It’s raining. Your weather app says it’s sunny. Why?

“THE WEATHER APP IS INACCURATE ! Why is this problem still not fixed?!” reads one of many similar complaints that have plagued Apple’s community discussion board for years.

Unhappy users include a commuter caught, sans umbrella, in drizzle after work last week, despite being sold zero millimetres of rain when she left her home in Melbourne’s inner north that morning. Work-from-homers in Sydney’s North Bondi found themselves equally aghast on Friday afternoon, their lunchbreak swims dampened by a bucketing forecast as nothing more than cloud cover.

Can we trust our default weather apps?

The applause that met what Steve Jobs and his trademark turtleneck promised to Macworld’s packed auditorium in 2007 – a revolutionary new handheld device with a pre-installed weather app – could have been the “just in case” umbrella’s death rattle. Yet almost 20 years after the iPhone’s debut, we’re still packing precautionary parasols.

To be fair, Android apps also weather their fair share of sprays online – “Samsung Weather is now garbage, weather app recommendations?” is a six-year-old Reddit thread’s title; “Pixel Weather App Called Out for Being Super Inaccurate” a headline from January – but Apple’s never quite recovered from its snow and sleet at 17°C in Sydney moment in April 2015.

Last month, users in St Louis, Missouri, were forecast another impossibility: 72°F (22.2°C) and snow.

How does this keep happening?

“It’s just whatever the numerical weather prediction model has spit out,” says Dr Kerryn Hawke, an atmospheric scientist and lecturer at Murdoch University in Western Australia. “It hasn’t gone through that quality checking process that gets done at the national weather services.”

Forecasting, Hawke says, is a science but “it’s still an art ... it still needs the meteorologist with the in-depth knowledge of individual places to … apply their understanding to improve those forecasts.”

How do weather apps work?

For severe weather, Apple says it uses national weather service information. Despite some recent high-profile misses from Australia’s Bureau of Meteorology (BoM) – and “climate whiplash”, plus US President Donald Trump’s climate cuts, sparking concern over the future of reliable forecasting – internal data says 80 per cent of the BoM’s flood peak forecasts were accurate within 0.3 metres in 2024-25.

In that same period, the BoM says its next-day maximum temperature forecasts had a 91.3 per cent accuracy rate (within 2°C), and overnight minimum temperature forecasts 82.8 per cent. Its rainfall range forecasts are considered reliable. How much Apple relies on that data, however, is unclear.

About 670,000 routine forecasts are issued annually by the BoM, and meteorologists in its employ, using its own global numerical prediction model, the Australian Community Climate and Earth-System Simulator (ACCESS).

Apple’s 10-day forecast, maps, next-hour precipitation, and air quality features use a mix of information from US-owned The Weather Channel, and its own proprietary WeatherKit (formerly Dark Sky) modelling, which amalgamates data from multiple sources and uses an algorithm to interpret it.

Google, which creates its forecast using an internal algorithm that ingests weather models and observations from global weather agencies, does not list the BoM as a data source. Samsung uses The Weather Channel.

Modern-day forecasting is computer-driven, and more accurate than you might think. Paradoxically, default smartphone weather apps’ overreliance on algorithms to interpret data – instead of using local meteorologists, as the BoM does – can contribute to its inaccuracy.

“Every single model does not represent reality perfectly,” says Hawke. “There’s always going to be something that’s not quite right.”

Thunderstorms, for example, are small-scale, very high-temporal-resolution events; they occur over a short time and a short space. They can happen between weather stations, meaning apps without a meteorologist on deck can miss them or get the time wrong, as they change rapidly.

Models also perform poorly in Hawke’s native New Zealand, she says, as they don’t account for the microclimates induced by its rugged topography. One side of a mountain may have different weather to the other, but a model may not understand that change in terrain between weather stations the way a local expert does.

‘It’s a science, but it’s still an art.’

Dr Kerryn Hawke on weather forecasting

As senior research fellow at Monash University’s ARC Centre of Excellence for 21st Century Weather Dr Michael Barnes puts it: “A forecaster sitting in Tasmania would know ‘If I get 20mm of rainfall over this catchment, that river’s going to flood’. Maybe AI will be able to do that one day or assist in that process. But I don’t think we’re there yet.”

Are we actually the problem?

Barnes says “the human element” is crucial to accurate forecasting. Human error, however, is also part of the problem. What “hurts us … what may confuse people”, Barnes says, “is what is possible with these weather forecasts”.

Italian science communicator Dr Federica Zabini first floated this in Meteorological Applications in 2016: excessively detailed forecasts supplied by apps, without sufficient disclaimers regarding the fact that weather forecasting is a game of probabilities, create unrealistic expectations of accuracy that damage public confidence.

Tomorrow’s displayed 29°C maximum temperature, for example, actually means there is likely a 50 per cent chance it will be 29°C, with the range of probabilities between 27°C and 31°C.

While the five to seven-day forecast is considered as accurate as the three-day forecast 30 years ago, default smartphone weather apps publish 10-day forecasts (the BoM, citing accuracy, prefers seven-day forecasts).

Each day increases the range of probabilities, diminishing the chances of the one outcome shown on an app actually occurring.

“You don’t actually know where the forecast is tending towards because you’re only getting one of the outcomes,” says Barnes. “You’re not getting all 200 possible outcomes that we know about.”

So which weather app is the best?

Birdwatchers, surfers and boaters habitually check WillyWeather, while teachers who don’t use the BoM app or Weatherzone swear by Windy or Rain Parrot.

Barnes says accuracy is “very much in the eye of the beholder”, depending on what aspects of a forecast – temperature, tide – you look at. He cuts out the middle man and makes his own forecasts from the BoM website, as does Hawke, who also likes to read the sky.

“It’s not that I don’t trust weather apps … I don’t think they provide everything, all the nuance that is needed,” she says.

If all else fails, look up. And pack an umbrella, just in case.

Hear the story behind the headlines on The Morning Edition podcast, every weekday from 5am on Apple, Spotify or your favourite podcast platform.