In July 2021, a slow-moving low-pressure system stalled over western Germany and Belgium, dropping 150mm of rain in 24 hours onto already saturated ground. The Ahr River — a modest tributary of the Rhine — rose 8 metres in hours, sweeping away bridges, roads, and entire villages. 220 people died. Economic losses exceeded EUR 40 billion. Insured losses were approximately EUR 13 billion — just 33% of the economic total, and even that figure was high by historical flood standards for Germany.
Weeks earlier, similar rainfall flooding had killed over 300 people in Zhengzhou, China — a city of 12 million in Henan Province — when subway tunnels flooded during a single-day extreme rainfall event. Economic losses were in the tens of billions of dollars. Insured losses were a fraction of a percent of economic loss. The protection gap there was essentially total.
Flood is the world's costliest natural peril by economic loss and the one with the largest gap between economic damage and insured loss. It is also the most technically demanding peril to model well — three distinct flood types with separate physical mechanisms, a hazard that varies at sub-street-block scale, a vulnerability function that is sensitive to inundation depth in centimetres, and a coverage landscape so fragmented across markets that the same event can produce very different insured loss ratios depending entirely on which policies are in force.
Flood encompasses three physically distinct hazard types that share a common damage mechanism — water inundating structures — but differ fundamentally in their formation, spatial scale, warning time, and the buildings they affect. A cat model that handles fluvial flood well may handle pluvial flood poorly. A flood policy that covers river flooding may exclude surface water. Understanding which flood type is driving a loss is a prerequisite for assessing whether the model and the policy are responding to the same event.
River or watercourse overflows its banks following sustained or extreme rainfall over the catchment upstream. Build-up time ranges from hours (flash flood on steep catchments) to days (major river systems). Spatial extent follows the floodplain — typically well-mapped. Warning times are longer for major rivers, shorter for small steep catchments. The primary flood type in most major flood loss events historically.
Rainfall exceeds drainage capacity before it reaches a watercourse — water accumulates on the surface in low-lying areas, underpasses, and urban depressions. Does not require a river to overflow. Can occur in locations that conventional floodplain maps show as low-risk. The primary mechanism in urban flash flood events. Warning time is very short — typically the duration of the rainfall event itself.
Sea water inundates low-lying coastal land driven by storm surge (covered in P.2), tidal extremes, or a combination of surge and tide. Distinct from tsunami-driven inundation (covered in P.7). Affects coastal floodplains, estuaries, and deltaic coastlines. Can penetrate far inland along low-gradient coastlines and river mouths. Inundation depth and duration governed by surge height and coastal topography.
Unlike wind (where peak gust speed drives damage) or earthquake (where PGA drives damage), the primary intensity metric for flood is inundation depth — the depth of water above the floor of an affected building. Depth governs how far water rises into the building, which determines what is damaged: contents on the floor, wall finishes and services in the lower metre, structural elements only at extreme depths. The relationship between depth and damage is not linear — the first 30cm of inundation in a building causes disproportionate damage to contents and ground-floor services relative to the incremental damage from additional depth above that level.
Velocity is a secondary intensity metric — fast-moving floodwater exerts hydrostatic and hydrodynamic forces on structures that slow-moving water does not. Flash floods on steep catchments and coastal surge events produce high-velocity flows capable of structural damage, partial collapse, and debris impact that depth alone does not capture. Most flood models treat velocity as a secondary modifier rather than a primary variable, which can understate loss in fast-moving flood scenarios.
Duration matters for contents and business interruption losses. A building inundated to 50cm for two hours sustains different losses than the same building inundated to 50cm for two weeks. Prolonged inundation enables mould growth, deeper penetration of water into building materials, and extended loss of access that drives business interruption claims well beyond the direct physical damage repair cost. Duration is rarely represented as a primary variable in standard flood vulnerability functions.
Flood is the most computationally demanding of the major natural perils to model at high resolution. The hazard calculation requires simulating water movement through a three-dimensional landscape — routing rainfall through river catchments, calculating floodplain inundation extents and depths, and in urban areas, modelling how water moves through drainage networks, streets, and built environments at sub-100m resolution. This is fundamentally different from the relatively smooth spatial interpolation that characterises wind and earthquake hazard models.
Fluvial flood hazard models work by simulating river discharge through hydraulic routing models that calculate inundation extent and depth across the floodplain. The input chain is: rainfall over the catchment → hydrological model to convert rainfall to river flow → hydraulic model to route flow through the river channel and floodplain → inundation map showing depth and extent. Each stage introduces uncertainty that accumulates through the chain.
