Kerry Plowright was watching television one evening late last year when her phone alerted her that it was full.
“When I walked out the door I was stunned because there was just a roar,” he says, describing the sound of hailstones hitting rooftops in the New South Wales town of Kingscliff. He had just enough time to protect his cars from damage by moving them under canvas sails.
Plowright is not alone in Australia’s seemingly relentless summer of extremes with little warning ahead of severe weather. This season could include a second tropical cyclone to hit Queensland.
Relating to: Hurricane forecast supported by artificial intelligence provides earlier path tracking
The Albanian government has launched an investigation into warnings issued by the Bureau of Meteorology and emergency authorities following complaints from councils and others that some warnings lacked accuracy and timeliness.
But Plowright’s situation is a little different; The hail warning was triggered by data generated by his company, Early Warning Network.
The Early Warning Network analyzes data from radars and remote sensors to detect and warn of extreme temperatures, precipitation and flooding. Its customers include local councils and major insurers.
Private companies have long offered services based on data from the BoM or organizations such as the European Center for Medium-Range Weather Forecasts (ECMWF). But the Early Warning Network is starting to test artificial intelligence models that promise to deliver much more weather information both quickly and at low cost.
“You have to pay a bucket load [ECMWF] data,” Plowright says. “We no longer need a supercomputer to provide a highly accurate forecast for up to 10 days, especially in extreme weather.”
He predicts that AI “will be absolutely phenomenal at weather and eventually climate as well when it gets to that point.”
How can AI help us prepare for extreme weather?
Water resources engineer Juliette Murphy is similarly excited. He founded FloodMapp to give communities more time to prepare after observing devastating floods in Queensland’s Lockyer region in 2011 and in Calgary, Canada two years later.
FloodMapp uses traditional physics-based hydrology and hydraulic models as well as machines that learn from each model run. Even relatively simple computers can quickly scan “really large data sets” to determine the likely effects of a flood, he says.
Its customers include Queensland’s fire and emergency services. Its results complement those of the BoM, helping authorities decide which homes to evacuate and which roads to close. “This is especially important because nearly half of deaths from flooding are from people in cars,” says Murphy.
Relating to: Flood warnings prompt evacuations in NT as Queensland braces for incoming cyclone
A BoM spokesperson said the bureau had “been proactively and safely engaged in AI capabilities for several years”.
“This area of research is one of many initiatives the bureau is actively pursuing to improve its services to government, emergency management partners, and the community,” he says.
Computer scientist Justin Freeman led BoM’s research team working on machine learning before leaving to start his own company, Flowershift, in late 2022.
Flowershift creates a geospatial model trained on existing observational data. “We will fill in the gaps in what the current forecast products are,” says Freeman, such as providing forecasts for remote parts of Australia or beyond.
“There’s a lot more flexibility to be able to explore things [outside BoM] and we use very new technologies,” says Freeman, who still does contract work for the office. “We have completely new, different model classes that are completely different from the others. [the bureau had] “We have been running for the last 50 years.”
Models that can inexpensively analyze data and then provide localized information have many potential uses. For example, farmers may ask: “Should I spray my crops this week?” Freeman says he will be told why or why not.
“It hasn’t been that long before we’ve had access to something like ChatGPT,” he says. “Look ahead, like two more years, five more years, and it will get faster and better and better.”
Limitations of artificial intelligence
But some BoM and climate researchers point out how much AI-based models like Google’s GraphCast or Nvidia’s FourCastNet can improve numerical models that reveal a variety of possibilities.
Relating to: ‘Very scary’: Mark Zuckerberg’s promise to create advanced AI alarm experts
“To downscale ‘simple’ weather forecast and physical model data I guess [there’s] tremendous potential,” says an office scientist. “I would be very careful as it warns us of real dangers when the atmosphere gets violent.
“With climate change, we need to better understand things that are way outside the norm.”
Despite the limitations, there are still benefits, says Sanaa Hobeichi, a postdoctoral researcher at the ARC Center of Excellence for Extreme Climate.
Current climate models often offer only “rough” solutions, such as predicting precipitation changes in areas of 150 km by 150 km. For example, in Sydney, a model of this size would be able to display the city, mountains and more, and thus its use would be limited.
While Google’s GraphCast prediction model has a resolution of up to 28 km by 28 km, Hobeichi says some AIs can only model 5 km by 5 km.
But one challenge is that machine learning techniques inherit and potentially predict flaws in the traditional models they are trained on.
Postdoctoral CSIRO researcher Jyoteeshkumar Reddy Papari notes that ECMWF was initially skeptical of AI but has recently launched its own experimental model. It also displays several more on its own website, including Google’s.
“Countries that do not have good meteorological organizations rely on these machine learning models because they are extremely easy to learn and publicly available.” “So some African countries use these estimates.”
Google researchers claimed last year that GraphCast “significantly outperformed the most accurate operational systems” on 90% of 1,380 targets. Tropical cyclones, atmospheric rivers, and extreme temperatures were all better predictions from traditional models, and improvements continue.
Relating to: AI will impact 40 percent of jobs and likely worsen inequality, IMF chief says
“One particular example that we often talked about was Hurricane Lee, because it was the first time we observed in real time how GraphCast predicted the trajectory of a hurricane differently than traditional systems, and it was eventually shown to be the correct trajectory,” he said. Alvaro Sanchez-Gonzalez, Goggle researcher.
“Detected in real time and verified by independent sources.”
The current tracking of the potential hurricane in the Coral Sea (which will be named Kirrily if it forms as expected by Monday) will also be monitored to see how the models compare.
Matthew Chantry, ECMWF’s machine learning coordinator, says AI models are “a very exciting avenue as a companion system for traditional forecasting” but the latter has some advantages.
“Tropical cyclone intensity forecasts are a good example,” he says. “Whether these flaws will persist as the technology matures is an open question; it’s too early.”
Authorities act based on probabilities calculated by traditional models, but this requires a very large supercomputer. “With AI predictions, this is being reduced significantly, with some estimates suggesting the energy spent making predictions is reduced by a factor of 1,000. So cheaper systems could be a force for equality.
“This reduced cost can also be invested in larger communities, meaning we have a better idea of low probability but extreme events that could occur.”
But when it comes to predicting the effects of a warming planet?
“The problem is much more difficult than forecasting weather with less data,” Chantry says. “However, in a changing climate where evidence points to an increase in extreme events, any help in predicting these events is of significant value.”