Flood hazard potential reveals global floodplain settlement patterns

Global flood damages and affected populations have risen markedly, driven by growth in exposure and climate change. Conventional top-down climate impact assessments propagate scenarios from global climate models through hydrological and flood models but accumulate large uncertainties at each stage, leading to divergent flood hazard and exposure estimates. As an alternative, the authors adopt a scenario-neutral stress-testing approach to identify where floodplains and populations are most sensitive to changes in flood event magnitude. The research questions are: how does flooded extent and population exposure grow with increasing flood magnitude across the global river network, what physical factors control this sensitivity, and how do human settlement patterns align with the sensitivity of flood hazards? The study aims to provide a first-order global guide to where changes in hazard would most critically affect exposure and to inform adaptation strategies.

Prior global flood risk studies commonly follow a model cascade from GCM projections to hydrological simulations and flood hazard mapping to derive exposure and impact estimates, but these approaches suffer from biases in extreme precipitation, model structural choices, and parameter uncertainties, resulting in large discrepancies among global flood models. Scenario-neutral or stress-testing methods have been proposed and applied to river flow sensitivity but rarely to flood hazard and exposure sensitivity at global scale. A recent global sensitivity study of exposure assumed linear growth between event magnitudes and excluded many smaller basins, potentially misrepresenting settlement dynamics. Socio-hydrological literature documents human–flood co-evolution, including adaptation and the levee effect, and regional studies highlight diverse settlement responses to flood hazards. This work builds on these strands by mapping flood hazard and population exposure sensitivities globally and relating them to geomorphology and regional settlement patterns.

The study uses fluvial flood hazard maps from the Fathom global flood model at ~90 m resolution for ten return periods (5, 10, 20, 50, 75, 100, 200, 250, 500, 1000 years). The hydraulic framework routes extreme river flows estimated via regional flood frequency analysis of GRDC gauge data through 1D subgrid river channels (MERIT Hydro) coupled to a 2D hydrodynamic model (LISFLOOD-FP, local inertial formulation) over MERIT DEM topography. Undefended fluvial inundation is simulated for all rivers with upstream drainage area >50 km2, globally between ~60°N and 56°S. Flood depth grids are thresholded (>0 m) to binary extents. The river network is segmented into reaches using upstream–downstream inflow points, yielding 1,380,430 inflow sites (~1.2 million retained after quality control). For each reach and return period, flooded area is computed within the reach footprint and then normalized to form a growth curve. A power law F = RP^b is fit to normalized flooded area (F) versus log-normalized return period (RP). The exponent b (reported as bf or br) quantifies flood hazard sensitivity: b < 1 indicates rapid growth at frequent, low-magnitude events; b ≈ 1 linear growth; b > 1 rapid growth at rare, extreme events. Fits with R2 < 0.9 are excluded (12% of sites removed). To classify floodplain types controlling sensitivity, relationships between channel slope and upstream drainage area are analyzed. Thresholds on channel slope delineate three categories: confined (steep, bedrock-dominated valleys), partially confined (transitional valleys with lateral constraints), and laterally unconfined (broad, flat, unbound floodplains such as deltas). Population exposure is calculated by intersecting binary flood extents with 2020 WorldPop constrained population counts (~90 m, country-level GeoTIFFs harmonized to hazard tiles). For each reach and return period, exposed population totals are extracted, normalized, and a power law E = RP^bpop is fitted. The bpop exponent indicates exposure sensitivity across flood zones: bpop < 1 implies higher fractions living in frequently flooded zones; bpop ≈ 1 implies even distribution; bpop > 1 implies higher fractions in rarely flooded zones. Country-level summaries compute mean sensitivity parameters by floodplain type. Additional analyses assess regional distributions, exposure shares by floodplain category, and exposure density per kilometer of reach.

• Physical controls: Flood hazard sensitivity exhibits clear spatial patterns primarily organized by local topography (channel slope) and upstream drainage area. Confined floodplains (8% of reaches) and laterally unconfined floodplains (3%) tend to be most sensitive to rare, extreme events (b > 1), while partially confined floodplains (89%) tend to be most sensitive to frequent, low-magnitude events (b < 1).
• Global exposure magnitude: Approximately 2 billion people live on floodplains (defined as the 1000-year extent) globally; about 1.4 billion live on the 100-year floodplain. Asia has about 1.5 billion people on floodplains (35% of the continent’s total population), accounting for ~75% of global floodplain population. China and India have about 490 million and 456 million people on floodplains, respectively. Bangladesh has ~105 million people on floodplains.
• Exposure by floodplain type: In Asia, laterally unconfined floodplains comprise only ~5% of reaches but account for ~20% of population exposure. Globally, laterally unconfined floodplains host ~412 million people (21% of total exposure) despite representing ~5% of populated reaches, and populations here are typically most densely settled in rarely flooded zones (bpop > 1).
• Settlement patterns vs hazard sensitivity: For partially confined floodplains, almost all countries plot above the 1:1 line relating hazard and exposure sensitivities, indicating higher population density in rarely flooded zones, yet many countries (notably in Europe and Asia) have bpop < 1, meaning more people in absolute numbers live in frequently flooded zones due to their larger area. This suggests adaptation or acceptance of frequent-flood risk (e.g., via defenses or livelihood choices). In South and Central America, several countries (e.g., Brazil, Paraguay, Costa Rica, Suriname, Argentina, El Salvador, Haiti, Puerto Rico) exhibit bpop > 1 with higher absolute exposure in rarely flooded zones, indicating settlement away from rivers to mitigate frequent flooding.
• Regional protection and pressures: In Europe, many countries show high standards of protection (≥50-year in 58%; ≥100-year in 26%), supporting more even density across zones but larger absolute exposure in frequently flooded areas, consistent with the levee effect. In Asia, very large exposed populations occupy frequently flooded zones on partially confined floodplains (~1.1 billion), likely due to urban growth and agricultural dependence, while defense data are sparse and possibly underestimated.
• Exposure density: Laterally unconfined floodplains in South, East, and Southeast Asia and North Africa have high exposure densities per kilometer of reach (~4519, 2536, 1973, and 902 people/km, respectively), compared to North America and Europe (~183 and 247 people/km). In China, 82% of laterally unconfined reaches have the highest exposure density in rarely flooded zones, with ~69 million people in ≥100-year zones. Across China, India, Bangladesh, and Vietnam, over 160 million people live in rarely flooded zones of laterally unconfined floodplains.
• Policy-relevant sensitivity: Populations on partially confined floodplains are particularly susceptible to increases in the intensity of frequent events (e.g., deeper floods during overtopping). Populations on laterally unconfined and confined floodplains are more at risk from increasing probabilities of rare, extreme events affecting areas without recent flood memory.

The analysis demonstrates that the sensitivity of flood extents is strongly governed by floodplain geomorphology, which in turn shapes human settlement behavior. Where floodplains are sensitive to frequent, low-magnitude events (partially confined), populations tend to distribute more evenly across zones, often resulting in greater absolute numbers living in frequently flooded areas, especially in regions with defenses or livelihood necessities. Where floodplains are sensitive to extreme events (laterally unconfined and confined), populations tend to concentrate most densely in rarely flooded zones, implying heightened vulnerability if the probability or magnitude of extremes increases. These results help resolve the research questions by linking physical sensitivity (bf) to exposure sensitivity (bpop) at scale, revealing systematic regional patterns: Europe and North America show evidence of protection and potential levee effects; Asia shows very high exposure with land and food system pressures; South America shows relatively hazard-averse settlement with higher density in less frequently flooded zones and potential adaptation effects. The interaction between changing hazard and settlement patterns highlights where adaptation should focus: raising protection and zoning on partially confined floodplains to manage frequent events, and enhancing preparedness for rare extremes on laterally unconfined and confined floodplains, particularly in rapidly urbanizing deltas. The framework offers actionable, first-order guidance for targeting interventions under climate non-stationarity and growing exposure.

This study introduces a global, high-resolution stress-testing framework to quantify sensitivity of flood hazard extents (bf) and population exposure (bpop) across ~1.2 million river reaches. It shows that floodplain geomorphology (channel slope and drainage area) organizes hazard sensitivity into confined, partially confined, and laterally unconfined types, and that human settlement patterns align with these sensitivities in systematic, regionally distinct ways. Key contributions include global estimates of floodplain populations (~2 billion on 1000-year floodplains; ~1.4 billion on 100-year floodplains), identification of high-risk settlement patterns on laterally unconfined floodplains (412 million exposed globally), and policy-relevant insights into where increases in frequent versus extreme events would most affect exposure. Future research should: improve representation of flood defenses and adaptation measures in global datasets; integrate coastal, storm surge, and compound flooding; address discharge and extreme precipitation uncertainties; extend sensitivity analyses to infrastructure and economic assets; and refine methods and data for regions with sparse observations, especially arid areas and parts of Asia.

• Model and data uncertainty: Flood hazard maps derive from a global model dependent on DEM quality, river morphology estimation, hydrodynamic assumptions, and regional flood frequency analysis of gauged flows. Gauged data are sparse and most uncertain during extreme events; arid regions show greatest uncertainty and lowest model accuracy.
• Assumption of invariant growth shape: The analysis assumes the shape of discharge growth with return period and the flooded area/exposure growth curves remain applicable under future conditions, despite potential non-stationarity and differing trends for small versus large floods.
• Defenses and adaptation: Undefended flood maps likely overestimate exposure where protection is extensive. Available defense data (e.g., FLOPROS) are coarse, sparse in many regions (notably Asia), and not differentiated by floodplain type or spatial planning policies.
• Hazard scope: Only fluvial flooding is included; coastal, storm surge, pluvial, and compound events are excluded, which can be critical for deltas and coastal reaches.
• Network coverage and quality control: Only rivers with upstream drainage area >50 km2 are modeled; 12% of reaches are excluded due to poor power-law fit (R2 < 0.9). Although the retained reaches cover 89% of the global population on floodplains, some complex behaviors may be omitted.