15Hough.qxd 523 ANNALS OF GEOPHYSICS, VOL. 47, N. 2/3, April/June 2004 Key words New Madrid earthquakes – intraplate – historic 1. Introduction The earthquake sequence that struck the New Madrid region of the North American mid-conti- nent in 1811-1812 had remarkably far-reaching effects. By some accounts, they are among the largest – if not the largest – to have ever oc- curred in a so-called Stable Continental Region (SCR) (Johnston, 1996). Ground motions from the three principal events were felt in places as far away as Canada, New England, and at a num- ber of locations along the Atlantic coast (Mitchill, 1815; Bradbury, 1819; Fuller, 1912; Nuttli, 1973; Penick, 1981; Street, 1984; John- ston, 1996). Contemporary accounts document three principal events: approximately 02:15 Lo- cal Time (LT) on 16 December 1811; around 08:00 LT on 23 January 1812, and approximate- ly 03:45 LT on 7 February 1812 (henceforth NM1, NM2, and NM3, respectively; see fig. 1). All three events were felt throughout much of the central and eastern United States. Addition- ally, a large aftershock to NM1 (NM1-A) oc- curred near dawn on 16 December 1811. Sub- stantial aftershock activity following all events was also documented (Fuller, 1912; Penick, 1981). Paleoseismic investigations suggest a repeat time of the order of 400-500 years for the New Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence Susan E. Hough US Geological Survey, Pasadena, California, U.S.A. Abstract The central and eastern United States has experienced only 5 historic earthquakes with Mw 7.0, four during the New Madrid sequence of 1811-1812: three principal mainshocks and the so-called «dawn aftershock» following the first mainshock. Much of the historic earthquake research done in the United States has focused on the New Madrid Seismic Zone (NMSZ), because the largest New Madrid earthquakes may represent the archetype for the most damaging earthquakes to be expected in intraplate regions. Published magnitude values ranging from 7.0 to 8.75 have generally been based on macroseismic effects, which provide the most direct constraint on source size for the events. Critical to the interpretation of these accounts is an understanding of their historic context. Early settlments clustered along waterways, where substantial amplification of seismic waves is expected. Ana- lyzing the New Madrid intensity values with a consideration of these effects yields preferred values of Mw 7.2- 7.3, 7.0, and 7.4-7.5 for the December, January, and February mainshocks, respectively, and of 7.0 for the «dawn aftershock». These values are consistent with other lines of evidence, including scaling relationships. Finally, I show that accounts from the New Madrid sequence reveal evidence for remotely triggered earthquakes well out- side the NMSZ. Remotely triggered earthquakes represent a potentially important new wrinkle in historic earth- quake research, as their ground motions can sometimes be confused with mainshock ground motions. Mailing address: Dr. Susan E. Hough, US Geological Survey, 525 South Wilson Avenue, Pasadena, CA 91106, U.S.A.; e-mail: hough@gps.caltech.edu 524 Susan E. Hough Madrid events; they also suggest that the New Madrid seismic zone tends to produce pro- longed sequences with multiple, distinct main- shocks, the magnitudes of which are compara- ble to those of the 1811-1812 events (e.g., Tut- tle and Schweig, 1996; Tuttle et al., 2002). Thus, the magnitudes of the earthquakes be- come a critical issue for the quantification of re- gional hazard in central North America. A re- peat of the 1811-1812 sequence would clearly have a tremendous impact. The New Madrid Seismic Zone (NMSZ) contributes a nontrival component of seismic hazard in relatively dis- tant midwestern US cities such as St. Louis, Missouri (Frankel et al., 1996). A second impetus to investigate the 1811- 1812 sequence stems from their implications for general issues related to intraplate earth- quake processes. The NMSZ is among the best- understood intraplate source zones in the world, largely because it has been so active throughout the historic and recent prehistoric past. This rel- ative abundance of data affords the opportunity to explore critical unanswered scientific ques- tions regarding large SCR earthquakes, most notably the questions of why such events occur in certain regions but (apparently) not in others. Because an evaluation of the magnitudes of the 1811-1812 events is so critical for several reasons, tremendous effort has been invested in gleaning quantitative information from the lim- ited available data. Available data include i) pa- leoliquefaction features preserved by the sedi- ments within the Mississippi embayment (e.g., Tuttle and Schweig, 1996); ii) the present-day distribution of seismicity in the NMSZ, which is assumed to illuminate the principal fault zones (e.g., Gomberg, 1993; Johnston, 1996); iii) first-hand reports («felt reports») of the shaking and/or damage caused by the events over the central/eastern United States (e.g., Nuttli, 1973; Street, 1984). Determination of magnitudes for the 1811- 1812 mainshocks hinges exclusively on the felt reports and their interpretation for Modified Mercalli Intensity (MMI) values. Nuttli (1973) drew isoseismal contours based on his compi- lation of approximately 40 felt reports. He de- termined mb = 7.2, 7.1, and 7.4 for NM1, NM2, and NM3, respectively, based on a relationship between ground motion and intensities from smaller and more recent instrumentally record- ed earthquakes in the central United States. With an exhaustive archival search, Street (1984) greatly expanded the number of reports (to approximately 100 for NM1) and assigned them intensity values. Street (1982, 1984) used these new data and the same method used by Nuttli (1973) to obtain mb = 7.1 and 7.3 for NM2 and NM3 and mb 7.0 for the 07:15 LT af- tershock of 16 December 1811. Street (1982) determined these values by assuming the mb value for NM1 determined by Nuttli (1973) and comparing the relative isoseismal areas of the other events. Johnston (1996) carried out a comparison be- tween intensity distribution and moment magni- tude Mw for large earthquakes in stable continen- tal regions worldwide. He compared areas with- in isoseismals of discrete intensities with instru- mentally measured moment magnitudes. On the basis of this calibration, he assigned 8.1 ± 0.31, 7.8 ± 0.33, and 8.0 ± 0.33 for NM1, NM2, and NM3, respectively. In this calculation, Johnston (1996) used the only published intensity con- tours; those