tracking not only the forces that drive drought but the damage it leaves behind has revealed that droughts have lasted longer in Australia in recent decades, especially in areas with the most people and farms.
UNSW researchers analysed drought trends across Australia between 1911 and 2020 based on rainfall shortages and falling river and dam levels.
Their analysis showed that, since 1971, the time spent under drought conditions has increased across most of Australia, especially in the southeast and southwest, which are densely populated and key breadbaskets.
The increasing dryness was especially felt during winter and spring, which are critical seasons for growing crops like wheat.
Lead author, UNSW PhD student Matt Grant, says the trend, which is expected to continue, is especially concerning because the regions where drought is more severe are also some of the most important to Australia's economy and people's welfare.
"Across early parts of the last century, it was wetter, but we can see that pattern has reversed since the 70s.
"In recent years, we've had this general sense that droughts are lasting longer. Our work puts data behind the intuition and confirms it.
"We expect this trend to continue, which increases water security risks for some of Australia's major cities and rural towns and puts serious pressure on farming."
He says it's not yet possible to say whether these conditions are caused by climate change, because Australia has historically had large variation in rainfall.
"It's not to say that climate change isn't influencing drought, it's that the natural swings in Australia's climate are so large that it may take longer for a clear signal to emerge from the noise."
The AI unravelling Australia's drought secrets
The researchers used AI to find relationships between these conditions and hundreds of official drought records of crop losses and threats to important water reservoirs in the southeast of the country.
Co-author Dr Sanaa Hobeichi, who pioneered an earlier "impact-based metric" in Texas says the approach helps answer long-standing questions in drought science.
"It wasn't really understood how closely weather was reflected in actual drought impacts, but we could show they clearly mirror one another," she says.
"The model links all the steps of a drought, from a lack of rain to documented consequences."
She says the research also shed new light on some of the less visible drivers of drought.
The study shows that what is happening in the soil and atmosphere is usually a better indicator of drought risk than rainfall across a month.
Evaporation on hot, dry days was the strongest predictor of crop failure, particularly in the driest regions like the Rangelands, which cover more than 80% of Australia's total landmass.
Whether or not water levels would run low, in turn, often hinged on soil moisture, because very dry soil soaked up rain before runoff could reach rivers and dams.
Dr Hobeichi says the results build on a growing body of research showing that drought risk is incredibly complex and can't always be predicted by major weather systems like El Niño.
"Different droughts have different profiles," she says. "AI lets us understand the complex dynamics of each one."
"We used what's called a Random Forest model, because it doesn't just look for one simple answer, like 'if rain is low, it's a drought'.
"It creates an enormous number of decision trees, which ask whether certain conditions have been met in terms of ocean systems, temperature or crop failure, and in complicated combinations.
"At the end, it combines all these different outcomes to give you the most likely result."
She says there is growing interest in developing impact-based approaches to better manage future climate risks.
"People want metrics that reflect conditions on the ground, not just anomalies in a single variable, because a lot of the risks we're facing are cascading and compounding."
However, she says, the AI would need further research and testing to adapt it for use as a tool to assess drought risk in a hotter future.
"Rainfall patterns, soil moisture and ocean drivers will all shift under climate change, altering the 'recipe' for drought.
"Any future applications will require stress-testing the model under conditions that resemble projected climates."
Even so, she says, the model has real potential for climate resilience planning.
"The AI could one day tell us not only when conditions are dry, but when it's going to start affecting people's lives, letting us make better decisions about where we need to allocate resources to adaptation."
Dr Hobeichi is part of a research team currently working on a nationwide drought-impact database, extending beyond NSW to all major cropping regions.