The stochastic event set for fluvial flood is built by sampling from the statistical distribution of extreme rainfall and river flow events at each catchment. Return period flow estimates — the 1-in-100 year flow, 1-in-200 year flow — are derived from gauged flow records at river monitoring stations. Where gauge records are short (under 30–50 years), extrapolation to the 1-in-200 year return period carries substantial uncertainty. The inundation maps for each return period event become the hazard footprints against which exposure is assessed.
Flood defence representation is a critical and contested component of fluvial flood models. Levees, floodwalls, retention ponds, and embankments physically alter the inundation extent and depth for events below the defence standard — but create discontinuous failure risk for events that exceed the defence design level. A levee designed to the 1-in-100 year standard provides complete protection below that threshold and potentially catastrophic exposure above it. Modelling this step-change in risk around defence thresholds is technically difficult and a major source of uncertainty in urban fluvial flood models.
Pluvial flood models require high-resolution digital terrain models of the urban surface to identify where surface water accumulates — depressions, underpasses, low points in street networks, basement-level entrances. The hazard calculation routes a design rainfall event (a rainfall depth over a specified duration and area) through the surface drainage network and identifies where drainage capacity is exceeded and water ponds. Because the relevant spatial variation is at the level of individual streets and building footprints, pluvial models require resolution of 1–5m — orders of magnitude finer than the 100–1000m grids typical of regional fluvial models.
The computational cost of high-resolution pluvial modelling means that full urban pluvial models are available for fewer cities than fluvial models. For locations without high-resolution pluvial modelling, cat models often apply a simplified surface water susceptibility index — a coarser proxy based on terrain and drainage data — which produces less reliable results than a full dynamic model.
Consider two cities: City A has no flood defences. City B has levees designed to the 1-in-100 year standard. At the 1-in-50 year return period, City B has zero flood loss (full protection) while City A has moderate loss. At the 1-in-150 year return period — just above the defence design level — City B may experience catastrophic loss as the levee overtops or fails, potentially exceeding City A's loss from the same return period event. The EP curve for City B has a near-zero loss for many return periods and then a sudden step-change to very large loss around the defence design standard. Standard flood models that do not explicitly model this discontinuity will produce smooth EP curves that understate the tail risk for well-defended cities and overstate the risk for return periods below the defence standard.
Flood vulnerability is depth-dependent in a way that no other major peril is. The damage a building sustains is directly governed by how high the water rises, which determines which building components are inundated. This means that a 10cm difference in inundation depth — easily within the uncertainty of the hazard model — can translate into a meaningful difference in damage ratio, because 10cm can be the difference between water staying below the floor level of a raised building and entering the ground floor.
The primary vulnerability tool for flood is the depth-damage function — a curve relating inundation depth above floor level to mean damage ratio for a given building type. The shape of the curve is broadly similar across building types: a steep rise in damage ratio across the first 0.5–1.0m of inundation (where the most damageable contents and ground-floor finishes are located), followed by a flatter increase at greater depths as incremental damage to already-inundated building fabric adds less marginal loss.
DR = damage ratio as percentage of ground-floor TIV. Illustrative only — actual curves vary by construction type, occupancy, and country-specific calibration datasets. Upper floors undamaged unless inundation exceeds storey height.
Flood is the natural peril with the largest global protection gap — the difference between economic loss and insured loss. Across most markets, the majority of flood economic loss is uninsured. The reasons are structural: flood risk is highly concentrated geographically (properties in floodplains bear most of the risk), which creates extreme adverse selection that makes private flood insurance economically unviable without either compulsion or government involvement. Most functioning flood insurance markets rely on some form of government backstop, mandated coverage, or cross-subsidy mechanism.
Figures are indicative and vary by flood type, property class, and year. Commercial property penetration is typically higher than residential across all markets.
France — Catastrophes Naturelles (Cat Nat): A government-mandated scheme under which flood coverage is automatically included in all standard property insurance policies. A natural disaster must be officially declared by inter-ministerial decree for claims to be paid under Cat Nat. The scheme is funded by a surcharge on all property premiums. Commercial property is covered alongside residential. This produces France's relatively high penetration — but the government declaration requirement creates uncertainty about which events qualify.
UK — Flood Re: A reinsurance pool established in 2016, jointly funded by the government and the insurance industry. Insurers can cede flood risk on high-risk residential properties to Flood Re at subsidised premiums, making flood coverage commercially viable for properties that private insurers would otherwise exclude or price prohibitively. Flood Re applies to residential property only — commercial flood remains in the private market and is often excluded from standard commercial property policies or subject to significant sub-limits and deductibles.
