When Science Goes Wrong (8 page)
People think of the task of weather forecasting as getting from the current conditions to the conditions tomorrow or the day after. But that actually omits a crucial first step, which is getting from the observational data to the current conditions. Forecasters call this step producing an ‘analysis’ – a mapping of pressure, temperature, humidity, wind strength and so forth – for the region of interest at the time the observations are made. It may be presented as a two-dimensional chart such as one sees on television, but the vertical dimension of the atmosphere is equally important. Why it’s called an analysis, I don’t know – it’s really more of a synthesis.
The analysis is always inaccurate to some degree. That’s because the observations themselves are not perfectly accurate, and because they don’t provide complete coverage. The latter problem is particularly severe for a country like Britain, most of whose weather comes off the empty expanses of the Atlantic Ocean. Up until the 1970s, Britain and other countries maintained a fleet of about ten weather ships that were stationed at various positions in the North Atlantic. Between bouts of seasickness, the crew would release balloons that ascended to the stratosphere, radioing back the atmosphere’s vital signs as they climbed. The ships gradually fell victim to fiscal belt-tightening, however, as well as to the belief that satellite-based meteorology was the wave of the future. That was a mistake, according to Jung, because satellites have a hard time measuring conditions at different vertical levels in the atmosphere. At any rate, by the time of the 1987 storm weather ships were history. Surface observations were restricted to those provided by a few automated buoys and by whatever commercial ships happened to be in the area of interest. These buoys and ships didn’t release weather balloons. What’s more, some of the commercial ships were so notorious for providing inaccurate data that they were put on an official blacklist of untrustworthy sources. Some of the missing data was supplied by commercial aircraft, but most of the transatlantic jetliners travelled to the north of the Bay of Biscay – the birthplace of many European storms, including the one of October 1987.
As if these problems weren’t bad enough, another factor may have made the data-acquisition process even worse than usual during the days leading up to the 1987 storm. Michael Fish told me that the meteorologists in France were on strike, and as a result they failed to send weather data from their country to the Meteorological Office (Met Office) in Bracknell near London, which was then the nerve centre of British weather forecasting.
Bill Giles, who was Fish’s boss at the BBC Weather Centre, confirmed this account to me. There is no mention of any such strike in the news reports that I read, however, nor in the official report that was published in the aftermath of the storm, so I sent an inquiry about it to Météo France. I got a reply from Jacques Siméon, who was head of the forecasting division there until 1987. He told me that a one-day strike of the entire French Civil Service was called for October 15, but that few people in Météo France heeded the call, and as a result there was no effect on forecasting or the transmission of data.
Having developed a representation of the current conditions – an analysis – forecasters have to predict how things will change in the future. By 1987, the Met Office had largely consigned this task to computers. In fact, the Bracknell centre already possessed a ‘supercomputer’ – a CDC Cyber 205 – capable of executing 200 million calculations per second. Still, this is only about one-millionth the speed of the fastest present-day supercomputers. Even at the time, it was outmoded in comparison to the Cray supercomputers being operated by the French meteorological service and by the ECMWF in Reading.
The Bracknell forecasters used two different models to transform the analysis into a prediction of future weather. To simplify the computing process, both models represented the atmosphere as a three-dimensional grid of boxes, each of which was assigned a single value of temperature, pressure and so on. In one model, the boxes were about 60 kilometres on a side. This so-called ‘fine-mesh model’ was restricted to the United Kingdom and surrounding areas. The other, coarser model used boxes of about 120 kilometres on a side. This ‘UK global model’, as it was called, covered the entire Earth. Both models divided the vertical dimension of the atmosphere into about 20 levels. (By comparison, the most detailed present-day models use boxes as small as four kilometres on a side, and they divide the atmosphere into about 70 levels.)
Applying the equations of fluid dynamics to the data in each box and its neighbours, the supercomputer changed the data in each box to represent the passage of some small increment in time. This was done repeatedly until the computer had arrived at the time of interest – say, 12 or 24 hours after the starting time. The result was a data set that human forecasters could use to draw charts and predict future conditions.
Several days before the storm struck, it was already clear that bad weather was heading toward Britain and France. Starting on Sunday, October 11, television and radio forecasts warned that wet and stormy conditions were likely toward the end of the week. This was not good news; Britons had experienced exceptional amounts of rain over the previous few weeks, with flooding in low-lying areas throughout southern England. By Wednesday, the actual weather over England still gave no hint of trouble to come, but the warnings handed out by the Met Office had grown more urgent. The lunchtime BBC TV forecast on that day included mention of a low-pressure system that ‘is going to deepen like mad and head up and give us an angry spell of weather, wet and windy’ on Thursday and Friday.
Even while the weatherman was speaking, however, things were starting to go wrong back at the weather operations centre in Bracknell. When the fine-mesh model was applied to the data for noon on Wednesday, the result was – no storm at all! A broad area of low pressure would cross the country and might drop some rain, but winds would be light. The global model did still show a storm, with a track crossing southern England, but the low-pressure centre was not deep enough to signify that exceptionally strong winds were on the way.
Meanwhile, out in the real world, conditions were rapidly deteriorating. By Wednesday night, the trough of low pressure in the Bay of Biscay had developed into a classic extra-tropical cyclone, and it was racing toward north-western France and the western approaches to the English Channel. Its central depression was deepening rapidly, even explosively, and the polar air that had been heading south was whipping around the depression in a counter clockwise direction and heading back toward the northeast. Ahead of it was a great mass of warm, subtropical air that had been bitten off and now formed an isolated, onrushing lobe of heat and moisture. Thus there were two fronts heading north-eastward: a warm front marking the leading edge of the subtropical air and, a couple of hundred miles behind it, a cold front marking the leading edge of the pursuing arctic air. (An animation of the advancing weather system is available online – see the list of sources for this chapter.)
