Executive summary
- Flood hazard is highly localised: metres of horizontal distance and centimetres of elevation change the result materially.
- Distinct flood types — fluvial, pluvial, coastal and groundwater — require different modelling and are often incompletely covered.
- Damage functions relating water depth to loss ratio are a major source of uncertainty and are frequently transferred between regions without recalibration.
- Flood defences alter the loss distribution non-linearly: near-total protection up to the design standard, then rapid escalation beyond it.
- Aggregation matters because a single event can affect an entire catchment simultaneously.
The hazard module
Hazard modelling produces, for each location, a relationship between event return period and flood depth. It combines meteorological inputs, hydrological modelling of runoff and river flow, and hydraulic modelling of how water spreads across terrain. Terrain data resolution is the limiting factor: a digital elevation model with coarse resolution cannot represent the small features that determine whether a given property floods.
- Fluvial flooding from rivers exceeding their channels, driven by catchment rainfall over days.
- Pluvial or surface-water flooding from intense rainfall overwhelming drainage, which can occur far from any watercourse.
- Coastal flooding from storm surge combined with tide and wave action, increasingly with sea-level rise as a baseline shift.
- Groundwater flooding, slower and more persistent, and frequently excluded from models and from policy wordings.
Pluvial flooding is the type most often under-represented, because it requires very high resolution terrain and drainage-network information that is rarely available. Portfolios can therefore appear to have limited flood exposure while carrying significant unmodelled surface-water risk.
Exposure and vulnerability
The exposure module records what is at risk and where. For flood, positional accuracy requirements are stricter than for almost any other peril: a property twenty metres from another and one metre higher can have a materially different hazard. Building attributes matter as much — the presence of a basement, ground-floor use, floor height above ground level and the location of plant and equipment.
The vulnerability module maps hazard intensity to expected damage, usually as a depth-damage curve giving a loss ratio for each water depth by building type and use. These curves are calibrated from claims and survey data where available, and transferred from other regions where it is not — a transfer that assumes comparable construction and comparable contents value, which frequently does not hold.
E[L] = Σ_r w_r × Σ_i V_i × D_i(h_{i,r})- E[L]
- — expected annual loss for the portfolio
- w_r
- — annual occurrence probability weight of event or return period r
- V_i
- — insured value of exposure i
- D_i(·)
- — damage function for exposure i, returning a loss ratio
- h_{i,r}
- — flood depth at exposure i under event r
Defences and non-linearity
Flood defences make the loss distribution strongly non-linear. Below the design standard, protection is close to complete and modelled loss approaches zero. Above it, water arrives rapidly and losses can exceed those of an undefended location because development has been permitted behind the defence.
- Model defences explicitly where their location and standard are known, rather than reflecting them implicitly in the hazard.
- Include a probability of defence failure below the design standard; breach and overtopping have different loss signatures.
- Record the condition and maintenance status of defences where available; design standard is not the same as current capability.
- Recognise that improved defences may increase exposure over time by enabling development in the protected area.
Climate change and the stationarity problem
Conventional flood models are calibrated on historical records and assume that the statistical properties of the hazard are stable. Changing rainfall intensity, sea level and catchment land use all undermine that assumption. A defence designed to a historical one-in-hundred-year standard may today protect against a considerably more frequent event.
- Apply climate adjustment factors to hazard frequency or intensity, and disclose them as an adjustment rather than embedding them silently.
- Present results under multiple climate pathways to show the sensitivity of the conclusion.
- Distinguish sea-level rise, which is comparatively well constrained, from changes in rainfall extremes, which are less so.
- Account for land-use change and urbanisation, which can affect surface-water risk faster than climate does.
Practical example
A commercial property is modelled with an expected annual loss of 0.05 per cent of insured value using postcode-centroid geocoding and a standard depth-damage curve. Refining to address-level coordinates places the building 1.2 metres higher than the centroid, reducing expected annual loss to 0.01 per cent. Adding the information that critical plant is located in a basement raises it again to 0.09 per cent.
Three estimates spanning almost an order of magnitude, all from the same hazard model, differing only in exposure detail. For portfolio steering this may be tolerable; for individual risk pricing, deductible setting or lending decisions it is not, which is why exposure data collection is the highest-return investment in most flood-modelling programmes.
Limitations and caveats
- Terrain and drainage data resolution limits achievable accuracy, particularly for surface-water flooding.
- Depth-damage curves transferred between regions may not reflect local construction, contents or repair costs.
- Defence information is often incomplete, out of date or unavailable at the required granularity.
- Historical calibration assumes stationarity that no longer holds.
- Model results differ materially between vendors for the same portfolio; single-model reliance is a control weakness.
Conclusion
Flood is the peril where exposure data quality matters most and where the gap between modelled and actual exposure is widest.
Improving location precision, building attributes and defence information generally produces a larger improvement in decision quality than any refinement of the hazard science.
References
- IPCC Sixth Assessment Report, Working Group I — Intergovernmental Panel on Climate Change, 2021
- State of the Global Climate reports — World Meteorological Organization
- Financial Management of Flood Risk — OECD, 2016
Author bio
Jonas Mohamed Osman Abdelghafour, known as Yonas Osman is an actuary, FRM and financial risk professional specialising in banking, insurance, model risk, capital modelling and quantitative risk management.