Vulnerability Functions and Damage Ratios

Why Vulnerability Functions Matter

A hazard intensity assigned to a location describes the physical event there; it does not, by itself, say how much of the location’s insured property is damaged. A vulnerability function supplies that link. It summarizes the expected proportion of a defined exposure value damaged at a given intensity, allowing the hazard module’s location-level output to be interpreted in terms of physical damage.

The function is conditional on the intensity at the location and on the exposure characteristics for which it was developed. A wind vulnerability curve for one building class is not automatically suitable for another class, and a curve expressed in wind speed cannot be read using a water-depth value. The intensity measure, units, peril, building characteristics, and damage definition all need to match.

The reference value \(V\) must be defined consistently with the damage measure. For example, a structure-only damage ratio should use a structure value, rather than silently using a combined building-and-contents value. Under a conventional physical-damage definition where damage is capped at the property’s full reference value, \(0\leq R\leq 1\), so \(M(h)\) also lies between 0 and 1. Other loss definitions can have different bounds; the curve’s documentation determines what its ratio represents.

Reading the Axes and the Curve

The horizontal axis gives hazard intensity, with its measurement units. It might be wind speed, flood depth, or another peril-specific measure. The vertical axis gives mean damage ratio, usually as a decimal from 0 to 1 or as a percentage from 0% to 100%. A plotted value of 0.18 means an expected damage ratio of 18% under the curve’s definition; it does not mean an 18% chance of damage.

The curve’s shape describes how expected damage changes with intensity. A low-intensity region with a nearly zero ratio may represent little expected physical damage at those intensities. A steep section means that a modest increase in intensity corresponds to a relatively large increase in mean damage ratio. A flattening section means the expected ratio is increasing more slowly, while a plateau indicates that the modeled mean ratio changes little across that range. These are interpretations of the function, not claims that every peril or building class has the same shape.

For the same peril and damage definition, a vulnerability curve will commonly be nondecreasing: greater intensity is not expected to reduce mean physical damage. A curve may nevertheless have irregularities because of estimation, limited data, or modeling choices. A user should not silently “correct” the supplied curve or assume monotonicity if its documented values do not support that assumption.

A curve is also specific to its exposure class and model conventions. Construction type, occupancy, building characteristics, and the definition of repairable damage can all matter. Comparing two curves is meaningful only when the intensity measure, units, and damage-ratio basis are compatible. A higher curve at a given intensity indicates a higher expected damage ratio for the specified classes and definitions; it does not establish that one class has more frequent hazard events.

From a Curve Value to Expected Location-Level Damage

A vulnerability function describes a conditional mean. To interpret the expected damage at one location, first identify the intensity at that location, then read the curve for the appropriate exposure class and damage definition. If the location has a fixed reference value \(V\), the expected physical damage on that basis is the mean ratio multiplied by \(V\). This gives a conditional expected amount, not a guaranteed amount for an individual event.

The identity follows directly from the definition of the ratio. If \(D=VR\) and \(V\) is fixed, then conditioning on \(H=h\) gives $$ E[D\mid H=h]=E[VR\mid H=h]=V E[R\mid H=h]=V M(h). $$ The second equality holds because a fixed value \(V\) can be taken outside the conditional expectation. This reasoning requires the same valuation basis for \(D\) and \(V\); if the reference value varies with the event or is defined differently from the damage measure, that relationship must be reconsidered.

The hazard intensity used here is the intensity assigned to the location, not necessarily the event’s regional maximum. An event footprint and spatial interpolation may produce different intensities at different locations. Once the location’s intensity has been established, the vulnerability function translates that value into a conditional mean ratio for the relevant class.

Interpreting an Expected Ratio Across Possible Intensities

If the intensity at a location has several possible values, the curve still gives a conditional mean ratio at each value. An overall expected ratio across those possibilities is a probability-weighted average of the curve values, provided the intensity probabilities refer to the same scenario being analyzed. This operation does not turn the vulnerability curve into an event-frequency model: the probabilities must come from a separate description of the possible intensities.

Interpolation, Extrapolation, and Model Boundaries

A curve may be supplied as plotted points or as a table. Between tabulated points, an interpolation convention is needed to obtain a value. Linear interpolation is common, but it is an assumption about the segment between points, not a universal property of vulnerability functions. If the model specifies another interpolation method, use that method instead.

Outside the supplied intensity range, extrapolation is even more consequential. Extending the last slope can produce implausible ratios, including values outside a physically defined range. Holding the curve flat, imposing a cap, or using a separate high-intensity rule are possible modeling conventions, but none should be assumed without support. State when an intensity lies outside the curve’s range and identify the documented treatment before reporting a result.

A curve’s apparent precision should also be interpreted in context. Its values are model estimates for a defined class and damage basis, not precise predictions of the damage to a particular building. The mean ratio does not describe the full distribution of possible damage outcomes. That distinction matters when a decision requires a probability of exceeding a loss amount or a measure of damage variability; a mean curve alone does not supply either one.