How do they predict the weather? James Burn, a meteorologist from IBM, explains how weather modelling works, how data is gathered and outcomes calculated.

Weather and climate modelling
Play video

James Burn is a meteorological solutions engineer with IBM. He has worked with climate data for around 20 years, is passionate about his subject and equally committed to answering the complex, yet fascinating question: how is weather predicted?

The weather, beyond being a quintessentially British obsession, is a topic of huge global importance. In 2020, the world experienced around $95 billion of weather related losses. 2020 was also the joint warmest year on record according to NASA. Oddly though, if you ask The National Oceanic and Atmospheric Administration (NOAA), it’ll tell you that 2020 was actually the second warmest year on record.

‘NOAA uses a different statistical framework,’ Burn says, hinting at an important weather prediction fact: statistical models are very important and there are many different frameworks.’ Indeed many forecasts, Burn reveals, actually draw on and combine different models when making their predictions.   

How do they predict the weather?

IBM, as an organisation, has a long history of investing in and developing solutions around climate modelling. It has a dedicated R&D programme, invests heavily in infrastructure and has taken a leadership position in using AI to improve weather forecasting.

Big Blue’s interest and investment in weather might, at the first glance, seem like an odd one for a technology firm. But, IBM has a huge number of clients across an equally wide selection of sectors and industries. Transport, agriculture, insurance companies, utilities and even policy holders and makers – they’re all IBM clients and are involved in work which can be disrupted by adverse weather conditions.

‘I also work with some really interesting start-ups,’ Burn says. ‘We’re working with [one] which is using weather data to help optimise satellite data transmission. When it rains it attenuates the signal… the data transmission. They’re using our weather data to help improve and optimise where base stations should be located.’

Facebook, Google and many of the world’s most popular weather apps also get their weather from IBM. 

Burn’s fascinating talk about how weather is forecast covers fundamentals, including: 

  • How does weather modelling work?
  • Data assimilation and quality control
  • IBM multi-model forecast methodology
  • Weighting or model input with machine learning
  • Probabilistic outputs: the ‘chances of rain’ and what does this mean?
  • Intelligent gridding - getting more for your computing resource
  • Short term weather modelling verses climate modelling.

The mega maths of how weather is predicted

So, how is weather predicted? How does weather modelling work? Burn says: ‘The atmosphere is a fluid - it’s a mixture of gas, clouds, water vapour and a few small particles. That fluid follows the laws of physics and, with relative ease, you can predict the movement of that fluid around the globe.’

For you

Be part of something bigger, join the Chartered Institute for IT.

At its heart, weather prediction is, of course, very mathematical. Explaining, Burn says: ‘You take a set of dynamic equations which look at the change over time in atmospheric variables. Essentially, you have a model which describes the state of the atmosphere now and it asks, “what will it be at a certain time in the future?” You then need to feed that model with a set of observations. That will let you understand what the weather will be doing at one given point.’

Of course, knowing what the weather will be at one place on Earth isn’t much use. Most people are interested in understanding weather in a region and being able to see how weather will move through that area. So, you need to start with a grid.

‘You have a grid of data points,’ Burn explains. ‘You grid up the atmosphere. Grids start on the surface and go up vertically into the atmosphere. Some of the best models today look at up to 200 layers up into the atmosphere and have a grid resolution of about one kilometre, or possibly less. You need to build up your observations. You need to build up your best estimate of what the weather is doing at individual grid points and then you use equations to project forward. What will the state of the atmosphere be in ten or fifteen minutes?’

Continuing, he explains: ‘Each time, to project forward, you are solving those equations and that lets you build up your predictions.’

How is weather predicted? A history lesson

The first numerical weather forecast was completed by Lewis Fry Richardson in the early 20th century. ‘He did a forecast for a point. He spent weeks and weeks on it and,’ Burn says, ‘he was terribly wrong.’ Undeterred, Fry Richardson proposed a forecast factory.

‘He wanted what he called computers - [human] mathematicians... He wanted 64,000 mathematicians sitting around an amphitheatre, all processing the equations… doing the calculations for every point in Earth. It didn’t happen because, a few years later, they invented computing and, as meteorologists we’ve being working very closely with computing technology ever since. And one of the main uses of super computing is using all that power to calculate forecasts.’

Burn, in his talk, goes into huge amounts of detail which is fascinating and equally easy to understand. Watch James Burn’s climate video now.