In 1839 the Victorian polymath John Ruskin, then a 20-year-old undergraduate, had a remarkable vision for the development of the infant science of meteorology. Inspired by the recent invention of telegraphy, he imagined a “vast machine” built on “perfect systems of methodical and simultaneous observation” that would combine weather data from around the world to show “the state of the atmosphere” everywhere. An individual’s “observations are useless; for they are made upon a point”, Ruskin wrote in Transactions of the Meteorological Society, “while the speculations to be derived from them must be on space.” Andrew Blum’s excellent book, The Weather Machine, describes the technological and geopolitical developments that have brought Ruskin’s dream to reality, creating a global meteorological machine that uses scientific models of the atmosphere to convert observations into ever more accurate speculations — or forecasts — of future weather. In the 1850s the Smithsonian Institution in Washington DC set up the first systematic meteorological programme, taking advantage of new telegraph networks to warn of bad weather. Observers across the US sent in reports that were displayed on a gigantic map with small paper discs of different colours indicating conditions in each place (white for clear skies, brown for clouds and so on). The Smithsonian’s directors reported in 1858 that the map not only showed visitors the weather experienced elsewhere “but is also of importance in determining at a glance the probable changes which may soon be expected.” Others followed. The UK’s new Meteorological Office initiated the world’s first routine weather forecasts in 1861. The International Meteorological Organisation began work in 1873 to standardise observing worldwide. Predictions made from the blossoming observing networks of the late 19th century were not very accurate as forecasters lacked a good understanding of the way weather systems developed. A firmer scientific basis for meteorology began to emerge in the early 20th century, thanks above all to a group of Norwegian meteorologists known as the Bergen School. Their methods, showing for example the development of “fronts” between air masses of different geographical origins, were important for forecasting during the second world war — including the famous prediction of a brief window of fine weather in Normandy, which enabled the Allies to launch the D-Day landings on a day when the Germans thought it would be too stormy. Blum describes vividly the weather war as Germany fought to overcome its geographical disadvantage: the movement of weather systems across the North Atlantic from west to east. Hostilities cut off the flow of observations of incoming weather to German meteorologists. In 1943, in the most dramatic example of an initiative to overcome that, a U-boat crew put a clandestine automatic weather station on an uninhabited hill on Canada’s north-east coast. The postwar world has seen a sustained improvement in forecasting. New observing stations on land, sea, air and above all in orbit have greatly extended the quantity, quality and range of meteorological data. Supercomputers have transformed data processing and also refined the “numerical models” of atmospheric physics that underlie the forecasts. Blum’s fieldwork for The Weather Machine included visits to several key meteorological sites, all described in lively detail. He sees the European Centre for Medium-Range Weather Forecasts in Reading, generally regarded as the world champion for predicting the weather a few days ahead, as the realisation of an influential “dream” that the British mathematical meteorologist Lewis Fry Richardson published in 1922. Richardson envisaged a forecasting factory where thousands of human “computers” worked interactively to solve equations for the future state of the atmosphere. “Perhaps some day in the dim future it will be possible to advance the computations faster than the weather advances,” he wrote. Blum shows how first Ruskin’s and then Richardson’s imagination anticipated both the global view and the scientific components at the heart of today’s weather machine. An analysis published by researchers at Penn State University earlier this year pointed out that a five-day forecast is now as accurate as a one-day forecast was in 1980. People love to carp when forecasters get it wrong, yet few seem to appreciate how well the machine is working.