Why the Deadliest Floods Strike Where Vulnerability Runs Deepest

Flood science has become astonishingly good at predicting where water will go. Two-dimensional hydraulic models now simulate flood routing and inundation patterns at resolutions fine enough to map individual streets and buildings, and machine-learning approaches are accelerating those computations to near-real-time speeds. Yet a troubling paradox persists: physically comparable floods keep producing radically different human outcomes depending on where they strike. The same depth of water that is an inconvenience in one neighbourhood becomes a catastrophe in another. Writing in Nature Water, Aiperi Stambekova and Avidesh Seenath of the Environmental Change Institute at the University of Oxford argue that flood risk governance has quietly inverted its priorities, obsessing over the physics of hazard while treating the social conditions that determine who dies, who loses everything, and who recovers as an afterthought. Their central claim is stark: anticipating where flood impacts will concentrate requires looking beyond where floodwater will flow, and instead identifying who lacks the capacity to avoid, withstand, and recover from flooding in the first place.

The case for this reframing is written in recent disaster statistics. When severe floods hit Indonesia in late 2025, the death toll passed 700 and roughly one million people were evacuated, according to reporting cited by the authors. Disasters on this scale are rarely products of water alone. They emerge from the interaction between a hazard and a population whose housing is fragile, whose warning systems fail, whose savings are thin, and whose political voice is muted. Risk analysts have formalised this insight for decades: risk is conventionally understood as the product of hazard, exposure, and vulnerability. Yet investment and policy attention remain overwhelmingly concentrated on the first of those three terms, because hazard modelling is tractable, fundable, and spectacularly visualisable, while vulnerability is messy, contested, and slow to measure.

The technical capabilities now exist to change this. Recent work in the flood modelling literature, including studies by Seenath and colleagues published in Risk Analysis in 2025, has pushed toward integrating social dimensions into flood risk assessment rather than treating hydraulic output as the end product. Research published in Nature Communications in 2024 by Fox and collaborators demonstrated that large-scale flood modelling can now capture inundation dynamics with enough fidelity to support decision-making at national and global scales. A 2025 study in npj Natural Hazards extended similar approaches to vulnerability-aware impact estimation. The computational machinery for vulnerability-informed flood governance is, in other words, no longer speculative. What remains missing, Stambekova and Seenath contend, is the governance architecture, and the political will, to place social vulnerability at the centre of flood risk management rather than at its margins.

The evidence base for why this matters is now substantial. A widely cited 2022 analysis by Rentschler, Salhab, and Jafino in Nature Communications estimated that roughly 1.8 billion people worldwide are exposed to substantial flood risk, and that a disproportionate share of those exposed live in low- and middle-income countries, with poorer households within those countries frequently occupying the most dangerous ground. Work on climate-related mobility, such as Sonja Ayeb-Karlsson’s research in Climate Risk Management, has documented how environmental stress interacts with poverty and immobility, trapping the least resilient in harm’s way. Studies by Rhein and Kreibich in Natural Hazards and Earth System Sciences in 2025 examined how the drivers of flood damage change over time, underscoring that vulnerability is dynamic and must be tracked, not assumed. Each of these strands points in the same direction: the distribution of flood consequences is a social variable, not a hydraulic one.

Measuring vulnerability, however, is far harder than measuring water depth. Vulnerability operates at multiple scales simultaneously. At the macro scale, it reflects national wealth, institutional quality, and infrastructure investment. At the meso scale, it reflects the characteristics of specific communities, including housing stock, access to emergency services, and social cohesion. At the micro scale, it reflects the circumstances of individual households: age, disability, income, tenure security, insurance coverage, and social networks. Stambekova and Seenath sketch a conceptual structure for micro-scale vulnerability profiles, arguing that flood governance needs to know not merely which neighbourhoods flood, but which households within those neighbourhoods will be unable to evacuate, will lose their livelihoods, and will still be displaced months after the water recedes. Bradford and colleagues, writing in Natural Hazards and Earth System Sciences, offered one of the earlier systematic frameworks for measuring flood vulnerability across such dimensions, and subsequent work by Moreira, de Brito, and Kobiyama reviewed indicator-based approaches to vulnerability assessment, revealing both progress and persistent fragmentation in how the concept is operationalised.