At about 2am on Thursday, the Met Office’s Cyber 205 began chewing on the data that described the weather situation at midnight, using the fine-mesh model. After an hour or so, it spat out the result: yes, the depression would be deep enough to cause stormy conditions after all, but the centre would track up the English Channel, perhaps grazing the very tip of Kent in southeast England, on its way to the North Sea and the Low Countries. This seemed like good news for Britain, because it meant that the right-hand side of the depression, where the strong winds were likely to occur, would affect the English Channel and northern France, sparing the British mainland.
This tendency for the strongest winds to occur to the right of the centre of an extra-tropical cyclone is a phenomenon that is also seen with hurricanes. It is caused by the fact that, on the right side, the speed of the entire advancing system adds to the speed of the cyclonic circulation, whereas on the left side it subtracts from that speed. This effect is not very pronounced with hurricanes, whose eyes typically progress quite slowly – perhaps 10 mph. But extra-tropical cyclones move much faster – the centre of the cyclone that caused the Great October Storm raced across Britain at more than 50 mph. This caused a very large difference between wind speeds on the left and right sides of the storm track: to the left, gentle breezes wafted from the northeast; to the right, hurricane-force gales blew from the southwest.
Having finished its work on the fine-mesh model, the Met Office’s computer turned its attention to the global model. Because it was now about 3am, some more recent data could be fed into the model that had not been available for the running of the fine-mesh model. Whether on account of the new data, or for some other reason, the output of the global model was quite different from that of the fine-mesh model. It predicted that the centre of the depression would veer left, make landfall in southwest England about midnight, and cross the country well north of London during the small hours of Friday morning. What was more, the model predicted that the pressure at the centre of the depression would be 965 millibars. This was 48 millibars below mean atmospheric pressure and, in combination with the predicted steep pressure gradient to the south of the advancing centre, it would be enough to generate very strong winds in the region of England to the southeast of the storm track.
Thus, when the Bracknell forecasting team for Thursday, October 15 came on duty, they were faced with a quandary: their two computer models predicted utterly different conditions for the following night. Either southeast England would experience a major windstorm, or it wouldn’t.
This was a situation where the task of weather forecasting was suddenly thrown back into the laps of human beings, and specifically into the lap of the senior forecaster on duty that day. (The position is now called chief forecaster.) I haven’t been able to find out that person’s name. Ewen McCallum, a senior forecaster who was present in the operations room on that day (but not on duty), wouldn’t tell me; others I spoke with professed not to know, and the reports issued after the storm didn’t identify him. Evidently, there has been some desire to protect the identity of the person who (in terms of his official status at least) was most responsible for the erroneous forecast. As McCallum commented, ‘it was a there-but-for-the-grace-of-God-go-I kind of thing.’
McCallum did give me one insight. The then head of the Central Forecasting Office (and thus the boss of the senior forecasters) was a man by the name of Martin Morris. During the previous week, McCallum told me, Morris had been sceptical of the idea that a storm was on the way, in spite of the models that indicated that it was. ‘He stressed to me that he was worried that we were perhaps overdoing things [in our forecasts],’ McCallum said. On Thursday, Morris got involved in discussions with the forecasting team. ‘I have no idea how much pressure he put on the senior forecaster, but I certainly think there were discussions taking place,’ said McCallum. This seemed to be a broad hint that Morris urged the senior forecaster to opt for a milder forecast than he might otherwise have done, but I haven’t been able to confirm this story from other sources.
In any event, the forecast that went out to the media on Thursday morning was a compromise between the results of the two models. It described the depression as likely to move up the Channel and across southeast England, with very high winds in the Channel and windy, but not exceptionally violent, conditions in the south-eastern counties. It also spoke of the likelihood of considerable rain.
It was on the basis of this guidance that Michael Fish issued his infamous ‘no hurricane’ forecast. Although many sources, including the Met Office’s own website, describe Fish’s forecast as having been delivered on Thursday evening, it was actually given at 1.25pm as part of the BBC’s lunchtime television news. This lunchtime broadcast was not very widely watched. Thus, when Fish told me that he was ‘not even on duty at the time’, he presumably meant that he didn’t give the more popular evening forecast – it was his boss, Bill Giles, who handled that one. Still, it was the crucial two sentences from Fish’s broadcast, endlessly replayed during the days and weeks after the storm, that became emblematic for the Met Office’s failure to properly predict the storm.
I was curious to know whether Fish regarded his job as requiring him to help develop weather forecasts, or whether it was more a matter of presenting forecasts generated at Bracknell. He was emphatic that his role included actual forecasting, and that all the BBC weather presenters were trained Met Office scientists. Knowing where I was calling from, he couldn’t resist a dig at his US counterparts. ‘The situation in America is pretty appalling,’ he said. ‘The weather should not be entertainment; it’s a life and death kind of thing.’ Bill Giles made the same point: the BBC presenters were independent forecasters, he said. He, too, bemoaned the trend, especially evident in the United States, toward weather forecasts as entertainment. ‘Television wants nubile young ladies,’ he said, ‘and meteorology likes fat old grey men with experience.’
Fish’s and Giles’s disparagement of the US television forecasters may not be entirely fair. A good number of American weather presenters, probably including some of the nubile ones, have undergraduate or advanced degrees in meteorology or related disciplines, according to the American Meteorological Society. Fish himself learned his trade while serving tea to the weather forecasters at Gatwick Airport, starting in 1962. He doesn’t have any college degree, although in a 2004 interview posted on the website of London’s City University he was quoted as saying that he earned a degree in physics at that institution in 1968.*