determined by Nuttli (1973). Hough et al. (2000) revisited the magnitude determination for the New Madrid mainshocks in two ways. First they reconsidered intensity assignments for the reports compiled by Nuttli (1973) and Street (1984). This reinterpretation focused on effects that were considered rela- tively objective, such as descriptions of damage to structures. The reinterpreted MMI values can be used to define new isoseismal contours using sub- jective as well as systematic approaches, and the isoseismal contours can then be used to ob- tain Mw estimates following the procedure and calibration established by Johnston (1996). The results can then be interpreted with a con- sideration of their historic context, most no- tably early American settlement patterns. The population of the United States was ≈ 7 000 000 in 1811, with sizable numbers in the states of Tennessee, Kentucky, and the re- gion including the present-day states of Mis- souri and Louisiana. The 1810 Census gives the population for several districts for which felt reports are considered, including the Dis- 525 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence Fig. 1. Map showing location of the New Madrid seismic zone as illuminated by microseismicity between 1974 and 1996. Locations are from the New Madrid catalog (see Taylor et al., 1991), which are reported only to two significant figures in decimal degrees. Epicenters of the three principal 1811-1812 mainshocks are shown with large open circles (after Johnston and Schweig, 1996). Solid line shows inferred location of Reelfoot Fault (af- ter Odum et al., 1998). Rupture scenarios for NM1, NM1-A, and NM3 are also indicated. Scenario for NM2, in- dicated with dashed line, is considered relatively uncertain. 526 Susan E. Hough trict of St. Louis (population 5667), Cincinnati (2540), New Orleans (24 552), Louisville (1357), and New Madrid (2103). Although present-day Missouri was relative- ly sparsely populated in 1811, available contem- porary accounts (e.g., Brackenridge, 1817, Bradbury, 1819) provide a fairly thorough docu- mentation of demographic and related informa- tion. These sources reveal that some towns were more than simple villages by 1811, with solidly constructed houses appearing by the turn of the century. The oldest brick building west of the Mississippi was built in the town of Sainte Genevieve in 1804; this town is along the Mis- sissippi River valley north of New Madrid. This house and ≈ 50 others that predate the New Madrid sequence, are still standing today. This paper summarizes and expands on the results published by Hough et al. (2000). The reader is referred to this publication for many of the details regarding the results summarized here. Additionally, I summarize evidence that the 1811-1812 sequence included a number of significant, potentially damaging earthquakes, that occurred well outside of the NMSZ. 2. Intensity reports 2.1. Original sources: general considerations Hough et al. (2000) concluded that many of the original MMI assignments by Nuttli (1973) were too high for two basic reasons: a general bias in the interpretation of reports whose dra- ma is belied by low levels of actual damage re- ported, and, to a lesser extent, a failure to take site response issues into account. Many of the original accounts describe the effects of long- period shaking; this kind of shaking can be dra- matic for large (Mw 7 and larger) events at re- gional distances even when the overall effects of the shaking is low. Overall, many of the accounts do not appear to support values as high as those originally as- signed. In St. Louis, Missouri, Fuller (1912) de- scribes reports (from the Louisiana Gazette, 21 December 1811) of people having been wak- ened by NM1 and furniture and windows hav- ing been rattled. He notes that «several chim- neys were thrown down», and a few houses «split». To understand such accounts one must be familar with the vernacular of the time; in this case, the word «split» seems to have been used in a number of accounts to mean «cracked» rather than destroyed. Consistently, as in the above example, the phrase «thrown down» is used to describe catastrophic damage to chimneys, walls, or houses. The Louisiana Gazette account goes on to note that «no lives have been lost, nor has the houses sustained much injury». This observation also suggests that the word «split» does not imply substantial damage. On the basis of these reports, a MMI of VI-VII appears to be more appropriate than the value of VII-VIII that Nuttli (1973) assigns for NM1. At many locations at regional distances (roughly 500-1000 km) event NM1 is generally reported as having been «distinctly» (often the word «sensibly» is used) felt but with no reports of damage. Instead, reports describe the rattling of washing stand pitchers, glass, china, and fur- niture. Reports from these locations also gener- ally indicate that «many» were awakened by the event. Such descriptions are consistent with an MMI value of IV-V, whereas higher values were assigned in the earlier study. In two instances (Arkport, in Western New York, and Lexington, Kentucky) it appears that Nuttli (1973) was sim- ply mistaken in either his reading of the original sources, or the MMI assignments. The reinterpretation of Hough et al. (2000) thus represents both a revision and a substantial expansion of the original intensity work by Nuttli (1973). 2.2. Site response In addition to the reassignments discussed above, Hough et al. (2000) also assigned – and interpreted – MMI values with a consideration of site response. Arguably, the key to under- standing the effects of the New Madrid earth- quakes lies with an appreciation for their his- toric context. As a first-order observation, the intensity data are very sparse and concentrated along major river valleys and other bodies of water. The latter observation reflects the distri- 527 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence 1815). However, the potential magnitude of site amplifications at regional distances has perhaps not been fully appreciated until it was so dra- matically demonstrated in a number of destruc- tive earthquakes in recent years (e.g., Singh et al., 1988). Recent dramatic examples of site re- sponse have tended to involve lake beds, valleys or basins, and coastal regions such as the San Francisco Bay area, but significant site amplifi- cations along river valleys have also been doc- umented (e.g., Stover and Von Hake, 1982). A close reading of original sources reveals that the role of site