On The Scientific Value of "Unscientific" Data

(Seismological Research Letters, Sept-Oct, 2000)
Susan E. Hough
U.S.G.S., 525 S. Wilson Ave, Pasadena CA, 91106

As research scientists we are justifiably inclined to be skeptical about data that is subjective, or qualitative, in nature -- data such as anecdotal accounts of earthquake effects, and the modified Mercalli intensity (MMI) estimates derived from them. Especially where perceptions of earthquake ground motions are concerned, non-instrumental data is generally considered suspect at best. "Everyone knows" that people exaggerate the severity of earthquake effects; "everyone knows" that subjective perceptions are not to be trusted.

The engineering community has been known to discount instrumental data as outliers in light of more general damage patterns, but the seismology community generally turns to MMI data only out of desperation, because we need to analyze important events for which no instrumental data exist. Even so, a certain cloud of suspicion always hangs over the results from such investigations in the minds of some. And there can be remarkably little interest in ever revisiting original historical accounts once someone has done the dirty job of wading through them once.

The first purpose of this note is to suggest that it bespeaks a certain arrogance, to suppose that we have nothing left to learn from the voices from the past. Historical accounts of earthquakes are a treasure trove of information that can be revisited to test and even develop new hypotheses as our collective understanding grows.

The second purpose of this note is to summarize the reasons why subjective accounts and MMI data are often discounted, and to respond to each one by drawing on arguments from the published literature as well as my own experiences with anecdotal/historical accounts, from the New Madrid sequence of 1811-1812 especially. One of the happy surprises of my New Madrid work has been discovering the extent to which contemporary observers displayed both credibility and insight. One is reminded that 200 years ago, as now, there were some very smart people around. It can be not only illuminating, but also a lot of fun, listening to their stories.

In no particular order, then:

(1) Anecdotal accounts are unreliable . Clearly, some some objective accounts of earthquake effects are quite exaggerated; others, no doubt, are moderately but still significantly biased. But, as summarized eloquently by geologist-turned-Congressman Samuel Mitchill in 1815, "[when] five or six witnesses, who seem to have been wholly unknown to each other, agree in so many particulars, [then] their united evidence may be considered as near to the truth as we can expect to arrive". Mitchill had set out shortly after the New Madrid sequence of 1811-1812 to compile accounts of the earthquakes and develop a satisfactory physical explanation for them. His 1815 publication provides an invaluable compendium of accounts from all over the United States (the extent of the United States at that time).

Original felt reports from historical earthquakes can establish their own consistency in one of two ways: (1) in the manner that Mitchill describes, involving agreement between independent observers, but also (2) by describing effects that fit well with a modern understanding of wave propagation. In our reevaluation of felt reports from the New Madrid sequence, my colleagues and I found that over a half-dozen individuals left explicit documentation of stronger shaking at sediment sites, such as the following observation by Daniel Drake of Cincinnati, "It seems to have been stronger in the valley of the Ohio, than in the adjoining uplands. Many families living on the elevated ridges of Kentucky, not more than 20 miles from the river, slept during the shock; which cannot be said, perhaps, of any family in town." Drake even went on to explain the phenomenon in more-or-less accurate terms, ``The convulsion was greater along the Mississippi, as well as along the Ohio, than in the uplands. The strata in both valleys are loose. The more tenacious layers of clay and loam spread over the adjoining hills, many of which are composed of horizontal limestone, suffered but little derangement."

Readers familiar with the New Madrid saga will recognize Daniel Drake to have been one of the most thorough and scientific observers to have witnessed and documented the events. But he was far from alone in his powers of observation, as site response was also documented explicitly in Asheville, North Carolina; Carthage, Tennessee; Chillicothe, Ohio; Newark, New Jersey, and Brownsville, Pennsylvania. And Drake was one of two individuals who endeavored not only to document every event they felt, but also to rank the shaking by an invented severity scale. (The other person being Jared Brooks of Louisville, Kentucky, another familiar name to those who have studied the 1811-1812 sequence.)

