The research, published in the International Journal of Disaster Risk Science, was led by David Karacsonyi, together with Andrew Taylor and Kerstin Zander, and it tackles a problem that has frustrated disaster scholars for decades: establishing whether an extreme weather event actually caused a population to decline, or whether the decline would have happened anyway. In sparsely populated regions, communities are routinely buffeted by forces that have nothing to do with weather, including mining boom-and-bust cycles, shifting public spending, and the constant churn of a highly mobile workforce. Any census-based analysis must therefore tease apart the signal of a hazard event from the noise of systemic economic volatility.

To do this, the team merged two hazard datasets. The first, the Disaster Events with Category Impact and Location database hosted on Australia’s Spatial Intelligence Network portal, contains records of 639 hazard events for Australia and the Indo-Pacific from 1753 to 2014, with details on injuries, deaths, and insured losses. The second, the Australian Disaster Resilience Knowledge Hub curated by the Australian Institute for Disaster Resilience, adds 728 more events with descriptions of timing, location, and impacts. After merging overlapping records and filtering for Northern Australia, the researchers identified 86 extreme weather events between 1870 and 2019, with roughly 65 percent overlap between the two sources.

These hazard footprints were then spatially intersected with the 181 remote settlements that existed in Northern Australia at the 2021 census, a set that includes 178 urban centres and localities in remote and very remote areas plus the spatially isolated towns of Darwin-Palmerston, Humpty Doo, and Howard Springs. Because a single cyclone or flood can strike several communities, the intersection produced 294 hazard impact locations. Against this, the researchers compiled 23 census data points spanning 22 consecutive intercensal periods from 1871 to 2021, generating 1,691 individual records of population change for the settlements involved. A decline was defined statistically as a fall in the census count between two consecutive periods while the later count remained above zero.

Three indicators were constructed to probe cause-and-effect relationships. The first measured the share of all recorded population declines that coincided with at least one hazard event. The second detected whether a shift from growth to decline occurred in the period bracketing a hazard, comparing trends before and after the event. The third calculated the proportion of all extreme weather events that coincided with a decline. These associations were cross-tabulated across three historical eras, three hazard categories, and settlement classifications based on socioeconomic profile and size, a design intended to guard against the ecological fallacy and modifiable areal unit problems that plague analyses aggregated at the wrong scale.

The headline result is striking. Although 38.9 percent of all intercensal periods for individual settlements recorded population decline, only 14.6 percent of those declines coincided with a hazard event. Meanwhile, 32.7 percent of hazard-affected periods saw decline, a figure significantly lower than the background rate of decline, with a Pearson’s chi-squared test yielding 5.76 on two degrees of freedom and a p-value of 0.0164. In other words, declines were actually more common when no extreme weather struck, suggesting that systemic issues such as resource-cycle economics drive the region’s overall demographic volatility far more than storms and floods do.

Yet the century-and-a-half timescale reveals a compelling evolution. During the frontier period from 1871 to 1947, hazard events and population declines were tightly linked: 40.6 percent of declines coincided with hazards, and decline following hazard events was more frequent than decline in general. Cyclone-related declines occurred in 45.5 percent of cyclone-affected periods in that era, and heat-related hazards such as droughts, heatwaves, and bushfires accounted for roughly a third of all population decline, reflecting the dominance of pastoralism in the regional economy at the time. By the postwar boom of 1947 to 1986, associations weakened dramatically as mining towns multiplied and Indigenous people were finally counted in the census. In the consolidation phase from 1986 to 2021, marked by fly-in-fly-out workforces rather than permanent mining settlements, demographic volatility peaked with 44.8 percent of periods in decline, yet hazard associations remained statistically insignificant. The pattern points to steadily increasing resilience, likely built on improved infrastructure, forecasting, governance, and adaptive capacity.

The case studies embedded in the data are vivid. Cyclone Yasi at Tully Heads in 2011, the Mount Isa floods of the 1973 to 1974 wet season, the 1907 cyclone at Cooktown, and Katherine’s floods of 1998 and 2006 all coincide with measurable population declines and divergences from previous trajectories. Darwin itself offers the most instructive paradox. Cyclone Tracy in 1974 devastated the city, yet the following census recorded growth, from roughly 37,000 people in 1971 to 44,000 in 1976, because reconstruction spending and public service expansion drew in workers and families. But the composition of the city changed permanently: around 60 percent of the cohort affected by the cyclone never returned, and Darwin emerged younger and more male. Compare this with the “Great Hurricane” of 1897, which produced a clear census-visible decline, or with smaller towns like Cooktown and Winton, which lacked Darwin’s reconstruction windfall and gradually bled population through repeated cyclones and floods.

Settlement type proved decisive. Diversified towns such as Longreach, Alice Springs, and Katherine, with older age structures and varied labour markets, were the most vulnerable to demographic disruption, with half of flood-affected intercensal periods recording decline compared with a background rate of 41.4 percent for that category. Indigenous communities, by contrast, appeared least disrupted overall, except by cyclones, whose timing and severity are unpredictable. The authors attribute this partly to the regularity of wet-season flooding, which these communities have long adapted to, and partly to the distinctive mobility patterns of Indigenous residents, who tend to move over short distances and periods, with a relatively low probability of permanent residential moves away from home communities. This raises a sobering concern: rather than migrating after disasters, Indigenous people may become trapped in place when at their most vulnerable, lacking the resources and networks to relocate.

Size mattered too, with volatility increasing inversely to population. Smaller urban centres and larger localities, places such as Cooktown, Wadeye, Kowanyama, and Roebourne, saw cyclones disrupt growth in 45 percent and 41.7 percent of affected periods respectively, while floods undermined growth in 40.9 percent of periods for larger localities. The authors caution, however, that their census-based method cannot capture settlements that ceased to exist entirely after disasters, since such places vanish from subsequent counts. Ghost towns like Wittenoom, Cossack, Mt Wells, and Burrundie fall outside the analysis, as do temporarily present victims such as the pearl divers killed by Cyclone Mahina in 1899, Australia’s deadliest natural disaster. Before 1971, Indigenous Australians were excluded from census coverage altogether, meaning some hazard impacts are almost certainly underestimated.

The study’s policy implications are direct. As climate change intensifies cyclones and floods across the tropical north, and as rising maximum temperatures threaten to push heat into the mobility calculus for vulnerable groups such as the elderly, the region’s settlements face pressures that past adaptation may not fully anticipate. The researchers argue that disaster mitigation and climate adaptation planning under frameworks such as the Northern Australia Action Plan 2024 to 2029 should be tailored to settlement type and size rather than applied uniformly, since a strategy suited to a diversified regional centre may be useless for an Indigenous outstation or a fly-in-fly-out mining hub. In demonstrating what 150 years of census data can and cannot reveal about disaster impacts at the demographic edge, the study offers both a methodological template and a warning: at the edges of habitation, where populations are small and connections tenuous, even a modest storm can bend the demographic arc of an entire town.