response in controlling ground motions from the New Madrid events is documented in several contemporary accounts of the events. For example, Fuller (1912) quotes an account by Daniel Drake of Cincinnati, Ohio: «(Event NM1) was so violent as to agi- tate the loose furniture of our rooms, open par- tition doors that were fastened with falling latches, and throw off the tops of a few chim- neys in the vicinity of the town». It was this ac- count that apparently prompted Nuttli (1973) to assign a MMI value of VI-VII for Cincinnati for NM1, yet Drake goes on to say that, on the «el- evated ridges» in Kentucky, less than 20 miles from the river, many people were not awakened by the event. This account (in particular the fact that many people away from the river slept through the event) suggests a MMI value of perhaps IV, certainly not as high as V. Consid- ering reported effects from the river valley and those from higher ground, one obtains a MMI range of IV-VI for Cincinnati, or an average of V. Equivalently, this approach corresponds to separate assignments for river valley and hill sites at Cincinnati. Of the felt reports from the New Madrid sequence, site response is explicit- ly documented at six different locations. The town of Sainte Genevieve, which had been moved onto a hard-rock site, provides a key example. No account of the earthquakes from this town was included in the compilation of Street (1982). A brief account was discov- ered by the author following a focused archival search. The account states that the earthquakes were felt in Sainte Genevieve but caused no damage (Rozier, 1890). The pristine, original appearance of brick and other masonry homes in the town also testifies to the absence of dam- bution of the overall population in the more sparsely populated parts of the central and southeastern United States in the early 1800s. Because the New Madrid sequence predates the construction of railroad lines into the midconti- nent, settlements tended to remain clustered in proximity to waterways. Westward expansion followed the major rivers, and virtually all early 1800s settlements in Missouri (the extent of the western frontier at that time) were within a few miles of the Mississippi River. In addition to the influx of settlers from the east, settlers of French descent also arrived in the area from Quebec to the north, primarily along the Wabash River. By the early 1820s, early settlers had begun to recognize the pitfalls associated with life on the immediate river banks, which included poor drainage, floods, and disease (Missouri Histor- ical Review, 1911). However, the very earliest settlements of the late 1700s and very early 1800s often were on fluvial sites, immediately adjacent to rivers. New Madrid was built so close to the river bank that even before the earthquakes, parts of the town regularly gave way under the continued assault of river cur- rents (Penick, 1981). One of the other sizable Missouri settlements of the time, Sainte Genevieve, had been moved to higher ground approximately a mile from the river after a flood in the late 1700s resulted in substantial erosion of the river bank upon which the town had originally been built (Brackenridge, 1817). This town, which is 160 km north of the town of New Madrid and 75 km south of St. Louis, provides a unique hard-rock sample point, as I will discuss later. Notwithstanding a handful of exceptions, at the time of the 1811-1812 sequence, the popu- lation of the US was clustered in proximity to waterways, especially throughout the sparsely populated mid-continent. Intensity data from river bank and other coastal regions will almost certainly reflect a significant site response re- sulting from the amplification of seismic waves in unconsolidated (and often water-saturated) sediments. The importance of site response in controlling earthquake ground motion has been understood for over a century (Milne, 1898), and even correctly inferred by one astute wit- ness to the New Madrid sequence (Drake, 528 Susan E. Hough age. This illustrates an important general point about poorly sampled intensity data from his- toric earthquakes: people are more likely to document their observations (by writing letters, etc) if their experiences were dramatic, than if a felt earthquake had little real impact. It also provides prima facie evidence that the hard- rock ground motions from the New Madrid earthquakes were not damaging (even to vul- nerable structures) at a distance of ≈ 160 km. For those cases where shaking and/or damage is reported to have been worse within a valley or along a riverbank than on adjacent higher ground, one can assign distinct MMI values for river- bank/valley sites and «hard rock» sites away from the waterways. Where the reports do not explicit- ly document relatively higher shaking along shorelines, Hough et al. (2000) do not attempt to correct the MMI values for site response. Howev- er, in some cases, it appears that high intensity values were assigned based solely on riverbank effects which may have been the result of agita- tion of the river itself; some of these values were downgraded. Clearly, differentiating between the effects of river disturbances along the Mississippi and ground shaking is difficult, if not impossible. In the final analysis, some level of bias will inevitably remain in any set of interpreted MMI values. However, in some cases the available data are sufficient to assign a more representa- tive regional MMI level based not on the maxi- mum effects reported at soft-sediment sites but on a full consideration of all available reports. Overall, Hough et al. (2000) assigned sig- nificantly more MMI IV-V values and signifi- cantly fewer VI-VII ones compared to the earli- er studies, although in a few instances their MMI assessments for a given location were higher than those of Street (1982). Clearly, however, the reinterpreted values were lower in general than those assigned in the earlier stud- ies, which implies that the differences are due to systematic differences in interpretation rather than random differences in interpretation of ambiguous accounts. A final map of reinterpret- ed MMI values for event NM1 is shown in fig. 2. These results include MMI values based on data from the following sources: Mitchill (1815), Fuller (1912), Nuttli (1973), and Street (1984), as well as a small number of additional sources, including a single point west of the Mississippi River, at the location of Fort Osage. 