Whether for hurricanes, heatwaves, wildfires or floods, preparing for disasters and issuing warnings and evacuation orders are essential to save lives. However, not everyone is able to respond to the same degree. Illness, disability, lack of transport and financial constraints can all exacerbate the risk for people in vulnerable groups. This can lead to disproportionate consequences. For example, between 1930 and 2015, people aged over 65 accounted for 46% of excess deaths associated with hurricanes in the United States (R. Young and S. Hsiang Nature 635, 121–128; 2024).
To help the people most at risk, local authorities and first responders need to know in advance who lives in the areas most affected, so they can deploy targeted resources. But they cannot easily access this information.
How AI is improving climate forecasts
To demonstrate what could be possible, my colleagues and I at the Environmental Inequality Lab have developed a framework for rapid-response analysis. Using the best available data, this method can deliver timely insights to local authorities and first responders before, during and after disasters.
Our approach combines demographic and socio-economic data provided by the US Census Bureau with forecasts released by the US National Hurricane Center and Weather Prediction Center. The population data are privacy-protected and consist of aggregated counts by age, race, sex and income decile, mapped in grid cells of approximately one square kilometre. This provides localized insights while ensuring that no individual or household can be identified.
Bringing these data sets together tells us in real time where and when a disaster is expected to strike, as well as who is likely to be most exposed to risks such as high winds, storm surges and flooding. We are developing similar tools to evaluate the risks of wildfires and heatwaves.
We tested our work flow during Hurricane Milton, which made landfall on Florida’s west coast late on 9 October 2024. On 8 and 9 October, we released a report (see go.nature.com/4jwkuro) showing who was forecast to be exposed to the greatest hazards. The next week, we released a follow-up report (see go.nature.com/3hicifv) comparing the forecasts with data on the storm’s actual extent. In this case, the forecasts turned out to be a reasonable guide to what happened.
Hurricane risk in a changing climate — the role of uncertainty
Our analysis highlighted how different populations faced different levels of risk. The people who experienced hurricane-force winds and excessive rainfall during Milton were disproportionately older, disproportionately Hispanic and disproportionately relatively poor (falling in the bottom 20% of the national income distribution).
