The Economist - The high-tech race to improve weather forecasting

Engine Room Number Five at the European Centre for Medium-Range Weather Forecasts’ data centre in Bologna houses a series of motors, each turning a three-tonne flywheel. Should the electricity cut out, the flywheels—and those in four other rooms elsewhere in the building—have enough momentum to keep the ECMWF’s newest supercomputer running until the back-up diesel generators fire up. Those generators have fuel for three days.
The World Meteorological Organisation (WMO), an arm of the United Nations, reckons that a five-day forecast today is about as accurate as a two-day forecast was a quarter of a century ago.
But no matter how powerful computers become, there is a limit to how far ahead a numerical forecast can look. The atmosphere is what mathematicians call a “chaotic system”—one that is exquisitely sensitive to its starting conditions. A tiny initial change in temperature or pressure can compound over days into drastically different sorts of weather. Since no measurement can be perfectly accurate, this is a problem that no amount of computing power can solve.
Meteomatics, a Swiss firm founded in 2012, allows its customers to crunch data from a range of sources in a way that suits their needs—such as “downscaling” the output of a numerical model by shaping it around the local topography.