computation, just keeping pace with real time.
The ENIAC machine in Philadelphia, Pennsylvania, c. 1947-1955, which produced the first computer-generated weather forecast. (Source: U.S. Army photo. Public Domain)
A modern mobile phone is approximately one million times more powerful than ENIAC, a machine that weighed 27 tonnes and cost almost seven million dollars to build. This was demonstrated in 2008 by an experiment performed by Peter and Owen Lynch, who successfully ran the first forecast produced by ENIAC on a Nokia mobile phone (PHONIAC). The increase in computing power over time has greatly improved the quality of weather forecasts, allowing larger areas to be forecast in greater detail.
Since 2007, Met Éireann has been running its forecast models at the Irish Centre for High-End Computing (ICHEC) which has two supercomputers, called Fionn and Indra. These machines each have approximately 7,600 processor cores and 20 terabytes (TB), or 20,000 gigabytes (GB), of RAM. A typical desktop computer has about 2 processor cores and 8 GB of RAM. These supercomputers allow Met Éireann to produce two sets of 54 hour forecasts four times daily, with horizontal grid spacings of down to 2.5 km. Each forecast takes approximately one hour to run.
The European Centre for Medium-Range Weather Forecasts (ECMWF) houses two supercomputers, both of which rank in the top 20 globally according to www.top500.org. They both contain around 130,000 processor cores with 500 TB of RAM. From early 2017, Met Éireann will run all of its forecast models on these machines. As computer technology improves further we can expect weather forecasts to become more detailed and more accurate than ever before.
References and further reading:
Peter Lynch, The Emergence of Numerical Weather Prediction: Richardson's Dream. Published by Cambridge University Press, 2006.
Peter Lynch and Owen Lynch, Forecasts by PHONIAC. Weather – November 2008, Vol. 63, No. 11.
For further information contact Ronan Darcy (ronan dot darcy at met dot ie).