Elsevier

Engineering Geology

Volume 66, Issues 3–4, November 2002, Pages 315-317
Engineering Geology

Opinion paper
PSHA: is it science?

https://doi.org/10.1016/S0013-7952(02)00039-XGet rights and content

Abstract

Probabilistic seismic hazard analysis (PSHA) is beginning to be seen as unreliable. The problem with PSHA is that its data are inadequate and its logic is defective. Much more reliable, and more scientific, are deterministic procedures, especially when coupled with engineering judgment.

Introduction

First it was the turn of earthquake prediction.

Not that it was all that unscientific. True, the science in earthquake prediction might have fitted into Peter Potter's cocked hat, and there were all the jokes about Chinese cookies and Parkfield capers making the rounds among seismologists (Lomnitz, 1994). But watching it sink so fast, without a bang or a whimper—that was a black eye for science. A lot of credible research went down with it.

Now it is the turn of probabilistic seismic hazard analysis (PSHA). Critical discussions of some aspects of PSHA are beginning to appear in the scientific literature (e.g., Atkinson et al., 2000, Krinitzsky, 2002, Newman et al., 2001). PSHA predicts, for part of the central United States at the 2% probability level in 50 years, a seismic hazard comparable with that of the San Andreas Fault, a notorious plate boundary. To what extent are these results based on sound science? Should earthquake hazard be assumed as high in Memphis as it is in San Francisco, and where does probability come in? The criterion of “2% in 50 years” has been sanctioned by the International Code Council (2000) as being realistic for the central United States. Why 50 years? Is it conservative or the opposite? People are beginning to wonder, “Is this science?”

Section snippets

Misunderstandings in probability and statistics

When the Senior Seismic Hazard Analysis Committee (1997) of the US Nuclear Regulatory Commission officially distinguished between “aleatory” and “epistemic” uncertainty their decision was a direct result of admitting expert opinion as evidence on the same level as hard earthquake data. An earthquake was regarded as a nonrepeatable natural experiment, which needed to be interpreted by a seismologist.

But it was Gauss (1823) who firmly established the foundations of probability in measure theory,

The “P” in PSHA

In short, the problem of PSHA may be attributable to the first letter in its acronym. PSHA cannot claim the rigor and objectivity of a statistical method as long as it countenances the view that an earthquake—the source of all our data—is not a statistic but a “nonrepeatable experiment”. Nonrepeatability necessarily implies a dependence on the vagaries of expert testimony—which is presumably what the NRC means by “epistemological error”.

Statistics was never intended to work that way. It cannot

Conclusions

The shortcomings of PSHA appear to be due to a series of misunderstandings. Statistics has sometimes been presented as a collection of recipes for data processing, and such an approach may occasionally be valid when there is a large amount of data. But the assumptions made about the connections of statistics with probability, and of probability with measure theory, become increasingly critical as one attempts to squeeze the data or extrapolate a small data set over regions of sample space where

References (8)

  • G. Atkinson

    Reassessing the New Madrid seismic zone

    Eos, Trans. Am. Geophys. Union

    (2000)
  • C.F. Gauss

    Theoria Combinationis Observationum Erroribus Minimis Obnoxiae

    (1823)
  • International Building Code

    (2000)
  • E. Krinitzsky

    How to obtain earthquake ground motions for engineering design

    Eng. Geol.

    (2002)
There are more references available in the full text version of this article.

Cited by (94)

  • Multiple-peak HVSR curves: Management and statistical assessment

    2022, Engineering Geology
    Citation Excerpt :

    Two common ways to deal with this problem are the PSHA (Probabilistic Seismic Hazard Assessment – Solomos et al., 2008; Musson, 2012) and the NDSHA (Neo-Deterministic Seismic Hazard Assessment) approaches (Panza et al., 2012; Panza et al., 2021; Nekrasova et al., 2014; Panza and Bela, 2020). While for a detailed discussion about these two approaches we refer to the mentioned literature and Castaños and Lomnitz (2002), Wyss et al. (2012) and Stein et al. (2012), we underline that while the knowledge of the correct subsurface VS model is necessary for both the approaches, the main differences lie in the way quakes used in the modelling are generated: statistically (based on historical series) for PSHA while through a scenario approach (modelling a series of appropriate faults that characterize the local seismotectonic regime) for NDSHA. In short, while assessing an experimental HVSR curve, several facts need to be properly considered and three points are often poorly considered:

  • Disaster prediction and civil preparedness

    2022, Earthquakes and Sustainable Infrastructure: Neodeterministic (NDSHA) Approach Guarantees Prevention Rather than Cure
  • The view of a structural engineer about reliable seismic hazard assessment

    2022, Earthquakes and Sustainable Infrastructure: Neodeterministic (NDSHA) Approach Guarantees Prevention Rather than Cure
  • Seismic hazard and earthquake engineering for engineering community

    2022, Earthquakes and Sustainable Infrastructure: Neodeterministic (NDSHA) Approach Guarantees Prevention Rather than Cure
  • Hazard, risks, and prediction

    2022, Earthquakes and Sustainable Infrastructure: Neodeterministic (NDSHA) Approach Guarantees Prevention Rather than Cure
View all citing articles on Scopus
View full text