Why are all weather apps different and which one should you trust?

Daily Telegraph Online | Tom Hoggins | 23 July 2019

If there is one thing that weather forecasters can agree on, it is that the UK is in for one of the hottest weeks on record.

Temperatures will soar in mid-week, topping 30 degrees in some parts of the country. Sweltering City workers that may be looking forward to a picnic in the park on Saturday to soak up the sun, meanwhile, look certain to be washed out.

Some weather forecasts are universal, to a degree, but even in this most extraordinary of weeks there are points of contention between different institutions.

The Met Office and the BBC seem to believe that the skies will be clear until showers on Friday. Norwegian app Yr is predicting light showers amid searing heat on Thursday afternoon. But AccuWeather is forecasting a thunderstorm as early as Wednesday morning, followed by a clear Thursday and more thunder on Friday.

So who to trust? And why are they so different anyway?

"All weather forecasts, no matter which app you look at are based on running "Numerical Weather Prediction" (NWP) models," says Dr Pete Inness, Associate Professor in Meteorology at the University of Reading.

"These are computerised models of the atmosphere which include all the fundamental laws of physics which determine how the atmosphere changes in time."

And with the inexorable advancement of technology, we now have more access to details on temperature, wind speed, cloud movements and rainfall than ever before. In tandem with readings on atmosphere taken on the surface by meteorologists and automated weather stations, there are now many newer tools used to collate more data.

There are radar networks dedicated to reading rainfall, countless weather satellites monitoring the Earth, aircraft readings, AI 'robots' used to clean up imagery to more accurately predict precipitation.

This glut of data is fed into supercomputers able to make trillions of calculations a second, which then forecasts several outcomes based on different variables. The most likely weather fronts (if a storm appears at the same time on the various outcomes, for instance) are then extrapolated from there.

"Most providers now will use output from several models, but each will be using a different combination. Some providers use ensemble forecasts as well, many do not yet." says Ken Mylne, Head of Verification at the Met Office. "The Met Office is one of several global forecasting centres running global models and ensembles, and also runs high resolution regional models and ensembles for the UK.  As a forecast and app provider the Met Office bases it forecasts primarily on Met Office models, but also blends data from other centres."

Mylne also sayas that most forecasters perform some "post-processing" to refine outputs, but says that process is not simple either.

"Different providers will have many approaches to dealing with these challenges, leading to differences in the forecasts they issue. Different approaches will work better under different situations, so it is very unlikely that any one provider is always the best."

Different models, different forecasts

With such processing power (the three most powerful computers in the UK are owned by weather forecasting centres) you might expect that the margin for error is becoming vanishingly small. But that just isn't the case. While the power of today's supercomputers has hugely improved the accuracy of forecasts, the scores of different processing models used around the world differ in their methodology and algorithms.

"All of these models include the same laws of physics, but may differ in the amount of detail they include, the way the computer solves the equations and the weather observations that are used to set the starting conditions for the forecasts," says Inness. "Different forecast models will produce similar forecasts, particularly at the start of the forecast period, but the details may differ. Two models may both be predicting showers across the UK during the day, but the exact location and timing of those showers may be different."

There are also differences in how apps display the information they are provided from their forecasts models. Inness gives the example of an app having access to four different forecasts, three predicting rain and another keeping it dry. The app may then display the symbol for a shower, but with the probability of 75pc. Another provider might only have access to the dry forecast.

Different apps also have different thresholds for deciding on whether to display a shining sun or thunderstorm symbol. While a thunderstorm may only be 20pc likely to happen, one app may decide that users need to be warned of the possibility. Another may only display the thunderstorm symbol at 30pc.

All of these fine margins and small details add up to a wide variety of forecasts from the countless apps that fill up smartphone stores. These variables and the volatile nature of the Earth's atmosphere mean that even the most reputable organisation will never be able to guarantee 100pc accuracy. Even if the computers could be brought into line to provide consistent forecasts across the board, Mother Nature might have other ideas.

"Model evolution can be very sensitive to small errors in the starting conditions," says Mylne. "This is a classic example of chaos theory. A good way to explain that is if a butterfly flaps its wings in one part of the world, it may have an effect on the weather in another part of the world."

The perfect forecast?

There are apparently two hundred tredecillion (200,000,000,000,000,000,000,000,000,000,000,000,000,000,000) molecules in our atmosphere in random motion and a computer would need to be capable of modelling them all to even approach fully accurate weather forecasts. That's even without the inevitable changes to the atmosphere that come as part of the natural world.

"Forecasts differ partly because the atmosphere is turbulent and chaotic," Paul Williams, Professor of Atmospheric Science at the University of Reading says.

"Try predicting the swirling vortices of cream when it is poured into your coffee, and you will immediately get a feel for why weather forecasting is so hard."

Mylne agrees, particularly when it comes to pinpointing thunderstorms and rain. "A good analogy is a saucepan of boiling water, and trying to predict where the next bubble will appear," he says. "There is a large element of randomness to where individual showers will develop."

However scientifically improbable the perfect forecast is, it seems unlikely that this will stop the time-honoured British tradition of complaining about the weather and the forecast. As for weather apps we can trust, it seems the best recourse is consulting a handful. And maybe hedge your bets on the way out.

"I personally consult several different forecasts from a variety of providers every day," says Williams. "If at least one of them indicates rain, then I take my umbrella. With this policy, it has been a long time since I got caught out by a shower."