WHEN you need a weather forecast you probably go to an internet site or an app on your phone and plan accordingly. Mistake number one. Never place your trust entirely in a single prediction.
The best way to illustrate this is to take a close look at what is on offer. There are at least 20 websites offering weather forecasts, about 50 apps for each of Android and iOS, and every newspaper, TV and radio news programme will have one.
The five-word radio summaries are useless and newspaper forecasts are well out of date before you read them. The phone and computer-based sites all follow the same pattern: symbols to show the type of weather expected, plus details which range from basic to more or less relevant to impossibly comprehensive, the latter intended to make you think their forecast must be pretty good. Mistake number two: more detail does not imply more accuracy.
Where do all these predictions come from? The big computers produce several forecasts with slightly different starting points because we still can’t tell them enough detail about the initial state of the atmosphere. The difference between these forecasts will indicate the degree of uncertainty in the output. No forecast is perfect, therefore, and the result is that small errors can grow in proportion to the time span of the forecast. Tomorrow’s prognosis will always be more useful than that of next week, and so on.
Once the forecast is agreed it is sent through the software mangle and stretched to cover every hour of tomorrow, next week, next month, for every city and town in the world. The result is that Wissleby-on-the-Wolds does not get a simple message such as ‘next Tuesday will be cloudy with occasional showers’. Instead the village folk will find something like ‘At 2 pm next Tuesday there is a 60 per cent chance of precipitation, 80 per cent cloud cover, NW wind gusting to 18 mph, temperature of 13 deg C which feels like 12, 85 per cent humidity, and 0.6mm of rain.’
On top of that kind of detail many forecasts will add dew point, air quality, visibility, cloud ceiling, times of sun and moon rise and set, and whatever else they can think of. That first version for Wissleby stands a very good chance of being correct. The second one will almost inevitably be wrong. It is simply too specific, both meteorologically and geographically.
If your app says there will be a 37 per cent chance of rain at 3 pm in 12 days’ time (and some of them do), you can be sure that the impetuous software chap has overruled the cautious meteorologist. If your app gives you the forecast for every day until January 2024 (yes, there is one) or offers a cloud amount and the exact quantity of rainfall (in tenths of a mm) for every hour over the next three weeks, you know the meteorologist has gone home in despair.
What use is all that to someone who simply wants to know whether it will rain or not? These predictions follow the Williams Laws of Forecast Confusion. First Law: No two will ever be the same, whether on websites, apps, TV, radio or newspapers, unless they are short statements in plain language. Some of this is possibly due to the major companies using different super-computers. Second Law: More detail does not imply more accuracy. Third Law: Tomorrow’s forecast is doubtful; after that they are even more dubious.
The Third Law came about because I discovered that even today’s forecast can change before your morning coffee break. As I said before, ‘cloudy with occasional showers’ would nearly always be accurate, so would ‘rain spreading from the west’. But to tell Wissleby there will be a ‘40 per cent chance of light rain at 10am tomorrow’ would almost amount to fraud except for that ‘40 per cent’. This guarantees that they can never be wrong. If it rains, the forecast was right, and if it doesn’t, it was still right. Whoever thought up that percentage idea made forecasters immune to any future grumbling.
Accuracy? Very subjective. Possibly the most important aspect of the future weather we need to know is: will it rain? If so, when and for how long? Alas, mostly all we have is a likelihood of the first, and a do-it-yourself interpretation of the rain radar sites (which I wrote about here) for the second, though some will give you slightly more detail on the day itself.
One way to check these apps is to wait for an approaching frontal system coming in off the Atlantic. The day before, note what time the rain will start tomorrow according to a selection of your favourite apps and websites. I tried this on nine of the better-known ones on Wednesday and Thursday this week. The forecast on Wednesday for the time of arrival of the rain on our market town the next day was variously given as 10am, 11, midday, 1pm, 2pm and 3pm. Six different out of nine. The answer was it began at 10.45, so only two of them were nearly right.
‘Prediction is very difficult,’ said quantum theorist Neils Bohr, ‘especially when it is about the future.’