3. Isoseismal areas Considering the data shown in fig. 2, it is clear that isoseismal contours are not well-con- strained. To obtain magnitude estimates using the equations derived by Johnston (1996), how- ever, one must estimate isoseismal areas. Hough et al. (2000) employed three different approach- es to contour the data from NM1: one subjec- tive, one based on the least-squares minimiza- tion schemes presented by Seeber and Arm- bruster (1987) (see also Armbruster and Seeber, 1987), and one in which the MMI values are treated as Boolean data. If a data point falls within the appropriate isoseismal area (e.g., a value of IV that falls between the MMI IV and V contours) the residual is zero. If a data point is outside the appropriate contour, the residual is equal to the (whole number) difference between the observed and calculated values. This ap- proach was designed to reproduce the usual con- ventions applied when intensity data are con- toured subjectively. That is, isoseismals are gen- erally drawn to outline areas of equal intensity. In both regressions, the starting model for the falloff of intensity with distance is derived from the empirical equations of Johnston (1996). The inversion schemes then allow for it- eration away from this model based on the dis- tance decay of the data. Using the regression approaches, the treat- ment of the «not felt» (NF) reports becomes a critical issue. Following Street (1984), Hough et al. (2000) assign a NF value to those loca- tions where a local newspaper is known to have not mentioned an earthquake as having been felt in that location. Because NM1 and NM3 occurred at times when people can be assumed to have been asleep, a NF report is taken to in- dicate a bound of MMI < IV (the shaking level at which «some» people are awakened (Stover and Coffman, 1993). For NM2, which occurred around 08:00 a.m. LT, a NF report is taken to imply a bound of MMI < III. Hough et al. (2000) do not attempt a subjec- tive contouring of the data for events NM2 and 529 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence Fig. 2. MMI values based on a reinterpretation of original felt reports from towns as documented by Nuttli (1973) and Street (1984). Interpolations are done using a standard mathematical algorithm (for details see: Hough and Martin, 2002); black circles indicate locations where MMI values are available, while outlined gray circles indicate locations where Hough et al. (2000) assigned a «not felt» value. < .17 .17-1.4 1.4-3.9 3.9-9.2 9.2-18 18-34 34-65 65-124 >124 < 0.1 0.1-1.1 1.1-3.4 3.4-8.1 8.1-16 16-31 31-60 60-116 >116 None None None Very light Light Moderate Moderate/Heavy Heavy Very heavy Not felt Weak Light Moderate Strong Very strong Severe Violent Extreme Boolean approach would be preferred given a sufficiently complete set of felt reports, it is yielding overestimates of isoseismal areas for the New Madrid events because of the biased sampling of site conditions. The least squares regressions, on the other hand, result in contours that are closer to what one would draw subjec- tively based on an assessment of site response. The least squares results for events NM2 and NM3 yield the preferred magnitude estimates. For event NM1 the preferred estimate is the one resulting from the subjective contouring. Al- though it is not possible to quantify the uncer- tainties precisely, the bootstrap results do pro- vide a good general indication of the appropriate error bars. The final, preferred estimates for the three events are Mw 7.2, Mw ≈ 7.0, and Mw 7.4, respectively, with uncertainties of ≈ 0.3 units in each case. Hough and Martin (2002) estimate Mw 7.0 for the dawn aftershock, and conclude that this event most likely occurred on a south- eastern segment of the Reelfoot thrust fault. 5. Remotely triggered earthquakes In his compilation of accounts of the New Madrid sequence, Street (1982) compiled a list of all events for which there are multiple accounts, identifying a number of «large after- shocks» that were widely felt. At the time of this earlier study the seismological communi- ty did not yet generally appreciate the fact that large earthquakes are capable of triggering events at distances far greater than those asso- ciated with classic aftershocks. Since the 28 June 1992, Landers, California, mainshock, however, numerous studies have documented the reality of so-called «remotely triggered earthquakes» (e.g., Hill et al., 1992). Trigger- ing appears to be associated with dynamic strain associated with the surface wave (Gomberg and Davis, 1996). Although the physical mechanism where by strains cause distant earthquakes remains unclear, remotely triggered earthquakes are generally assumed to be earthquakes that would have happened at some point, but that were «nudged along» by the triggering event. It is possible, however, that remotely triggered earthquakes represent 530 Susan E. Hough NM3. Given the sparsity of the data for these events, both the ellipticity and the shape of the distance decay are fixed to match that determined for NM1. The decision to fix ellipticity is a prag- matic one; allowing another free parameter with the sparse data results in unstable solutions. Once the isoseismal contours are deter- mined, Mw values can be estimated from each individual isoseismal contour using the equa- tions derived by Johnston (1996). Johnston (1996) derives western correction factors for extrapolation of isoseismals from the New Madrid sequence to the west, using the 1843 Marked Tree, Arkansas earthquake to derive correction factors for NM1 and the 1895 Charleston, Missouri, earthquake to derive a different set of factors for NM2 and NM3. Hough et al. (2000) used the same factors. The method of Johnston (1996) yields inde- pendent estimates of Mw from each isoseismal area (MMI 4-8) from each event. To obtain an average Mw for each event, one can estimate seismic moment, Mo using the standard formu- la, log (Mo) = 1.5 Mw + 16.05, and compute an average Mo value that we then translate it to Mw. To investigate the uncertainties associated with each regression, we apply a bootstrap analy- sis in which isoseismals are fit using 50 random- ly resampled sets of data points. For each intensi- ty, the five most extreme results are discarded and bounds are estimated from the remaining 45 sets. The uncertainty ranges resulting from the boot- strap analysis are approximately ± 0.1-0.2 units for NM1 and ± 0.2-0.4 for NM2 and NM3. The reader is referred to Hough et al. (2000) for a more thorough discussion of these results. 