PGA vs Spectral Acceleration: Which Intensity Measure Should You Use for Fragility Analysis?
Every fragility curve is plotted against an intensity measure (IM), and the two most familiar candidates are peak ground acceleration (PGA) and spectral acceleration, Sa(T). Both are well established, both have ground motion models and hazard maps, and both appear in published fragility models. They are not interchangeable. This post compares them, using results from our study on converting fragility curves between intensity measures.
What is peak ground acceleration (PGA)?
Peak ground acceleration (PGA) is the largest absolute acceleration of the ground recorded during an earthquake, read directly from the acceleration time history.
It is the simplest IM there is. It needs no information about the structure, and it has a long history in seismic codes and hazard maps. Together with peak ground velocity (PGV) and peak ground displacement (PGD), it is one of the three classic peak measures. Our post on computing intensity measures from ground motion records shows how to obtain all three in Python.
What is spectral acceleration, Sa(T)?
Spectral acceleration, Sa(T), is the peak pseudo-acceleration response of a linear single-degree-of-freedom oscillator with natural period T, usually with 5% damping, subjected to the ground motion.
Plotting Sa(T) over a range of periods gives the response spectrum. Because Sa(T) filters the ground motion through an oscillator, it measures the shaking at the frequencies a structure with period T actually responds to. As T tends to zero, the oscillator becomes rigid and Sa(T) tends to PGA.
How do you judge an intensity measure?
The most useful criterion for fragility analysis is efficiency: how well the IM explains the structural response. An efficient IM leaves little unexplained scatter, so the fragility curve has a small dispersion, , in the lognormal form
A second criterion is sufficiency: the response, given the IM, should not depend much on other properties of the earthquake, such as its magnitude or distance. A third is practical: you need a ground motion model and a hazard curve for the IM at your site. PGA and Sa(T) both satisfy the practical test easily.
PGA vs Sa(T): what the numbers say
In the study by O'Reilly, Ozsarac and Shahnazaryan (2026), the same single-degree-of-freedom system with a period of 1.0 s was analysed directly in many IMs, at a site in L'Aquila, Italy. For one damage state, the fragility dispersions were:
| Intensity measure | Dispersion, |
|---|---|
| PGA | 0.72 |
| Sa(0.5 s) | 0.62 |
| Sa(1.0 s) | 0.46 |
| Saavg3(1.0 s) | 0.17 |
| FIV3(1.0 s) | 0.17 |
PGA is the least efficient IM in the table. Sa at the structure's own period does much better. Sa at the wrong period, 0.5 s for a 1.0 s structure, sits in between.
The reason is that a structure responds to the ground motion through its own dynamics. Sa(1.0 s) measures shaking at the period of this structure; PGA is governed mainly by the short-period content of the ground motion, well away from 1.0 s. There is a second effect too. As a structure is damaged and yields, it softens and its effective period lengthens. An IM tied to short periods misses both the structure's period and that elongation.
The two IMs also disagree on where the curve sits, not just its slope. At the L'Aquila site, the 475-year values are 0.19 g for Sa(1.0 s) and 0.31 g for PGA, a ratio of 1.65. Yet the ratio between the medians of the two reference fragility curves was 2.4. A fixed scaling factor between PGA and Sa does not translate one fragility curve into the other.
Does the choice affect losses?
Yes, and more than the fragility curves suggest. When vulnerability models built on other IMs were converted to PGA and integrated with the site hazard, the average annual loss (AAL) was overestimated in every case: by 30% for the Sa(1.0 s) model, by 23% on average and by up to 55%. The same Sa(1.0 s) model converted to FIV3(1.0 s) stayed within 5%.
The cause is dispersion again. A wide PGA-based model raises the expected loss at each level of shaking, and the error accumulates when it is integrated over the hazard. If you want to see how that integral works, read how to compute average annual loss from a vulnerability function and a hazard curve.
Can a PGA fragility be converted to Sa(T), or the other way round?
Converting into PGA works well: in the study's heat map, the PGA column had a mean Hellinger distance of 0.24 between converted and reference curves. Converting out of PGA does not: the PGA row had a mean of 0.62. Information that an inefficient IM never captured cannot be recovered by any conversion. That is why the main post describes PGA as a universal receiver and FIV3(T) as a universal donor.
So which should you use?
- For a single building or a well-defined building class, Sa(T) at the fundamental period is a clear improvement over PGA, and it is widely supported by ground motion models, hazard software and record selection tools. Next-generation IMs do even better: average spectral acceleration, the geometric mean of Sa over a range of periods (see Eads et al. 2015), and filtered incremental velocity (Dávalos and Miranda 2019), introduced in what FIV3 (filtered incremental velocity) is.
- For very stiff, short-period structures, the gap between PGA and Sa(T) narrows, because Sa(T) approaches PGA as the period approaches zero.
- For a regional portfolio, a single Sa(T) cannot match the period of every building class. That has historically pushed portfolio models towards PGA. A better route is an IM that works across periods: FIV3(T) converted well between its own definitions at different periods because it is nearly period-independent.
- If you inherit a model built on PGA, keep it on PGA or compute a hazard curve in PGA for it. Do not expect a conversion to a more efficient IM to be reliable.
Whatever you choose, use the same definition of the IM in the fragility model, the ground motion model and the hazard curve. Kohrangi et al. (2017) also suggest that next-generation IMs such as Saavg(T) may be less site-dependent than Sa(T), which makes models built on them easier to reuse at other sites.
Frequently asked questions
Is PGA a good intensity measure for fragility curves?
PGA is simple and widely available, but it is usually an inefficient intensity measure for structures with periods well above zero. For a 1.0 s structure, a PGA-based fragility had a dispersion of 0.72 against 0.46 for Sa(1.0 s), and harmonising vulnerability models on PGA overestimated average annual loss by 23% on average.
Which period should I use for spectral acceleration?
Use the fundamental period of the structure, or a period representative of the building class. Using a period well below it, such as Sa(0.5 s) for a 1.0 s structure, raises the dispersion and makes the model harder to convert later. Average spectral acceleration avoids the choice of a single period.
Can I convert a fragility curve from PGA to Sa(T)?
You can run the calculation, but the result will not be reliable. PGA does not capture the response information that Sa(T) would, and no conversion can recover it. Converting from Sa(T) to PGA, on the other hand, works well with total probability methods.
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
- Eads L, Miranda E, Lignos DG (2015) Average spectral acceleration as an intensity measure for collapse risk assessment. Earthquake Engineering & Structural Dynamics 44(12): 2057–2073. doi:10.1002/eqe.2575
- Dávalos H, Miranda E (2019) Filtered incremental velocity: A novel approach in intensity measures for seismic collapse estimation. Earthquake Engineering & Structural Dynamics 48(12): 1384–1405. doi:10.1002/eqe.3205
- Kohrangi M, Vamvatsikos D, Bazzurro P (2017) Site dependence and record selection schemes for building fragility and regional loss assessment. Earthquake Engineering & Structural Dynamics 46(10): 1625–1643. doi:10.1002/eqe.2873