Extreme weather: will we ever see it coming?
In October 1987, the Great Storm wreaked havoc across the south of the UK, taking everyone by surprise. Now, 30 years on, why do we still have trouble predicting the weather?
Thirty years after the Great Storm of 1987, we find out why weather is so hard to predict.
O n the morning of 16 October 1987, southern England woke to a scene of devastation. Eighteen people were dead, and thousands of homes were without power. Fifteen million trees had fallen across the country, irreparably changing the landscape and blocking roads and railways. In Folkestone, a 110-metre Sealink ferry was discovered marooned on the beach. A pier on the Isle of Wight was almost totally demolished.
The culprit was the Great Storm of 1987, as it became known. Forming in the Bay of Biscay north of Spain and sweeping up across the country, it brought gusts of up to 185km/ h (115mph) and sustained winds of more than 80km/ h (50mph) across the south-east. The passage of the storm’s warm front increased temperatures by up to 10°C, while barometric pressure fell to 951 millibars. It was the worst storm to have hit southern England and northern France in almost 300 years.
Forecasters had seen it coming. Five days earlier, the BBC’s farmers’ forecast warned that the weather would be “becoming very windy later in the week”. By the middle of the week, however, the computer models that meteorologists use to make their predictions had
become more uncertain. Shipping forecasts continued to warn of dangerous weather, but the inland forecasts cautioned of rain rather than winds. On the lunchtime news, weather presenter Michael Fish issued a now infamous proclamation that there would be no hurricane, and by the time most people went to bed, no warnings of unusually strong winds had been issued on TV or radio. As the night continued, however, the storm grew stronger and stronger. Meteorologists began to realise what was happening and issued ever-worsening alerts to emergency services and the government. At 10:35pm on 15 October, Force 10 gales over the channel were forecast. At 1:40am on 16 October, warnings of Force 11 were issued. The military was notified that its assistance may be required to deal with the storm’s consequences.
So why was this particular storm so damaging? “One thing people talk about when they remember the storm was just how warm it was that night,” says Dr Pete Inness, a meteorologist at the University of Reading. “The subtropical air mass sitting in the middle of the storm provided a huge amount of heat energy to spin the storm up.”
“There was also a very small feature developed within it, which we today call a sting jet, but was unknown to meteorologists in 1987,” adds Mel Harrowsmith, head of civil contingencies at the Met Office. “A sting jet is, effectively, a small core of very strong winds, and they tend to happen towards the back end of a storm. So you get this very narrow, very strong core of wind that touches down on the ground and can cause a lot of devastation.”
VIRTUAL WEATHER
Computer simulations of the atmosphere work in pretty much the same way today as they did back in 1987, and it’s a three-step process. The first step is to collect data from meteorological stations on the ground, buoys in the ocean, weather balloons in the sky and satellites up in orbit. This can tell us what the state of the atmosphere is at any given time.
That data is then plugged into the computer model itself – a set of equations that reflect how air and water (the chief components of weather) behave under different circumstances. Hot air tends to rise over cold air, for example, and holds more water vapour. But as hot air rises it cools, and the water vapour condenses out into water droplets – just like on your mirror when you take a shower. That’s how a cloud is made.
Finally, to get a forecast, the atmosphere is split up into a series of ‘boxes’. Each box is given data from real-world observations and then allowed to develop according to the equations defined in step two. By seeing what happens in those different boxes over time, we get a weather forecast.
“On the lunchtime news, Michael Fish issued a now infamous proclamation that there would be no hurricane, and by the time most people went to bed, no warnings had been issued”
The reason why forecasts can go wrong is the potential for mistakes in all three of these stages. In the first stage, we might not collect enough data: it’s impossible to know the exact weather in every part of the globe – particularly over the ocean – so sometimes we have to make an educated guess on what numbers to put in. Plugging in numbers that are even slightly wrong will create an error that grows larger over time – just like the proverbial butterfly flapping its wings and eventually causing a hurricane. Meteorologists refer to this as ‘chaos’, and it’s the main reason why forecasts get less accurate over time.
Even if you did have perfect data, you might not get a correct forecast because the equations aren’t quite right. Meteorologists are constantly tweaking the physics of their models in response to the latest research, and the complexity of the atmosphere means that it’s going to be a long time, if ever, before we can perfectly predict what the weather will do in every circumstance.