4. Interpretation For NM1 the range from both regressions is 0.3 units. For NM2 and NM3, ranges of 0.35-0.7 are inferred. However, considering the range of results from both the Boolean and least squares approaches for each event, one obtains uncer- tainties of approximately a full magnitude unit for all three events. Hough et al. (2000) con- cluded, however, that the magnitudes of each event are constrained to better than ± 0.5 magni- tude units. They concluded that while the 531 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence ly strengthening to «tremendous», which is the descriptor Brooks reserved for the most severe levels of shaking (McMurtrie, 1819). According to Brooks, the strongest shaking from NM3-B lasted only a few seconds, suggestive of an event in the midwest rather than the New Madrid region. Daniel Drake also described the ground motions from NM3-A and NM3-B as having been qualita- tively different from those caused by other events. Evaluating the distribution of intensities with the Johnston (1996) regressions, one obtains magni- tudes of ≈ 4.5 and 5.0-5.5 for NM3-A and NM3- B, respectively, and locations well outside the NMSZ (fig. 3). Event NM2-A, which followed NM2 by 4 days, was apparently smaller than NM3-A and NM3-B; Hough (2001) does not determine a magnitude for this event. Additionally, a hand- ful of accounts describe the NM2 mainshock as having comprised multiple episodes of shaking within a few minutes. While not definitive, this suggests that remote earthquakes of substantial size could have been triggered in the immediate wake of the S/surface wave arrivals generated by NM2. Such immediate triggering is often observed following modern earthquakes (e.g., Hough and Kanamori, 2002). In addition to the inferred remotely trig- gered earthquakes discussed above, Hough and Martin (2002) analyzed accounts from a large events that would not have occurred other- wise. Hough (2001) reexamined three of the large «aftershocks» of the New Madrid sequence, events occurring at approximately 08:45 a.m. (LT) on 27 January 1812; 08:30 p.m. (LT) on 7 February 1812; and 10:40 p.m. (LT) on 7 Feb- ruary 1812 (hereinafter, NM2-A, NM3-A, and NM3-B, respectively; table I). The first of these events followed NM2 by approximately four days; the second and third events occurred the night following the NM3 mainshock (which occurred at approximately 03:45 a.m., LT). Many of the accounts describe the shaking from NM3-B as «severe» or «violent». The NM3- A event is generally described as less severe than NM3-B, but still strong. Two individuals experi- enced the New Madrid sequence and endeavored to not only document every event they felt, but al- so to rank the events by severity of shaking: Daniel Drake of Cincinnati, Ohio, and Jared Brooks, of Louisville, Kentucky (see, McMurtrie, 1819; Fuller, 1912). Brooks describes NM3-A as having been, «violent in the first degree, but of too short duration to do much injury». (He presum- ably means short in relation to the shaking from the New Madrid mainshocks, which are typically described as lasting for 2-4 min. in the Ohio-Ken- tucky region). Brooks describes the shaking from NM3-B as «violent in the second degree», quick- Table I. New Madrid sequence: mainshocks, principal aftershock, and triggered events. Event Year Month Day hr:min Long. Lat. Mw NM1 1811 12 16 02:15 –90.0 36.0 7.2 NM1-A 1811 12 16 07:15 –89.5 36.3 7.0 NM1-B 1811 12 17 noon –89.2 34.6 6.0 NM2 1812 1 23 08:45 –89.7 36.6 7.0 NM2-A 1812 1 27 09:00 –84.0 38.9 NE NM3 1812 2 7 03:45 –89.6 36.4 7.4 NM3-A 1812 2 7 20:30 –84.0 38.9 ≈ 4.5 NM3-B 1812 2 7 22:40 –84.0 38.9 5.0-5.5 Event; year, month, day; local time; crudely estimated longitude and latitude in decimal-degrees north and west; preferred moment magnitude estimate, NE: no estimate. 532 Susan E. Hough Fig. 3. Map showing inferred locations of principal and remotely triggered earthquakes that occurred during the 1811-1812 «New Madrid» sequence. Black stars indicate locations given in table I; gray star indicates pos- sible source zone for event NM2 as proposed recently by Mueller et al. (2004). event that occurred near noon (LT) on 17 De- cember 1811. They obtain a preferred magni- tude estimate of 6.0 and a location well south of the NMSZ (table I; fig. 3). 6. Discussion and conclusions The magnitude of the principal New Madrid mainshocks has been the matter of some debate in recent years. As is often the case with historic earthquakes, macroseismic data provide the most direct constraint on mag- nitude. As summarized in this paper, the key to interpreting such data in this case is twofold: 1) to assign intensity values based on objec- tive observations such as damage to structures rather than on the apparent drama of anecdotal accounts, and 2) to consider the earthquakes’ ef- fects in light of their historic context. The pre- ferred estimates for the moment magnitudes of the three principal events are Mw 7.2, ≈ 7, and 7.4-7.5, respectively, and 7.0 for the dawn after- shock. These values are more consistent with 533 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence other lines of evidence, such as geomorphology, that provide indirect constraint on event size. The geometry and extent of the Cottonwood Grove Fault, which is assumed to have generat- ed NM1, is established primarily from recent microseismicity (e.g., Gomberg, 1993; John- ston and Schweig, 1996). Event NM2 is diffi- cult to analyze in any detail because the in- ferred causative fault, the northern strike-slip limb of the NMSZ (Johnston, 1996), is the least well-understood part of the zone. Also, al- though the hour of the event provided a better characterization of the low-intensity (MMI III- IV) field, reporting of the event was likely ham- pered by a cold spell that had frozen the river and halted boat traffic along the Ohio River and Mississippi River until 22 January 1812. Very recently, Mueller et al. (2004) have reexamined this event and questioned whether it occurred within the NMSZ at all. At a minimum, this event remains the least well-understood of the three principal mainshocks. However, recent investigations have provid- ed significant constraint of the Reelfoot Fault, the thrust fault in between the two strike-slip limbs of the NMSZ that is inferred to have pro- duced NM3 (e.g., Russ, 1982; Kelson et al., 1992; Johnston, 1996). Structure of the Reelfoot fault has been elucidated in recent years with seismic reflection profiling. Odum et al. (1998) infer an overall fault length of at least 30 km and constrain the dip to be ≈ 31°. More recently, Champion et al. (2001) concluded