PALEOSEISMIC INFORMATION IN USGS SEISMIC-HAZARD MAPS OF THE CENTRAL AND EASTERN UNITED STATES AND THE NORTHERN CARIBBEAN

Probabilistic seismic-hazard computations require estimates of magnitudes (M), locations and ages of individual prehistoric earthquakes, and the uncertainty of each value. The U.S. Geological Survey’s seismic-hazard maps use paleoseismic data and interpretations to constrain values of these six variables. Analytical uncertainty tends to be better characterized than geologic uncertainty but smaller. Paleoseismic values that have large uncertainties can yield large uncertainty in the computed hazard. We draw on experience with seven sources of large earthquakes to suggest how geologic uncertainty might be characterized precisely for hazard computations.

The New Madrid (Missouri), Wabash Valley (Illinois-Indiana), and Charleston (South Carolina) seismic zones have paleoseismic records that consist largely of several generations of widespread liquefaction features. In contrast, along the Meers (Oklahoma), Cheraw (Colorado), Septentrional (Dominican Republic), and South Lajas (Puerto Rico) faults, records consist of scarps and related landforms. Typically hazard analysts use a paleoseismic estimate of M from a single-earthquake scarp length or slip, or the distance to the farthest liquefaction feature. Location is determined from scarp maps or the geographic distribution of liquefaction features. Age comes from structural and stratigraphic relations exposed in trenches, or the offset of a surface of loosely constrained age. Often M, location, and age are reported as single values or ranges of values. It would be more helpful if each were reported as a probability distribution (pdf). If a pdf cannot be calculated, might it be estimated qualitatively by the paleoseismologist?

We pose three questions to improve uncertainty characterization. (1) Can paleoseismologists qualitatively estimate the geologic uncertainties in M, location, and age so that the estimates could serve as proxies for standard deviations? (2) How could uncertainty be quantified if its distribution is not Gaussian or not symmetrical? (3) Could measures of sizes of liquefaction features yield more robust estimators of M than the distance to the farthest liquefaction feature? Uncertain hazard translates into increased risk of either overly high construction costs or overly low protection.