To give a more recent example of the power of observation demonstrated by some individuals, on June 28, 1992, a 15-yr old resident of Yucca Valley reported having been wakened by the Landers mainshock, running outside, and watching the Eureka Peak Fault rupture through his backyard some 20 seconds later. By the time this account was relayed to me, I was inclined to be skeptical -- I had analyzed instrumental data from an array of instruments not too far away and concluded that the Eureka Peak surface rupture had been caused by a large aftershock approximately three minutes--not 20 seconds--after the mainshock. But, largely because of the eyewitness account, I was inspired to take a closer look at the very early coda of the Landers event. Indeed, a slant-stack analysis revealed significant energy in the early coda arriving not from the direction of the mainshock rupture (due north), but from the direction of the Eureka Peak fault (almost due east). With, then, strong evidence for two ruptures of the Eureka Peak fault, I concluded that it had, in fact, re-ruptured in rapid succession, perhaps reloaded over a time frame of minutes by late-arriving energy from the mainshock.

As another example, Tom Heaton has observed that a handful of eyewitness accounts of large earthquake ruptures do corroborate the so-called "slip pulse" model, in which slip at any one point on a fault is achieved very quickly, far faster than the duration of the event as a whole. As first documented by Bob Wallace, an eyewitness to the 1983 Borah Peak, Idaho earthquake, Mrs. Lawana Knox, reported seeing the fault scarp grow to a height of 1-1.5 meters in about one second. She also reported seeing the scarp tear along the flank of the mountains, "just as though one took a paint brush and painted a line across the hill."

I note, in closing, that the individuals whose observations I have discussed in this section were all well ahead of the scientific community. That is, by virtue of what they'd seen with their own eyes, they understood something about earthquakes that the seismological community would come to appreciate only some time later (in the case of Daniel Drake, much later). This is not to say that every eyewitness account is equally credible or insightful, but rather to suggest that it behooves us to at least consider the fact that they might be. This is especially true for observations presented in extended documents that provide evidence of the authors' reliability, such as those from Mitchill, Drake, and Brooks.

(2) Intensity data are imprecise. Here again, there is a measure of truth to the criticism -- the MMI scale provides only 12 steps with which to classify the level of shaking, the upper two of which are reserved for very extreme ground motions. Instrumentally-based measures of strong ground motion can, in contrast, be measured with considerable precision.

In a recent abstract, I showed that unit steps in MMI imply factor of two steps in ground motion parameters such as peak acceleration (pga). This can be understood via a consideration of the range spanned by both scales. At the lowest end, barely perceptible ground motions correspond to an MMI of I or II and a pka of 0.1-1% g. At the high end, an MMI of XI or XII corresponds, presumably, to the strongest levels of shaking observed, perhaps 100-200% g. Thus, 10 or 11 steps in MMI must span a factor of approximately 1000 in pga. If the factor corresponding to one unit change in MMI is X, then we have X^10 = 1000, or X=2.

MMI values therefore quantify, or bin, ground motion parameters by, approximately, factors of 2. While this may appear imprecise, one might consider a different seismological parameter, magnitude. Seismologists may recognize moment to provide the best quantification of earthquakes, but magnitude estimates are used routinely in various kinds of analysis, such as the derivation of b-values. Yet magnitude is related to moment according to, log(Mo)=1.5m + 16.05, where Mo is the physically meaningful quantity. Thus, by determining magnitude to at best +/- 0.1, we are effectively binning moment quite crudely, by factors of approximately 1.4.

Also, there is an advantage to using a scale that is crude -- determinations can be made with more confidence (the familiar trade-off between resolution and precision). Even using a scale as coarse as magnitude, the reality is that a certain level of uncertainty always plagues magnitude determinations, for small-to-moderate events especially. But each value on the MMI scale corresponds to ground motions that differ in a substantial manner from the adjacent levels: the level at which dishes rattle (IV) versus the level at which plaster cracks and unstable objects overturn (V); the level at which some chimneys are broken (VII) versus the shaking that leads to partial collapse of ordinary structures (VIII).