that the Reelfoot fault does not exhibit clear geo- morphic expression to the southeast of the northern terminus of the Cottonwood Grove fault. Given their inferred spatial extent of the actively deforming Reelfoot fault, they con- clude the fault could host plausibly a low- to mid-Mw 7 earthquake. Gomberg (1993) reached similar conclusions based on the fault area esti- mated from current microseismicity. Although no direct measurements of fault scarp height are available for NM3, contempo- rary accounts from boats on the Mississippi de- scribe waterfalls forming on the river. As dis- cussed by Odum et al. (1998), these observa- tions correspond to points where the inferred fault rupture crossed the river. The height of these waterfalls is not well constrained, although some information can perhaps be gleaned from available reports. In light of these accounts and the established geometry of the Reelfoot Fault, one obtains an average slip of 4-5 m. One can account for a Mw of 7.5 on the Reelfoot fault with plausible rupture parame- ters: a length L of 40 km, a width W of 30 km (consistent with the maximum depth of micro- seismicity, see Gomberg, 1993), and an average slip D of 5.0 m. These rupture parameters are consistent with established scaling relationships derived from worldwide events (Wells and Cop- persmith, 1994). Given a rupture length of 40 km and an average slip of 5 m, one obtains Mw 6.9 and Mw 7.5 from the scaling relationships for rupture length versus magnitude and slip versus magnitude, respectively. The faulting parameters illustrated in fig. 1 are thus consistent with the lateral and depth ex- tent of the NMSZ faults as inferred from micro- seismicity. The static stress drop values are also consistent with those inferred from finite fault in- versions of more recent large earthquakes (e.g., Hough, 1996). Clearly one cannot prove that the New Madrid earthquakes did not have excep- tionally high static stress drop values. However, Hanks and Johnston (1992) showed that high- frequency shaking, and thus isoseismal areas, will depend on stress drop. Thus if the New Madrid events had higher stress drop values than the average of those used to obtain the regression results of Johnston (1996), the magnitudes of the New Madrid earthquakes would be lower than the estimates derived from these results. Because the regression results of Johnston (1996) were calibrated with similarly subjective data, one critical question is the extent to which the Hough et al. (2000) assignments are consis- tent with those on which the regressions were based. To answer this question, one must con- sider both our general approach to the MMI as- signments as well as our treatment of site re- sponse issues. In general, there is some prece- dent for keying an MMI assessment on the most dramatic effects described. However, consider- ing the MMI assignments made for the 1968 mb 5.3 southcentral Illinois earthquake (Gor- don et al., 1970) as an example, it is clear that an MMI of VI is typically assigned when there are multiple instances of the damage usually as- 534 Susan E. Hough sociated with this level of intensity: broken windows, cracked plaster, damage to brick chimneys, etc. At some locations the specific report suggesting a high MMI value in the New Madrid sequence is one that suggests relatively long-period shaking effects. There is ample precedent for not assigning an MMI value based on such a report when the effects related to higher-frequency shaking (i.e. toppling of small objects and furniture) indicate a much lower value (e.g., Armbruster and Seeber, 1987). In general, there is a fundamental distinc- tion between the 1811-1812 New Madrid events and those used by Johnston (1996) to de- rive the isoseismal area-moment magnitude re- gressions: the latter events are those for which instrumental magnitudes are available, which means they are from the 1900s (1925 onward). The New Madrid sequence is upward of 100 years less recent, and so its collection of felt re- ports is considerably more sparse than the oth- ers. Systematic differences in sampling of site conditions can clearly introduce substantial bi- ases. In 1811-1812, logistical constraints in- duced most of the population to live along river banks (or coasts), which are often characterized by alluvial near-surface geological conditions. Later in the nineteenth century, the introduction of round transportation allowed settlement to shift to higher ground, away from potential flooding hazard. Sediment-induced amplifica- tion is therefore much more likely to affect re- ports from the early part of the nineteenth cen- tury than those from the twentieth century (or even the mid-nineteenth century). Although this probably results in a systematic bias in the 1811-1812 intensity data, I do not correct for it systematically in our assignment of MMI val- ues. It would, indeed, not be appropriate to «correct» MMI values for site response and then apply the Johnston (1996) regressions be- cause the MMI data used to derive the regres- sions were not similarly corrected. I have, however, addressed the issue of site response in two ways: i) by revising the MMI assignments where contemporary accounts do document significant site response, which we view as consistent with the usual practice of as- signing site-specific MMI values based on site- specific information, and ii) by using judge- ment in choosing preferred isoseismal contours. The subjective contouring approach is con- sidered to be the most reasonable proxy for the ideal procedure, which one is unable to do in this case: to allow the MMI values to define the shape of the contours, with clear definition of high-intensity lobes. We conclude that it would clearly be inappropriate to allow the contours to «balloon» out, as was done by Nuttli (1973), based on values that surely represent «spokes» of anomalously high ground motions. A systematic site response correction could be done via a careful consideration of intensity distributions from more recent events. Hopper et al. (1983) present a map of isoseismals ex- pected from a repeat of a New Madrid main- shock in which site response is included im- plicitly. I note, however, that site corrections for the 1811-1812 data would require a very de- tailed analysis because settlement patterns changed so drastically in the decades following the New Madrid sequence. Hough et al. (2000) focused on the moder- ate intensity contours because their isoseismal areas are the critical inputs to the area-based magnitude determination