PITY the poor weather forecaster. Behind the TV presenter鈥檚 smiling face and
phraseology lies some of the most sophisticated technology in the world. Off
screen, they can tap into a stunning worldwide information-gathering system.
Every 24 hours, it logs more than a million real-time weather observations from
satellites, ships, aircraft, buoys and the 1300 or so weather balloons released
around the world each day.
This data is then crunched by some of the most powerful computers on Earth.
Two of the biggest are in Britain: the Meteorological Office鈥檚 Cray
supercomputer in Bracknell, Berkshire, and the Fujitsu machine at the European
Centre for Medium-Range Weather Forecasts in nearby Reading, which can make 250
billion calculations a second.
The forecasters have plenty of hard data to fall back on. And yet we still
don鈥檛 believe them. So what has gone wrong? Is it us, them, or their computers?
A recent trip to the European Centre convinced me that they at least know their
synoptics. For the past 19 years, this little-known outfit has been supplying
national meteorological services with daily forecasts of the world鈥檚 weather for
up to 10 days ahead. And with increasing accuracy. It claims an 80 per cent hit
rate for its current five-day forecast鈥攎ore accurate than two-day
forecasts were in the early 1970s.
Advertisement
The centre also boasts that its forecasts extend half a day further ahead
than those of the Met Office. Number crunching is the name of the game. Staff at
Reading convert the global data into a map of world weather containing 4 million
grid points at 31 different heights in the atmosphere. Then, using a
mathematical model of the atmosphere, the computer forecasts the weather at each
of those points in 15-minute blocks, up to 10 days ahead. Oh, and it repeats
this exercise 51 times 鈥攋ust for luck.
The weather people call this repetition 鈥渆nsemble forecasting鈥�, and it has
been all the rage since an unexpected storm blew in late one evening and ripped
through southern Britain in October 1987. Never again, said the Met people. So
now they duplicate the forecast, slightly altering the starting data each time
to see what happens. If all the forecasts look the same, then the main one is
robust. If they are all over the shop, then the situation is chaotic.
Recently, the European Centre launched a six-month service of 鈥渁verage
weather鈥� spread over a month. This is marginally easier to do than the 10-day
forecasts. Random factors become less important, while more predictable
underlying factors鈥攕uch as ocean temperature and snow cover on
land鈥攖ake hold. By coupling an ocean model with its standard atmospheric
model, the centre successfully predicted the current El Ni帽o back in
December 1996, six months before the first signs appeared. And it forecast that
this climatic anomaly in the faraway Pacific Ocean would cause heavy rains and
floods in central Europe last summer. Nice work. Picking the unexpected storms
has always been the hard bit.
For the record, the Centre predicts more 鈥渨eird weather鈥� when El
Ni帽o鈥檚 twin sister, La Ni帽a, takes hold late this year. And yet,
and yet . . . It鈥檚 still the mistakes people remember. And often unfairly. I
planned to watch some test-match cricket in Birmingham five days after visiting
the Reading centre. So I checked its forecast. No rain, it said鈥攆or
Reading, anyhow. But it drizzled all morning at the match鈥攂arely enough to
register in a rain gauge or on a weather map, but enough to stop play for two
hours.
And here is the problem. The forecasters think in synoptic charts. Their
world is composed of depressions and anticyclones and weather fronts. We think
of rain and clouds and sun. And we don鈥檛 live on their maps. We live in frost
hollows or on windy hilltops; in the rainless lees of mountains or in urban
鈥渉eat islands鈥�. Our weather is often influenced by local features that don鈥檛
figure on the charts.
Often where you are matters as much as what is happening in the atmosphere.
That is why the best forecast of tomorrow鈥檚 weather in any one place often comes
not from a supercomputer, but from the rule of thumb that says: tomorrow it will
be similar to today.