AI weather startups claim more accurate forecasts than Bureau of Meteorology

AI weather startups claim more accurate forecasts than Bureau of Meteorology
  • PublishedJune 8, 2024

Mulgowie Farming Company agronomy manager Andrew Johanson was faced with a predicament as rain clouds rolled in off the Pacific last month.

The Bureau of Meteorology (BOM) was predicting 16 millimetres of rain — not a lot, but enough to cause machinery to bog in a paddock. 

If it was accurate, he needed to race to finish harvest and get his machinery back in the sheds. 

But he didn’t know if he could trust it.

Mr Johanson said it had been a common predicament for the company in the past 12 months as forecasts repeatedly failed to accurately predict the weather.

“This year was supposed to be a hot, dry year, and it ended up the wettest, muggiest year we’ve had for many years,” Mr Johanson said.

Andrew Johanson
Andrew Johanson grows corn and green beans at Mulgowie.(Supplied)

Mulgowie, nestled in Queensland’s Lockyer Valley, produces corn and beans for major supermarkets.

Mr Johanson said it ran to a tight planting and harvest timetable. 

He said the company had lost several crops this year due to the wet weather that he would not have planted if he had accurate forecasts.

“It’s not just the amount the crop has cost you to grow and get to that point, which is quite a bit, it’s the loss of sale,” he said.

“It’s hard to keep the customers happy if you aren’t consistent with your supply. 

“And if we get [inaccurate] forecasts, it can really affect our ability to do that.”

Localised forecasts

Perceived inaccuracies in traditional weather forecasting have prompted tech startups to rush in to what they hope will be a lucrative new market — offering precise, hour-by-hour forecasts for areas as small as the size of a football field.

Mr Johanson said he had been trialling a private weather forecasting service, Jane’s Weather, for the past four weeks.

It claimed to use traditional weather modelling and tap into the private weather station on the Mulgowie property to generate a forecast using machine learning.

Mr Johanson said the service had been more accurate than BOM forecasts at predicting rainfall totals.

“It is taking into consideration our local microclimate,” he said.

Lightning strike and storm clouds in the distance at a farm at drought-stricken Dululu, west of Gladstone.
Farmers rely on accurate weather forecasts to run their businesses.(Supplied: Neal Johansen)

Reliance on BOM

Monash University ARC Centre of Excellence for Climate Extremes research fellow Michael Barnes said the number of private weather companies had been growing rapidly over the past few years.

He said they were making money by filling in the “gaps” and providing additional information that might not be available through national services such as the BOM.

But he said the technology would not function without national forecasts.

“All of them rely completely or at least in part on the products of national meteorological and hydrological services [such as BOM],” Dr Barnes said.

“National meteorological and hydrological services are and will always be critical and provide an invaluable service to the community at large.”

Giles Weather Station in remote Western Australia.
The Bureau of Meteorology provides regular forecasts, warnings, monitoring and advice across Australia.(ABC News: Nick Hose)

In a statement, a spokesperson for the Bureau of Meteorology defended the accuracy of the organisation’s forecasts.

“The bureau’s forecast accuracy has consistently ranked in the top five in the world,” the spokesperson said.

“The bureau’s Australian weather model (ACCESS) ranks in the top four alongside the European Union, United Kingdom and United States. 

“On a seasonal timescale, the bureau’s system is recognised as one of the best in the world.”

The spokesperson said the bureau had been “proactively and safely engaging with artificial intelligence (AI) capabilities for several years”.

“This area of research is one of many initiatives the bureau actively pursues to improve its services to government, emergency management partners and the community,” the spokesperson said.

AI ‘revolution’

Speaking at a forum organised by the US-based Environmental and Energy Study Institute in January, Columbia University geophysics professor Pierre Gentine said artificial intelligence was ushering in a “second revolution for weather forecasting”.

“What people have shown recently, over the last year … is that those AI based models can do a better job than weather forecasts of physically based models that we’ve been using for the past decades. And that’s really exciting,” he said.

But he said AI-based models did not work for long-term forecasting.

“They do a great job for a couple of days up to two weeks, but then they do actually a very poor job,” he said.

“They are actually worse than just the mean seasonal cycle.”

One of the new generation of weather forecasters, US-based, turned on two weather radar satellites earlier this year, and plans on launching a new “constellation” of satellites in November, which will feed data to a machine learning algorithm which claims to offer accurate forecasts to areas the size of an airport runway. satellites
US-based company launched two weather radar satellites in 2023.(Supplied) chief marketing officer Dan Slagen said the company was banking on increased demand for their services as weather events become more extreme under climate change.

“We’re seeing major events happening on the front pages of every newspaper every day,” Mr Slagen said. 

“And so every business is facing more of this.” 

Early Warning Network chief executive Kerry Plowright, whose company offers a weather forecasting app for Australian users based on data from’s satellite service, said with prices starting at more than $2,000 per month, the cost was “probably too much for an individual”.

He said interest in the technology was coming from freight and logistics companies, councils, construction companies and government departments.


Leave a Reply

Your email address will not be published. Required fields are marked *