What Is a Fragility Curve? Fragility vs Vulnerability in Seismic Risk
Every seismic risk calculation answers two questions. How often, and how strongly, will the ground shake at this site? And what happens to the structure when it does? The first is answered by the seismic hazard. The second is answered by fragility and vulnerability models. This post explains what those models are, how they differ, and how they are used.
What is a fragility curve?
A fragility curve is a function that gives the probability that a structure, or a component, reaches or exceeds a given damage state when it is subjected to a given level of ground shaking.
The level of shaking is described by a single number, the intensity measure (IM). Common choices are peak ground acceleration (PGA) and spectral acceleration at a period T, Sa(T). The damage states are discrete levels of damage, ordered from slight damage up to collapse. A fragility model normally contains one curve per damage state; the case study in converting fragility curves between intensity measures uses five damage states per structure.
Fragility curves are most often described with a lognormal cumulative distribution function:
where is the standard normal cumulative distribution function, is the median, the intensity at which there is a 50% probability of reaching the damage state, and is the dispersion, the standard deviation of the logarithm of the intensity that causes the damage state.
The median tells you where the curve sits. The dispersion tells you how steep it is. A small gives a steep curve: the structure moves from "probably undamaged" to "probably damaged" over a narrow band of intensity. A large gives a flat curve, which means a lot of the structure's response is left unexplained by the IM.
What is a vulnerability curve?
A vulnerability curve (or vulnerability function) gives the expected loss for a given level of ground shaking, .
The loss is usually expressed as a loss ratio: the repair cost divided by the replacement cost of the structure, so it runs from 0 to 1. It can also be expressed as a monetary value, or as another decision variable such as casualties or downtime. A full vulnerability model also describes the scatter of the loss around its expected value, not just the mean.
How do fragility and vulnerability curves differ?
| Fragility curve | Vulnerability curve | |
|---|---|---|
| Answers | How likely is this damage state? | How much will it cost? |
| Output | Probability, 0 to 1 | Expected loss (ratio or money) |
| Number of curves | One per damage state | One per structure or building class |
| Typical use | Damage and collapse rates, damage scenarios | Loss estimates, average annual loss |
The two are closely linked. Given a fragility model and a damage-to-loss (consequence) model, which assigns an expected loss ratio to each damage state, the vulnerability function follows from the total probability theorem:
The probability of being in damage state is the difference between two consecutive fragility curves, . This is exactly how the vulnerability functions in the intensity measure conversion study were obtained from their fragility functions.
Going the other way is not possible in general: a single expected-loss curve does not tell you which damage states produced it. That is one reason why some risk models are delivered as vulnerability only.
Vulnerability can also be built without building-level damage states at all. Storey loss functions relate engineering demand parameters, such as peak storey drift and peak floor acceleration, directly to repair costs at each storey. You can read how they are generated from a component inventory in our post on storey loss functions (EDP-DV functions), and O'Reilly and Shahnazaryan (2024) discuss their use for regional vulnerability modelling.
How are fragility curves derived?
There are three broad sources, often used in combination:
- Empirical: fitted to damage observed after past earthquakes.
- Analytical: obtained from numerical models of the structure subjected to recorded ground motions.
- Expert judgement: elicited from engineers where data and models are lacking.
Silva et al. (2019) review the current challenges and trends in analytical fragility and vulnerability modelling. A typical workflow selects ground motion records, for example with the conditional spectrum method, runs non-linear analyses at several intensity levels, counts how many records exceed each damage threshold, and fits a lognormal curve to the result. In the study behind the main post, multiple stripe analysis with 40 records per intensity level produced fragility functions in 18 different IMs for each structure.
Where non-linear time-history analysis is too expensive, simplified demand-intensity models estimate the intensity needed to exceed a given demand from the structure's static pushover curve. See demand-intensity models for RC frames, infilled frames and base isolation.
How are fragility and vulnerability curves used in seismic risk assessment?
On its own, a fragility curve says nothing about how often damage will occur. It must be combined with the seismic hazard curve of the site, , the annual rate at which the IM exceeds a value . The annual rate of reaching or exceeding a damage state is:
When the hazard curve is approximated by a power law and the fragility is lognormal, this integral has a well-known closed-form solution (Cornell et al. 2002). The same integral with the vulnerability function in place of the fragility gives the average annual loss (AAL). We go through that calculation step by step in how to compute average annual loss from a vulnerability function and a hazard curve.
Both integrals only make sense if the fragility or vulnerability curve and the hazard curve use the same IM. For a single building that is easy to arrange. For a regional portfolio, where models come from different sources and are built on different IMs, it is not. You then have to simulate several cross-correlated ground motion fields, or convert every model to a common IM, as described in converting fragility and vulnerability models between intensity measures.
Why does the choice of intensity measure matter?
The IM is not just a label on the horizontal axis. It decides how much uncertainty the curve carries. In the main post, the same 1.0 s structure and damage state gave a fragility dispersion of 0.72 with PGA, 0.46 with Sa(1.0 s), and 0.17 with both average spectral acceleration, Saavg3(1.0 s), and filtered incremental velocity, FIV3(1.0 s).
An IM that gives a small dispersion is called efficient. Efficiency matters twice: it makes the risk estimate less uncertain, and it decides whether the model can later be converted to another IM. For a direct comparison of the two classic choices, see PGA vs spectral acceleration for fragility analysis; for one of the next-generation alternatives, see what FIV3 (filtered incremental velocity) is.
Frequently asked questions
What is the difference between a fragility curve and a vulnerability curve?
A fragility curve gives the probability of reaching or exceeding a damage state for a given intensity of ground shaking. A vulnerability curve gives the expected loss, usually as a loss ratio, for that same intensity. A vulnerability curve can be built from a set of fragility curves and a damage-to-loss model.
Why are fragility curves usually lognormal?
A lognormal distribution is described by just two parameters, a median and a dispersion, it is defined only for positive intensities, and it fits analytical and observed data reasonably well. The dispersion then gives a direct measure of how well the intensity measure explains the structural response.
Can I combine a fragility curve with any hazard curve?
Only if both use the same intensity measure, defined in the same way. If the fragility is in Sa(1.0 s) and the hazard curve is in PGA, the model has to be converted to the hazard's intensity measure first, or a hazard curve in the model's own intensity measure must be computed.
References
- O'Reilly GJ, Ozsarac V, Shahnazaryan D (2026) Conversion of Seismic Fragility and Vulnerability Models to Alternative Intensity Measures for Regional Risk Analysis. Earthquake Spectra 42: e70098. doi:10.1002/esp4.70098
- Silva V, Akkar S, Baker J, Bazzurro P, Castro JM, Crowley H, Dolsek M, Galasso C, Lagomarsino S, Monteiro R, Perrone D, Pitilakis K, Vamvatsikos D (2019) Current Challenges and Future Trends in Analytical Fragility and Vulnerability Modeling. Earthquake Spectra 35(4): 1927–1952. doi:10.1193/042418EQS101O
- O'Reilly GJ, Shahnazaryan D (2024) On the utility of story loss functions for regional seismic vulnerability modeling and risk assessment. Earthquake Spectra 40(3): 1933–1955. doi:10.1177/87552930241245940
- Cornell CA, Jalayer F, Hamburger RO, Foutch DA (2002) Probabilistic Basis for 2000 SAC Federal Emergency Management Agency Steel Moment Frame Guidelines. Journal of Structural Engineering 128(4): 526–533. doi:10.1061/(ASCE)0733-9445(2002)128:4(526)