method of Johnston (1996). Isoseismal contours for MMI levels IV- VII can be constrained by relatively objective reports of damage to structures and the percep- tions of individuals who (it can generally be as- sumed) were asleep at the time of events NM1 and NM3. The felt reports closer to the New Madrid seismic zone are relatively incontro- vertible in documenting the extent of damage and ground failure. However, interpretation of these reports is greatly complicated by the vast extent of poorly consolidated and largely water- saturated sediments within the Mississippi em- bayment. Once again, the natural settlement patterns would have resulted in a strong corre- lation between population density and proximi- ty to the Mississippi River. According to the results of Hough et al. (2000), NM1 was also similar in magnitude to the 1886 Charleston, South Carolina, earthquake, which Johnston (1996) estimates to have been Mw 7.3. Although perhaps at odds with «conven- tional wisdom» regarding the relative sizes of the two events, we note a striking degree of reciproc- 535 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence ity between our results and the isoseismal con- tours from the Charleston event determined by Bollinger (1977). Both events generated values of MMI ≈ V for areas midway between New Madrid and Charleston, and Charleston generat- ed a small swatch of MMI ≈ VI values in the im- mediate vicinity of New Madrid. Clearly, magnitude estimates for the New Madrid mainshocks will always be plagued by a certain level of uncertainty; a level that is, more- over, difficult to even quantify. We argue that the central issue is not one of precisely determined mathematical uncertainties but rather overall con- sistency and credibility. Magnitudes of 7.2-7.3, 7.0, and 7.4-7.5 for the three principal main- shocks, NM1, NM2, and NM3, respectively, are consistent with both known and plausibly in- ferred faulting parameters and the shaking distri- bution as inferred from our reinterpreted MMI values. Magnitude values significantly lower than these estimates strain credulity for two reasons: i) comparisons with more recent, better-con- strained events, such as the 1886 Charleston, South Carolina, and 1926 Grand Banks earth- quakes (Bent, 1995), and ii) the evidence, dis- cussed above, that NM3 was associated with significant surface faulting. On the other hand, significantly larger mag- nitude values are problematic for reasons that have been addressed at length in other studies; primarily, the lack of sufficient fault area and slip to generate three separate events with Mw close to 8.0. (Although one could plausibly ar- gue for a greater depth extent of large earth- quake ruptures in the NMSZ, even a factor of 2 increase in fault width would increase Mw esti- mates by only 0.3 units). One interesting consequence of our reinter- pretation concerns the relative magnitudes of the three principal mainshocks. In the interpre- tation of Johnston (1996), NM1 is larger than NM3, and NM2 is of appreciable size. In con- trast, our results reveal NM3 to be substantially larger than the other two. This implies that rather than being a mostly strike-slip system with thrust faulting associated with a compres- sional stepover, thrust faulting may have been the dominant mechanism associated with the 1811-1812 New Madrid sequence. At least, this would be the case if the mechanism of NM3 was predominantly thrust, as has generally been inferred. A predominant thrust mechanism is consistent with the hypothesis that post-glacial rebound provides the driving force for large late Holocene earthquakes in the NMSZ (e.g., Wu and Johnston, 2000). Although the results of Hough et al. (2000) represent a «down-grading» of the magnitudes of the principal New Madrid mainshocks, sev- eral lines of evidence argue for substantial dis- tributed hazard throughout the North American mid-continent. First, the hazard is a function of the expected ground motions, which, in the case of the New Madrid sequence, appear to have been significantly elevated in many cases by site response. An evaluation of site response may therefore be critical for seismic hazard as- sessment at many locations in the central/east- ern United States, particularly those immediate- ly adjacent to major rivers and the Atlantic seaboard. Secondly, remotely triggered earth- quakes potentially represent an additional source of distributed hazard. Finally, the recent earthquake history of western India reveals that large earthquakes can occur close together in time, not on the same fault but on neighboring faults (e.g., Hough and Bilham, 2003). Al- though the 1811-1812 New Madrid sequence provides a unique and critically important data set, a more thorough investigation of potential neighboring and regional source zones in the midcontinent appears to be warranted. Acknowledgements I thank Jim Dewey, Lucy Jones, Hiroo Kanamori, Bill Bakun, Jim Dolan, Paul Bodin, Ruth Harris, and Jerry Hough, for helpful com- ments and suggestions. I also appreciatively ac- knowledge the constructive criticism from two anonymous reviewers, as well as substantial ef- fort on the part of the editors to make this vol- ume happen. REFERENCES ARMBRUSTER, J. and L. SEEBER (1987): Seismicity 1886- 1889 in the southeastern United States: The aftershock sequence of the Charleston, S.C. earthquake, U.S. Nu- 536 Susan E. Hough cl. Regul. Comm., Washington, DC, Rep. NUREG/CR- 4851, pp. 153. BENT, A. (1995): A complex double couple source mecha- nism for the Ms 7.2 1929 Grand Banks earthquake, Bull. Seismol. Soc. Am., 85, 1003-1020. BOLLINGER, G.A. (1977): Reinterpretation of the intensity data for the 1886 Charleston, South Carolina, earth- quake, U.S. Geol. Surv. Prof. Pap. 1028, 17-32. BRACKENRIDGE, H.M. (1817): Views of Louisiana: Contain- ing Geographical, Statistical and Historical Notices of that Vast and Important Portino of America (Schaeffer and Maund, Baltimore), Md. BRADBURY, J. (1819): Travels in the Interior of America in the Years 1809, 1810, and 1811 (Sherwood, Neely, and Jones, London). CHAMPION, J, K. MUELLER, A. TATE and M. GUCCIONE (2001): Geometry, numerical models and revised slip rate for the Reelfoot Fault and trishear fault-propaga- tion fold, Eng. Geol., 62, 31-49. DRAKE, D. (1815): Natural and Statistical View, or Picture of Cincinnati and the Miami County, Illustrated by Maps (Looker and Wallace, Cincinnati). FRANKEL, A., C. MUELLER, T. BARNHARD, D. PERKINS, E.V. LEYENDECKER, N. DICKMAN, S. HANSON and M. HOP- PER (1996): National seismic