United States — NFIP: The National Flood Insurance Program is a federal programme that provides flood insurance primarily for residential and small commercial properties in communities that adopt FEMA flood management regulations. Coverage is limited to USD 250,000 for residential buildings and USD 500,000 for commercial. Properties outside the mapped Special Flood Hazard Area (SFHA) often do not purchase NFIP policies — creating the situation where a major pluvial flooding event, which by definition affects areas outside the mapped floodplain, generates large economic losses with very limited insurance coverage.
Germany — voluntary opt-in: Flood coverage (Elementarschadenversicherung — elemental damage insurance) is available as an optional endorsement to standard German property policies. Despite significant loss experience and growing risk awareness, only approximately 40% of German residential properties carry this endorsement. The 2021 Ahr valley floods demonstrated this gap starkly — many affected properties in the river valley had declined or never been offered flood coverage, leaving residents dependent on emergency government aid rather than insurance settlements.
| Region | Dominant Flood Type | Insurance Significance | Modelling Maturity |
|---|---|---|---|
| Central Europe (Rhine, Danube, Elbe basins) |
Fluvial — major river systems with large catchments. Snowmelt interaction with spring rainfall a key driver of extreme flows. Urban pluvial events increasingly significant. | High economic loss potential but mixed penetration. France (Cat Nat) and Austria have better penetration than Germany. Events like 2002 Elbe and 2021 Ahr floods illustrate the scale of economic loss relative to insured loss. | High for fluvial on major rivers. Pluvial modelling improving. Flood defence representation remains a key model uncertainty — particularly for the Rhine corridor. |
| United Kingdom | Fluvial and pluvial. UK has numerous small, fast-responding catchments prone to flash flooding. Coastal flood risk significant on east coast (North Sea surges). Urban surface water flooding increasingly prominent. | Significant — Flood Re has expanded residential penetration but commercial flood remains patchy. Lloyd's and London market key for commercial flood. Several major flood years (2007, 2013–14, 2019–20) calibrated the UK market's loss experience. | Very high. JBA, Ambiental, and vendor model flood offerings for the UK are among the most detailed globally. High-resolution pluvial models available for major urban areas. Environment Agency's national flood mapping is world-class. |
| United States | All three types — fluvial (Mississippi, Missouri, major rivers), pluvial (urban flash floods), coastal (Atlantic and Gulf surge covered separately). Each type has a distinct insurance structure. | Very high — but fragmented across NFIP, private flood market, and self-insurance. Harvey (2017) demonstrated the enormous pluvial flood gap — the majority of flooded properties in Houston were outside the mapped SFHA and had no flood insurance. | High for fluvial on major rivers. FEMA's National Flood Insurance Rate Maps (FIRMs) are the regulatory framework but are often outdated. Private flood market growth has driven investment in higher-resolution models. |
| Australia | Fluvial flood in Queensland and NSW river valleys. Coastal flood risk in cyclone-affected regions. Pluvial urban flooding increasingly significant in Brisbane and Sydney. | High penetration for standard home and commercial policies, but flood exclusions are common in older policy forms. 2010–11 Queensland floods and 2022 east coast floods drove major market and model development. | High and improving rapidly. Post-2011 investment in high-resolution flood models for key river catchments. ICA-commissioned flood mapping for many catchments now publicly available. |
| South and Southeast Asia (Bangladesh, India, Vietnam, Thailand) |
Fluvial monsoon flooding dominates — river systems fed by Himalayan snowmelt and monsoon rainfall flood extensively each year in Bangladesh, eastern India, and the Mekong delta. Flash flood and urban pluvial growing in significance as urbanisation increases. | Very low insurance penetration. Monsoon flooding is an expected seasonal event rather than an insurable catastrophe in much of the region — the concept of flood as an insurable peril requires catastrophic events above background flood levels. The 2011 Thailand floods (primarily fluvial, affecting industrial estates in the Chao Phraya basin) were an exception — significant insured losses from industrial property with Western-style coverage. | Low to moderate. Hazard mapping exists for major river systems but at lower resolution than European or North American models. Vulnerability functions for local construction poorly calibrated. |
| China | Fluvial flooding along major river systems (Yangtze, Yellow River, Pearl River) — some of the world's largest river systems with enormous flood volumes. Urban pluvial flooding increasingly significant as rapid urbanisation has increased impervious surface area in major cities. | Growing but still low penetration relative to exposure. Government-sponsored schemes are expanding. Zhengzhou 2021 and Henan Province events demonstrated the scale of uninsured urban flood loss even in a major developed city. | Improving. Major investment in Chinese flood modelling following high-profile events. Limited loss data availability constrains calibration. Government flood mapping data not always publicly available. |
The majority of standard commercial cat model flood offerings focus on fluvial flood — river overflow events that can be mapped to a floodplain. Pluvial flood, which affects urban areas outside the mapped floodplain, is either absent from these models or represented by a simplified susceptibility index rather than a dynamic hydraulic model. For portfolios with significant urban commercial exposure, this systematically underestimates the frequency and severity of surface water flooding events — which are increasing in frequency as urbanisation increases impervious surface area and climate change intensifies extreme rainfall events.