(3) Intensity data are unscientific. Although MMI data cannot fully characterize ground motion nearly as completely as instrumental data, numerous studies have shown intensity data to be of enormous value in quantified analyses. By comparing intensity data with seismic moment for a set of twentieth century earthquakes in eastern North America and other Stable Continental Regions (SCR), Arch Johnston has demonstrated a very good correlation between isoseismal areas (at different shaking levels) and seismic moment. This analysis provides a critical calibration with which pre-instrumental events can be analyzed.

In another study, Art Frankel showed that observed relationships between felt area and moment magnitude for SCR were consistent with a simple model incorporating geometrical spreading, attenuation, and stress drop. Frankel also showed that one could determine a high-frequency (2-4 Hz) spectral level corresponding to minimum perceptible ground motions, and estimated a value of 0.35 cm/sec at 100 km distance. Other authors, meanwhile, have concluded that, while low MMI values were controlled by fairly high frequencies (7-8 Hz, in their estimation), high MMI values correlate with energy at 0.7-1 Hz. Studies such as these indicate that information about shaking at different frequencies can potentially be teased out of MMI data.

It has even been suggested (by Paul Spudich, to name names) that MMI values in some specific instances are, essentially, instrumental data. If, for example, one knew that a specific bell in a specific church was rung for a certain length of time, one could presumably model the church tower as a single degree-of-freedom oscillator and obtain a quantified estimate of ground motion at the natural frequency of the structure. Similar analyses could be done of buildings--especially simple structures--that were or were not damaged during large earthquakes.

To sum up, then, my experience has been that the usual suspicions about intensity data are over-rated. In some cases, the reservations are also not as different as one might imagine from the ones associated with other types of data. To play a seismologist's game of Devil's advocate, consider this question -- how subjective are any number of geologic observations/interpretations?

In recent years, Internet-based tools have been developed to collect intensity data in a far more efficient manner than has ever been possible. Efforts such as this will greatly increase the quantity and quality of intensity data; these will undoubtedly be analyzed to fruitful ends. But as we focus on recording the data--instrumental and otherwise-- of the future, I suggest that we should also take the modest steps required to archive the data of the past. For pre-instrumental earthquakes, preserved accounts from eyewitnesses provide an important component of the data. As someone who has struggled to read old newspapers on microfilm and to find copies of felt-report compilations published in hard-to-find (usually "gray") reports, the solution seems obvious -- that historical accounts, once found, be transcribed and made available over the web. This could be done via a systematic or a grass-roots approach, as URLs can be linked easily enough to form a virtual database.

Indeed, a spattering of historical accounts have already been made available over the web by scientists, educators, and others. For the 1811-1812 New Madrid sequence, an exhaustive web-based compendium is already under construction under the auspices of the Center for Earthquake Research and Information (CERI).

To ensure a truly lasting record, accounts of these and other notable preinstrumental events could and should also be archived in a conventional, and easily accessible, manner, via non-electronic publication.

Just within the last decade or so, researchers have gone back to seemingly outlandish descriptions of waterfalls being created on the Mississippi River and teased out compelling evidence for thrust motion on the Reelfoot Fault in the event of 7 February, 1812. I have taken a closer look at an infamous "hoax" associated with the New Madrid sequence--that a volcano erupted in North Carolina at the time of the 12 December, 1811, mainshock--and concluded that perhaps a measure of truth can be found even in such a notorious tall tale. I concluded that at the root of this hoax were credible observations of a phenomenon known as "earthquake lights" -- a phenomenon that the seismological community has tended to dismiss as folklore until only very recently.

Other documented observations from the New Madrid sequence still seem fanciful given what is known about earthquakes and wave propagation, but maybe they also contain seeds of truths that we don't yet understand. As Samuel Mitchill wrote in 1815, after lamenting his inability to formulate a satisfactory theory to explain the New Madrid events, "I console myself...that the history which I have written will give valuable information to the curious on these subjects, and assist some more happy inquirer into nature, to deduce a full and adequate theory of earthquakes." In keeping with this spirit, Mitchill's own account can now be found at http://pasadena.wr.usgs.gov/hough/mitchill.html.