hazard maps: documenta- tion, U.S. Geol. Surv. Open File Rep. 96-532, pp. 69. FULLER, M.L. (1912): The New Madrid earthquakes, U.S. Geol. Surv. Bull., 494. GOMBERG, J.S. (1993): Tectonic deformation in the New Madrid seismic zone: Inferences from map view and cross-sectional boundary element models, J. Geophys. Res., 98, 6639-6664. GOMBERG, J. and S. DAVIS (1996). Strain changes and trig- gered seismicity at The Geysers, California, J. Geo- phys. Res., 101, 733-749. GORDON, D.W., T.J. BENNETT, R.B. HERRMANN and A.M. ROGERS (1970): The south-central Illinois earthquake of November 9, 1968: Macroseismic studies, Bull. Seismol. Soc. Am., 60, 953-971. HANKS, T.C. and A.C. JOHNSTON (1992): Common features of the excitation and propagation of strong ground mo- tion for North-American earthquakes, Bull. Seismol. Soc. Am., 82, 1-23. HILL, D.P., P.A. REASENBERG, A. MICHAEL, W.J. ARABAZ, G. BEROZA, D. BRUNMBAUGH, J.N. BRUNE, R. CASTRO, S. DAVIS, D. DEPOLO, W.L. ELLSWORTH, J. GOMBERG, S. HARMSEN, L. HOUSE, S.M. JACKSON, M.J.S. JOHNSTON, L. JONES, R. KELLER, S. MALONE, L. MUNGUIA, S. NA- VA, J.C. PECHMANN, A. SANFORD, R.W. SIMPSON, R.B. SMITH, M. STARK, M. STICKNEY, A. VIDAL, S. WALTER, V. WONG and J. ZOLLWEG (1992): Seismicity remotely triggered by the magnitude 7.3 Landers, California, earthquake, Science, 260, 1617-1623. HOPPER, M.G., S.T. ALGERMISSEN and E.E. COBROVOLNY (1983): Estimation of earthquake effects associated with a great earthquake in the New Madrid seismic zone, U.S. Geol. Surv. Open File Rep. 83-179, pp. 94. HOUGH, S.E. (1996): Observational constraints on earth- quake source scaling: Understanding the limits in reso- lution, Tectonophysics, 261, 83-96. HOUGH, S.E. (2001): Triggered earthquakes and the 1811- 1812 New Madrid, central US earthquake sequence, Bull. Seismol. Soc. Am., 91, 1574-1581. HOUGH, S.E. and R. BILHAM (2003). A shock heard ‘round the world, Natural History, February. HOUGH, S.E. and H. KANAMORI (2002): Source properties of earthquakes near the Salton Sea triggered by the 10/16/1999 M 7.1 Hector Mine, California earthquake, Bull. Seismol. Soc. Am., 92, 1281-1289. HOUGH, S.E. and S. MARTIN (2002): Magnitude estimates of two large aftershocks of the 16 December 1811 New Madrid earthquake, Bull. Seismol. Soc. Am., 92, 3259- 3268. HOUGH, S.E., J.G. ARMBRUSTER, L. SEEBER and J.F. HOUGH (2000): On the Modified Mercalli Intensities and Mag- nitudes of the 1811-1812 New Madrid, Central United States earthquakes, J. Geophys. Res., 105, 23, 839-23, 864. JOHNSTON, A.C. (1996): Seismic moment assessment of earthquakes in stable continental regions III, New Madrid 1811-1812, Charleston 1886, and Lisbon 1755, Geophys. J. Int., 126, 314-344. JOHNSTON, A.C. and E.S. SCHWEIG (1996): The enigma of the New Madrid earthquakes of 1811-1812, Annu. Rev. Earth Planet. Sci. Lett., 24, 339-384. KELSON, K.I., G.D. SIMPSON, R.B. VANARSDALE, C.C. HA- RADEN and W.R. LETTIS (1992): Multiple Holocene earthquakes along the Reelfoot fault, central New Madrid seismic zone, J. Geophys. Res., 101, 6151-6170. MCMURTRIE, H. (1819): Sketches of Louisville and its envi- rons; including, among a great miscellaneous matter, a Florula Louisvillensis; or, a catalogue of nearly 400 Genera and 600 Species of plants, that grow in the vicinty of the town, exhibiting their generic, specific, and vulgar English names, S. Penn, Jun. Main-street, Louisville. MILNE, J. (1898): Seismology (Kegan Paul, Trench, Trubn- er and Co. Ltd., London), 1st edition, pp. 320. MITCHILL, S.L. (1815): A detailed narrative of the earth- quakes which occurred on the 16th day of December, 1811, Trans. Lit. Philos. Soc. NY, 1, 281-307. MUELLER, K., S.E. HOUGH and R. BILHAM (2004): Analysing the 1811-1812 New Madrid earthquakes with recent instrumentally recorded aftershocks, Na- ture, 429, 284-288. NUTTLI, O.W. (1973): The Mississippi Valley earthquakes of 1811 and 1812: Intensities, ground motion, and magnitudes, Bull. Seismol. Soc. Am., 63, 227-248. ODUM, J.K., W.J. STEPHENSON, and K.M. SHEDLOCK (1998). Near-surface structural model for deformation associ- ated with the February 7, 1812 New Madrid, Missouri, earthquake, Geol. Soc. Am. Bull., 110, 149-162. PENICK, J.L., Jr. (1981): The New Madrid Earthquakes (Univ. of Mo. Press, Columbia), rev. edition. ROZIER, F.A. (1890): Rozier’s History of the Early Settle- ment of the Mississippi Valley (G.A. Pierrot and Son, Saint Louis). RUSS, D.P. (1982): Style and significance of surface defor- mation in the vicinità of New Madrid, Missouri, in In- vestigations of the New Madrid, Missouri, earthquake region, edited by F.A. MCKEOWN, L.C. PAKISER, U.S. Geol. Surv. Prof. Pap. 1236, 95-114. SEEBER, L., and J.G. ARMBRUSTER (1987): The 1886-1889 aftershocks of the Charleston, South Carolina, earth- quake: a widespread burst of seismicity, J. Geophys. Res., 92, 2663-2696. 537 Scientific overview and historical context of the 1811-1812 New Madrid earthquake sequence SINGH, S.K., J. LERMO, T. DOMINGUEZ, M. ORDAZ, J.M. ES- PINOSA, E. MENA and R. QUASS (1988): The Mexico earthquake of September 19, 1985 – A study of amplifi- cation of seismic waves in the Valley of Mexico with re- spect to a hill zone site, Earthquake Spectra, 4, 653-673. STOVER, C.W. and J.L. COFFMAN (1993): Seismicity of the United States, 1568-1989, U.S. Geol. Surv. Prof. Pap. 1527 (revised). STOVER, C.W. and C.A. VON HAKE (Editors) (1982): Unit- ed States Earthquakes, 1980, (US Dep. of Inter., Gold- en, Colo.). STREET, R. (1982): A contribution to the documentation of the 1811-1812 Mississippi Valley earthquake sequence, Earthquake Notes, 53, 39-52. STREET, R. (1984): The historical seismicity of the central United States: 1811-1928, Final Report, Contract 14- 08-0001-21251, Appendix A, U.S. Geol. Surv., Wash- ington, DC, pp. 316. TAYLOR, K., W. STAUDER and R. HERRMANN (1991): A com- prehensive unified data set for the New Madrid, seis- mic array, Seismol. Res. Lett., 62, 187. TUTTLE, M.P. and E.S. SCHWEIG (1996): Archaeological and pedological evidence for large prehistoric earthquakes in the New Madrid seismic zone, Central United States, Geology, 23, 253-256. TUTTLE, M.P., E.S. SCHWEIG, J.D. SIMS, R.H. CAFFERTY, L.W. WOLF and M.L. HAYNES (2002): The earthquake potential of the New Madrid seismic zone, Bull. Seis- mol. Soc. Am., 92, 2080-2089. WELLS, D.L. and K.J. COPPERSMITH (1994): New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement, Bull. Seismol. Soc. Am., 84, 974-1002. WU, P. and P. JOHNSTON (2000): Can deglaciation trigger earthquakes in North America?, Geophys. Res. Lett., 27, 1323-1326.