But they warn early attempts to expand predictions using AI will not be perfect, and they may need to experiment with the technology to find its strengths and limitations.
Scientists, academics and infrastructure experts issued the warnings at a CSIRO event on Tuesday that discussed ways physical climate risks could be addressed using technology.
The discussion comes after a Victorian agency warned $57 billion of infrastructure such as roads, rail and hospitals could be put at risk by climate hazards, and after a report from the University of Sydney found climate change would raise prices for housing and insurance.
The Australian Climate Service, which uses data from the Bureau of Meteorology, CSIRO, Australia Bureau of Statistics and Geoscience Australia, had already started testing AI tools in climate forecasts, chief scientist Dr Judith Landsberg said.
The technology had huge potential to analyse large amounts of data related to climate risks, she said, but also presented challenges when verifying its results.
"We're exploring the use of machine-learning and AI throughout our value chain, starting from data production through to the development of insights and to advice," Dr Landsberg said.
"We still haven't really figured out what the greatest opportunity is, what the risks are and how we manage those risks in what we provide so that we can remain an authoritative source for other organisations."
AI could be used to fill in gaps in existing climate information, she said, such as generating thousands of possible tracks for tropical cyclones.
"The data set of tropical cyclones that we have experienced is quite small," she said.
"We get a couple a year, in a bad year you get five or six, and that observational data set doesn't give us the near misses, what if Tropical Cyclone Alfred had actually hit Brisbane, for example."
But one of the challenges of using AI tools for climate forecasting was its lack of transparency, CSIRO senior research consultant Dr Geoff Lee said.
Any AI tools used for climate modelling would need to reveal how it had arrived at its conclusions, he said, so they could be verified by scientists.
"Being evidence-based and data-driven doesn't mean we want a black box that tells us what the solution is," he said.
"We want to be able to be really clear about how we reached these decisions."
AI tools were also unlikely to provide reliable predictions in their first iterations, Dr Lee said, but that did not mean they would not prove valuable in the long term.
"It's not going to be perfect the first time you make these decisions," he said.
"We are experiencing climate change, we have an imperative to do things about it differently than we have been doing."