Finally, there are those boxes, which meteorologists refer to as the ‘resolution’ of the model. Think of them like the pixels of your laptop or smartphone: the more you have, the better the picture you get, but the more computing
power it takes to run them all at the same time. The same is true for weather models. This may be why the Great Storm of 1987 was so poorly forecast. “The boxes back then were about 150km by 150km,” Harrowsmith says. “So if you have a small feature inside that box, the model cannot see it.” The sting jet that made the storm particularly damaging was about 50km across, making it impossible for the model to take it into account.
ON THE UP
Since 1987, meteorologists have made substantial progress in all three areas of weather modelling, making forecasts far more reliable. “Our four-day forecast now is as accurate as our one-day forecast was 30 years ago,” says Harrowsmith.
Some of that progress is down to increased supercomputing power. In 1982, the Met Office computers could handle 200 million calculations per second. In 1997, that figure was one trillion calculations per second, and earlier this year, the Met Office installed a new supercomputer capable of a stunning 14 quadrillion (that’s 14 with 15 zeros after it) calculations per second.
But that’s not all. We now have access to more weather observations, chiefly due to improvements in satellite and radar technology. The Met Office is in the middle of upgrading its radar network to be able to see not only where it’s raining but also the size and shape of raindrops. This will help forecasters to differentiate between rain and snow – revealing whether that oncoming storm is a downpour or a blizzard.
We also have much more data from the Atlantic. “As a result of the Great Storm, the Met Office realised they didn’t have enough observations over the ocean,” says Dr Hannah Christensen at the US National Center for Atmospheric Research. “So they set up a lot more observational buoys out to the west and south, to fill in some of the gaps.” Plus, new techniques like ensemble forecasting – where several forecasts are run at the same time with slightly different starting data – can give us an idea of how certain a forecast is. If all the models still come up with the same forecast, meteorologists can be more sure it’s correct.
On the subject of uncertainty, one of the biggest changes for forecasters over the last three decades has been in the way that a forecast is communicated. Where once we were told it was going to rain, for example, we’re now told there is a 60 per cent chance of rain. “For a long time the Met Office was reluctant to put out probabilistic information, and some of their reluctance came from the idea that the public wouldn’t know how to use the information,” says Christensen. Different weather forecasting services use probabilities in different ways, but the Met Office
“A number of meteorologists are investigating the potential for artificial intelligence and machine learning to improve forecasts”
explains on its website that its probabilities indicate how likely it is that precipitation will fall at some point during a specified period in a specified location. So a 70 per cent chance of rain in Scotland, for example, means that there’s a seven-in-ten chance that some rain will fall anywhere north of the border. “If you have the communication channels to present those uncertainties, this extra information can help people to make decisions to protect themselves and their property,” says Harrowsmith.
FUTURE FORECASTING
With continued improvements in computer power and our understanding of atmospheric physics, there’s little doubt that our forecasts will also continue to improve. Inness says that some meteorologists are investigating automated rocket- or drone-launching buoys that could gather data on developing systems across the Atlantic. “You wouldn’t use them every day,” he says, “but they could come in handy if you knew there was a weather system in that area and were particularly uncertain about how it was going to develop. It’s not fantasy by any means.”
Meanwhile, a number of meteorologists are investigating the potential for artificial intelligence and machine learning to improve forecasts, by using past weather data to learn from inaccurate forecasts and make predictions about future weather patterns. “At the moment it seems to be in a very exploratory phase,” says Christensen, “but there are results coming out which show that AI has the potential to help us process our enormous amount of observations to improve our simulations of the atmosphere.”
But the better we get at forecasting, the harder it is to tease out further improvements – and the more demanding the audience gets. “These days, we’ll do a 10-day forecast and we can predict that timescale with reasonable accuracy,” says Rebekah LaBar, a consultant meteorologist for MetraWeather in New Zealand. “But there are always people that ask us: what’s the coming summer going to be like? How many hot days are we going to have?”
Ultimately, there’ll probably never be such a thing as a perfect weather forecast. No matter how good our models and how powerful our supercomputers, chaos means that even tiny mistakes in our observations will grow out of control over time, causing some degree of uncertainty in our predictions. But as atmospheric science continues to progress, a mistake on the scale of the Great Storm is looking increasingly like a thing of the past. Storm clouds today rarely take us by surprise. But it’s always worth packing a raincoat, just in case.