The Weather   Wednesday, 13/03/2024

With all well here, the one thing that is noticeable is that the temperature is cooling as the season advances into autumn. There is still nice warm sunshine, but if it clouds over it feels cool. And, depending on whether the skies are clear or cloudy at night, it can get fairly cold in the early hours – some nights the temperature has fallen as low as 10 or 11 degrees. 

The weather is far from an irrelevant subject for me. I do find that I notice it more – especially wind direction and how it changes! When I’m sailing I have the backup of PredictWind, and the various weather models it’s based on. A few days ago I got thinking, thanks to an interesting article in the UK’s Daily Telegraph. This set me off, and after a couple of hours of reading various things that I found online, I thought I’d take the gist of them and write about it here… So if you have no interest in weather forecasting, you can stop reading! (Oh, and if you wonder when on earth I do things like this… it was late Saturday evening, done while drinking – appropriately – a Dark and Stormy: rum and ginger ale). 

Weather forecasting has come a long way. First, a little history (and you are likely to recognise some of the names). The earliest attempts to explain meteorology took place in India and, in particular, the Islamic world – which until the Middle Ages was at the forefront of rational scientific enquiry. 

By the late 1400s, the Europeans were getting started, in parallel with the voyages of discovery. In 1643, Evangelista Torricelli invented the mercury barometer; in 1648 Fernando de Medici sponsored the first cross-Europe weather observing network, centred on Florence. In 1724 Gabriel Fahrenheit created a reliable scale for measuring temperature with a mercury thermometer, followed closely in 1742 by Anders Celsius and his alternative scale. By 1802 Luke Howard was identifying and naming three broad cloud types: cirriform, cumuliform and stratiform, and classifying further on whether they are lower or upper level. In 1806, Francis Beaufort introduced his system for classifying wind speeds: the Beaufort Scale, still in use today. 

And from there, scientific progress was rapid. By 1849 the Smithsonian Institution began a network of 150 weather observers across the United States, connected by the new telegraph. This was followed in 1861 by the use of the telegraph to gather and publish daily observations in England (in The Times), a world first. The project was led by Vice-Admiral Robert Fitzroy, who had captained Charles Darwin’s HMS Beagle on his voyage of scientific discovery, and then went on to establish the U.K. Met Office. The aim was to improve the understanding of the marine climate and therefore the safety of life and ships at sea. Its first storm warning service also came in 1861, eventually becoming the “shipping forecast”. Indeed, Fitzroy coined the term “weather forecast”. 

In those early days, gathering data was laborious and pretty basic, relying on measurements of atmospheric pressure and wind speed and direction. By the early 1920s, maths and physics were being incorporated, using the study of fluids to better understand both the oceans and the atmosphere. In 1950, there was the first successful numerical weather prediction experiment; by the late 1950s, computers were being used, revolutionising forecasting. The power and speed of computers has of course increased enormously, eventually allowing machine learning. And now there are high hopes of artificial intelligence being able to make new sense of the huge streams of data generated worldwide, from deepwater buoys, automated weather stations, satellites, weather balloons and transponders on aircraft and ships. The oceans have a huge influence because of the way that they move heat from the equator to the poles and influence how storms develop. 

It is said that in recent decades, weather forecasting has improved on average by a day a decade, which means that a seven-day forecast in 2020 was as accurate as a three-day forecast in 1980. There are huge implications: not only for the safety of ships and planning daily life. Construction (for example, planning the laying of foundations), agriculture (workforce planning for harvesting in dry weather, or covering crops to prevent frost damage) and now for planning energy generation from renewables: wind, solar and hydroelectricity.

Artificial intelligence represents a huge step change. In December, Google DeepMind scientists reported on a new model called GraphCast, which aims to predict weather up to 10 days in advance. The model’s output can be generated in less than a minute on a desktop computer – which really is quick: this compares to the six hours taken by the European Centre for Medium-Range Weather Forecasts to produce the most accurate current forecasts on supercomputers (their ECMWF model is a favourite of mine on PredictWind). And the Google people claim that GraphCast performs better, beating the current best weather prediction systems on 90% of 1,380 metrics. It also proves better at forecasting severe weather events, which appear to be becoming more frequent. This AI model works by being trained on 40 years of historical weather data, identifying patterns in the data that are not easy to see in the equations used in current models. It then learns from these patterns, increasing its own predictive ability. Oh, and it uses only one-thousandth of the energy used by a supercomputer. It’s early days, but another revolution is underway…