Flood Models

The Most Widespread Natural Hazard

Flood is the most widespread natural hazard on the planet. It affects more countries, more people, and more property than any other peril. Yet despite its reach, flood has historically been one of the most underinsured perils globally and one of the most challenging to model accurately. The protection gap for flood — the difference between economic losses and insured losses — is among the largest of any natural peril, regularly exceeding 70–80% in many regions. Understanding why this gap exists, and how flood cat models attempt to bridge it, is essential for any cat modelling professional.

Why Flood Is Uniquely Challenging to Model

Compared to earthquake and hurricane, flood presents a distinct set of modelling challenges that have historically made it harder to address:

• Multiple flood mechanisms: Unlike a single physical process, "flood" encompasses several fundamentally different hazard types — each requiring a different modelling approach
• High spatial variability: Flood inundation depth can vary dramatically over very short distances — a property on the ground floor versus one on the first floor, or a property 50 metres from a river versus 500 metres, can face very different risk levels
• Dependence on defence infrastructure: Flood losses depend critically on the condition and capacity of flood defences — levees, embankments, sea walls — which are themselves subject to failure
• Data scarcity: Detailed topographic data (digital elevation models), river flow records, and historical flood footprints are less available in many regions than seismic or meteorological data
• Insurance market structure: In many markets, flood is excluded from standard property policies, making insured loss data sparse and harder to use for model validation

The Four Flood Mechanisms

Professional flood models address multiple distinct flood types. Each has different physical drivers, different geographic distributions, and different implications for the insurance industry.

1. Fluvial Flooding (River Flood)

Fluvial flooding occurs when a river's flow exceeds the capacity of its channel and water spills onto the surrounding floodplain. It is driven by prolonged or intense rainfall over a river's catchment area, often combined with factors such as snowmelt, saturated soils, or antecedent wet conditions. Fluvial floods are typically slow to develop — giving some warning time — but can inundate large areas and persist for days or weeks. The Rhine floods in Europe, the Mississippi floods in the United States, and the Indus floods in Pakistan are examples of large-scale fluvial events.

Key parameters for fluvial flood modelling include river discharge (flow rate in cubic metres per second), channel geometry, floodplain topography, and the presence and condition of flood defences. Hydrological models simulate rainfall-runoff processes across the catchment to estimate peak river flows, which are then input to hydraulic models that simulate how water spreads across the floodplain.

2. Pluvial Flooding (Surface Water Flood)

Pluvial flooding — also called surface water flooding — occurs when intense rainfall overwhelms urban drainage systems and water accumulates on the surface before it can drain away or infiltrate into the ground. It is not directly related to proximity to rivers or coasts: it can occur anywhere that rainfall intensity exceeds drainage capacity. Pluvial flooding is particularly significant in urban areas with extensive impervious surfaces (roads, car parks, buildings) that prevent infiltration.

3. Coastal Flooding (Storm Surge)

Coastal flooding arises from storm surge — the abnormal rise in sea level driven by the low atmospheric pressure and strong onshore winds of a tropical or extra-tropical storm. Storm surge can raise sea levels by several metres above normal tide levels, inundating low-lying coastal areas. It is often the most destructive component of a landfalling hurricane — Katrina's surge reached over 8 metres in parts of Louisiana — and is also significant in extra-tropical storms affecting coastlines in northern Europe and elsewhere.

Coastal flood modelling requires simulation of storm surge heights along the coastline (combining the meteorological event with ocean response models), combined with inundation modelling across the coastal floodplain. Tidal conditions at the time of landfall can significantly affect total water levels — a storm surge coinciding with a high spring tide produces far greater flooding than the same surge at low tide.

4. Groundwater Flooding

Groundwater flooding occurs when the water table rises above the surface, typically following prolonged periods of heavy rainfall that saturate the ground. It is most common in areas underlain by permeable geology such as chalk or limestone. Groundwater flooding is slow to develop — sometimes taking weeks or months of wet weather to trigger — but can also be slow to recede, with flooding persisting long after rainfall has stopped. It is the least well-modelled flood type and often excluded from commercial cat models, though its significance in regions such as southern England is well-documented.

The Flood Modelling Framework

A complete flood cat model integrates three main components, each corresponding to a stage in the physical process:

Hydrological Modelling

The hydrological component simulates how rainfall becomes river flow — the rainfall-runoff process. Key inputs include precipitation data (historical and stochastic), catchment characteristics (soil type, land cover, topography), and antecedent soil moisture conditions. Models simulate how rainfall is partitioned between infiltration, surface runoff, and subsurface flow, ultimately producing estimates of river discharge at key points in the river network. For coastal flooding, meteorological and ocean models simulate storm surge heights along the coastline.