Standard flood models struggle to accurately represent the discontinuous protection offered by flood defences — full protection below the defence design standard and potentially catastrophic loss above it. Many models either ignore defences entirely (overstating risk for well-defended areas) or treat them as providing continuous partial protection rather than a binary threshold. The tail risk for well-defended cities — where a defence-exceeding event could cause enormous losses to properties that have experienced decades of zero flood loss — is one of the most significant sources of epistemic uncertainty in European and US fluvial flood modelling.
The threshold effect — the flood depth at which water enters a building — is among the most important variables governing whether a shallow inundation event causes any building damage at all. For a building with a raised ground floor 30cm above street level, a 20cm inundation event causes no building loss. For a slab-on-grade building with no threshold, the same event causes ground-floor damage. This information is rarely available in standard exposure data, and models typically apply a default threshold assumption across all properties. The resulting error is location-specific and directionally uncertain.
Standard flood vulnerability functions estimate physical building damage. Business interruption losses from flood — which can substantially exceed physical damage for businesses that cannot operate while flooded properties are repaired — are either excluded from model output or estimated through a simple uplift factor. For commercial portfolios with significant BI coverage, the physical damage model output may materially understate the total insured loss, particularly for prolonged inundation events or events that cut access to undamaged properties through road closures and infrastructure disruption.
Flood frequency is changing in most regions as climate shifts alter precipitation patterns, snowmelt timing, and storm intensity. Historical flow records used to calibrate stochastic flood catalogues may no longer be representative of the future hazard distribution. This is particularly acute for regions experiencing intensification of extreme rainfall events, where the 1-in-100 year flow estimated from a 50-year gauge record may systematically understate current flood frequency. Models calibrated to the historical record are increasingly being supplemented with climate-adjusted scenarios, but these carry their own uncertainty.
River or watercourse overflow following rainfall or snowmelt that exceeds the channel's capacity. Inundation follows the mapped floodplain. Build-up time ranges from hours (flash flood) to days (major rivers). The dominant flood type modelled in most commercial cat platforms.
Surface water flooding caused by rainfall exceeding drainage capacity before reaching a watercourse. Affects low-lying urban areas outside the mapped floodplain. Short warning time. The primary mechanism in urban flash flood events. Often absent from standard fluvial flood cat models.
The primary intensity metric for flood — the depth of water above the floor level of an affected building. Governs which building components are inundated and drives the depth-damage function. A 10cm difference in inundation depth can translate into meaningful differences in damage ratio.
The flood vulnerability function relating inundation depth above floor level to mean damage ratio for a given building type and occupancy. Steep rise in the first 0.5m (where most damageable contents and finishes are located), flattening at greater depths. Calibrated from post-event damage surveys.
The difference between economic loss and insured loss from a natural catastrophe. Largest globally for flood — typically 70–90% of flood economic losses are uninsured across most markets. Driven by adverse selection, affordability constraints, and the structure of government-backed flood schemes.
Catastrophes Naturelles — the French government-mandated natural disaster insurance scheme. Flood coverage is automatically included in all standard property insurance policies; claims are triggered by inter-ministerial declaration of a natural disaster. Produces France's relatively high flood insurance penetration.
A UK reinsurance pool established in 2016 allowing insurers to cede flood risk on high-risk residential properties at subsidised premiums. Makes flood coverage viable for properties otherwise uninsurable at affordable premiums. Applies to residential property only — commercial flood remains in the private market.
National Flood Insurance Program — the US federal flood insurance scheme providing coverage primarily for residential and small commercial properties in communities adopting FEMA flood management regulations. Coverage limited to USD 250K residential / USD 500K commercial. Properties outside mapped Special Flood Hazard Areas frequently uninsured.
FEMA's designation for areas with a 1% annual chance of flooding (the 1-in-100 year floodplain). The primary regulatory zone for NFIP mandatory purchase requirements. Properties outside the SFHA are not required to carry flood insurance — a significant gap for pluvial flood events that affect areas outside mapped floodplains.
The component of a fluvial flood model that converts rainfall over a catchment into river flow at specific points along the river system. Accounts for soil saturation, evapotranspiration, and routing of rainfall through the catchment. The first stage in the fluvial flood model calculation chain.