The fragmentation is itself a governance problem. Different agencies hold different pieces of the vulnerability puzzle: census bureaus hold demographic data, social services hold welfare data, housing authorities hold tenure data, and meteorological and hydrological services hold hazard data. Rarely are these streams integrated into a single, actionable picture of who is most at risk before a flood arrives. Recent methodological work by Yousaf, Seenath, and Speight in the Journal of Flood Risk Management has explored how uncertainty in flood risk assessment can be handled more rigorously, and the same rigour, the authors argue, needs to be applied to the social layer of risk. A warning issued 48 hours before a flood is only as effective as the capacity of recipients to act on it, and that capacity varies enormously across a population in ways that current warning systems seldom account for.

The economic argument for vulnerability-centred governance is compelling as well. Hallegatte and colleagues, writing in Economics of Disasters and Climate Change, showed that losses from natural disasters fall disproportionately on poor people, and that well-designed resilience policies can substantially reduce this unequal burden. Investment that reduces vulnerability, through safer housing, social protection floors, accessible early warning, and inclusive planning, delivers returns that pure hazard-defence investment cannot, because it protects people rather than merely repelling water. Jonkman, Curran, and Bouwer, analysing global flood fatality patterns in Natural Hazards in 2024, found that mortality from flooding remains heavily concentrated in lower-income contexts, a pattern that hardened infrastructure alone has not reversed. Meanwhile, work by Vestby and colleagues in the Proceedings of the National Academy of Sciences linked flood exposure to broader social instability, illustrating how unmanaged flood risk radiates outward into conflict, displacement, and political strain.

Susan Cutter, one of the founding figures of vulnerability science, argued in the International Journal of Disaster Risk Reduction in 2024 that the field has matured to the point where vulnerability assessment should be embedded routinely in disaster risk governance rather than confined to academic analysis. Stambekova and Seenath’s commentary extends that argument into the specific domain of flood governance, calling for a reorientation in which vulnerability profiling becomes a first-order input to decisions about levee placement, warning system design, evacuation planning, and post-flood recovery financing. The figure accompanying their article depicts the conceptual structure of micro-scale vulnerability profiles, a template for how household-level social data could be layered onto hydraulic hazard maps to produce risk pictures that predict human outcomes rather than merely water depths.

None of this diminishes the achievements of flood modelling, which have transformed what authorities can know about an impending event. The point, the authors insist, is that knowing where water will go answers only half the question. The other half, who will be there when it arrives, and whether they can survive and recover, demands data, institutions, and political commitment that flood governance has historically delegated to chance. As climate change intensifies hydrological extremes and urbanisation pushes more people onto floodplains, the gap between the physically modelled hazard and the socially distributed consequence will widen. Closing that gap, Stambekova and Seenath argue, is no longer a research aspiration but a governance imperative: flood risk management must put the most vulnerable at its centre, or it will continue to be measured, grimly, by the losses it fails to prevent.

Subject of Research: The integration of social vulnerability assessment into flood risk governance and management

Article Title: Flood risk governance must put social vulnerability at the centre

Article References: Stambekova, A., & Seenath, A. (2026). Flood risk governance must put social vulnerability at the centre. Nature Water. https://doi.org/10.1038/s44221-026-00713-x

Image Credits: AI Generated

DOI: 10.1038/s44221-026-00713-x

Keywords: flood risk, social vulnerability, flood governance, flood modelling, disaster resilience, hazard exposure, early warning systems, climate adaptation, community resilience, natural hazards, risk assessment, Nature Water