Stochastic event generation for flood differs significantly from earthquake or hurricane. Rather than generating individual point-source events (a magnitude and location for earthquakes, a track and intensity for hurricanes), flood models generate stochastic rainfall fields — spatially correlated patterns of rainfall over a catchment — that drive river flows. Multi-day rainfall accumulations are often more relevant than peak hourly intensities for fluvial flooding.

Hydraulic Modelling

The hydraulic component simulates how water moves across the floodplain, given the peak river flows or storm surge heights from the hydrological step. Hydraulic models solve the equations of fluid dynamics to estimate water depth and velocity at each point across the modelled area. They account for the geometry of the river channel, the topography of the surrounding floodplain, and the presence of flood control structures.

Flood Defence Representation

One of the most important and difficult aspects of flood modelling is representing flood defence infrastructure — levees, embankments, sea walls, flood gates, and urban drainage systems. Defences fundamentally alter the spatial pattern of flood risk: a property behind a high-standard levee may have near-zero flood risk in most model simulations, while a very similar property on the other side of the levee faces frequent, severe flooding.

The challenge is twofold. First, comprehensive, accurate databases of defence locations, crest levels, and condition are not always available. Second, defences can fail — either by overtopping (when flood levels exceed the defence crest) or by breaching (structural failure before overtopping). Defence failure can dramatically change the loss landscape: a levee breach on the Mississippi in 1993 or the levee failures in New Orleans during Katrina (2005) turned a large flood into a catastrophe of a different order. Cat models must represent both the protection provided by defences under normal conditions and the probability and consequences of defence failure.

Flood Vulnerability and Damage Functions

Flood vulnerability functions relate inundation depth — the primary intensity measure for flood — to expected damage ratios for different building and content types. Unlike earthquake (where the primary driver is structural behaviour under dynamic loading) or wind (where envelope failure is the dominant mechanism), flood damage is primarily driven by water depth inside the building and the duration of inundation.

Key factors in flood vulnerability include:

• Inundation depth: The depth of water inside the building relative to the ground floor level. This is the single most important damage driver — shallow flooding may cause only minor damage, while deep flooding causes severe structural and contents damage
• Flow velocity: High-velocity flooding (as in flash floods or near levee breaches) can cause structural damage and scour foundations, in addition to inundation damage
• Flood duration: Prolonged inundation leads to greater saturation of walls and floors, greater mould damage, and more extensive contents loss
• Flood water quality: Contamination by sewage or chemicals significantly increases damage and remediation costs
• Building type and construction: Masonry buildings absorb water into walls, leading to prolonged drying times and structural degradation. Timber-framed buildings are particularly vulnerable to prolonged inundation. Flood-resilient construction features — raised floor levels, flood barriers, water-resistant materials — can significantly reduce damage
• Finished floor level: The height of the ground floor above external ground level determines at what water depth flooding enters the building. A property with a raised threshold may remain dry even when streets around it are flooded

The Protection Gap and Why It Persists

Flood has a larger protection gap than almost any other natural peril. Several structural factors explain this:

• Exclusion from standard policies: In many markets (including the United States), flood is excluded from standard homeowners policies. Separate flood insurance — often through government schemes such as the U.S. National Flood Insurance Program (NFIP) — is required, but take-up rates outside mandatory zones are low
• Adverse selection: Unlike earthquake, flood risk is highly spatially concentrated and relatively well-known to property owners. Those in high-risk areas are far more likely to purchase cover, making the insured pool highly adverse and premiums high — which further suppresses take-up among lower-risk properties
• Government provision: In many countries, governments have historically provided post-disaster relief for flood victims, reducing the perceived need for private insurance
• Historical losses exceeding expectations: Several major flood events have produced losses far exceeding what insurers had modelled, leading some to withdraw from flood markets or apply very high loadings

Climate Change and Flood Risk

Flood is the natural peril most directly and demonstrably affected by climate change, through two primary pathways:

• Intensification of extreme rainfall: A warmer atmosphere holds more water vapour, leading to more intense rainfall events. The relationship is roughly described by the Clausius-Clapeyron equation — approximately 7% more water vapour per degree of warming — which translates to more intense short-duration rainfall and increases in pluvial and flash flood frequency
• Sea level rise: Rising sea levels increase the baseline from which storm surges operate, effectively increasing the frequency and severity of coastal flood events at any given location. A 50cm rise in sea level could convert a 1-in-100 year coastal flood event into an annual occurrence in many low-lying coastal areas

For cat modellers, climate change introduces a non-stationarity problem: the historical record — already short for calibrating flood frequency relationships — may no longer be representative of future flood risk. This is driving the development of climate-conditioned flood models that explicitly incorporate projected changes in rainfall intensity and sea level into the hazard component.

Flood vs. Earthquake and Hurricane — Key Differences for Modellers

Having now studied all three major perils, it is worth summarising the key differences that affect how they are modelled and how model outputs are used: