- © 2014 by the Seismological Society of America
In southern California, where fast slip rates and sparse vegetation contribute to crisp expression of faults and microtopography, field and high‐resolution topographic data (<1 m/pixel) increasingly are used to investigate the mark left by large earthquakes on the landscape (e.g., Zielke et al., 2010; Zielke et al., 2012; Salisbury, Rockwell, et al., 2012, Madden et al., 2013). These studies measure offset streams or other geomorphic features along a stretch of a fault, analyze the offset values for concentrations or trends along strike, and infer that the common magnitudes reflect successive surface‐rupturing earthquakes along that fault section. Wallace (1968) introduced the use of such offsets, and the challenges in interpreting their “unique complex history” with offsets on the Carrizo section of the San Andreas fault; these were more fully mapped by Sieh (1978) and followed by similar field studies along other faults (e.g., Lindvall et al., 1989; McGill and Sieh, 1991). Results from such compilations spurred the development of classic fault behavior models, notably the characteristic earthquake and slip‐patch models, and thus constitute an important component of the long‐standing contrast between magnitude–frequency models (Schwartz and Coppersmith, 1984; Sieh, 1996; Hecker et al., 2013). The proliferation of offset datasets has led earthquake geologists to examine the methods and approaches for measuring these offsets, uncertainties associated with measurement of such features, and quality ranking schemes (Arrowsmith and Rockwell, 2012; Salisbury, Arrowsmith, et al., 2012; Gold et al., 2013; Madden et al., 2013). In light of this, the Southern San Andreas Fault Evaluation (SoSAFE) project at the Southern California Earthquake Center (SCEC) organized a combined field activity and workshop (the “Fieldshop”) to measure offsets, compare techniques, and explore differences in interpretation. A thorough analysis of the measurements from the field activity will be provided separately; this paper discusses the complications presented by such offset measurements using two channels from the San Andreas fault as illustrative cases. We conclude with best approaches for future data collection efforts based on input from the Fieldshop.
For the field activity, 27 participants met at the southern San Andreas fault near Palmdale, California in September 2012 to map and measure geomorphic offsets (Fig. 1). Participants had a mix of experience, but the majority was familiar with the goals and procedures for measuring offset landforms. Participants included graduate students (8), early career scientists (6), and midcareer scientists (8). The remaining participants (5) were less experienced in this type of study (undergraduate geology students or other earth scientists). Groups of three participants were provided with base maps and tape measures and were distributed along two sections of the fault (Fig. 1). We did not instruct the groups which features to measure, or how to measure offsets, anticipating this would reveal the approaches and experience of individual geologists and generate discussion. Participants were instructed to provide assessments of the quality of each offset feature. Over the day, the participants made dozens of measurements on 32 different features ranging in separation from ∼1 to 25 m.
In order to document the qualitative and methodological aspects of the field effort, seven people acted as external observers, recording discussion voiced by each group, methods of measuring, and discrepancies between the groups as they worked along the fault. After presentation of the measurements and a report from the observers, the workshop participants discussed issues relating to (a) recording and classification of measurements, (b) measurement uncertainties, (c) field techniques, (d) external impacts on offset quality, (e) implications of the measurements and uncertainty representations in along‐fault averaging, and (f) limitations in reconstructing landforms. The topics emphasized during the discussion can be categorized as mapping, measuring, and quality ranking.
GEOLOGIC AND GEOMORPHIC MAPPING
The field site was chosen for its concentration of displaced features with a range of offset magnitudes, and because it presented a common challenge of deciphering the contribution of multiple fault traces to offset measurements. Fault trace delineation became a primary concern among the teams; not all participants agreed on the presence of small subsidiary fault strands or which faults should be considered active (Fig. 2a). Many groups debated subtle misalignments in stream channels that may have represented additional offset but were ultimately not measured, because the deflections could not be clearly continued along strike and limited time precluded more careful mapping (Fig. 2b). This led to a common recognition that measurements sometimes provided minimum estimates of slip, because discrete offsets along subsidiary faults that may be apparent soon after the earthquake are subsequently obscured. These issues are not restricted to this type of effort; because of time limitations, distributed deformation is often missed even in immediate post‐earthquake measurements due to absence of continuous piercing lines except in rare cases (Rockwell et al., 2002, Rockwell and Klinger, 2013).
Another common debate within groups was the interpretation of the stream misalignment itself. (We differentiate “deflections,” abrupt changes in stream course caused by the tectonic fabric or geomorphic history of the stream, from “offsets,” which are the product of earthquake(s) separating the upstream and downstream portions of the channel [Wallace, 1968; Weldon et al., 1996].) Specifically, groups voiced uncertainty about the genesis of individual deflections, and how to determine if a deflection was produced by an earthquake (thus an offset) or if the deflection was the result of nontectonic, earth surface processes. These concerns were amplified in areas where the fault was multistranded, leading to discussion about the significance of each potential (but small) jog through a wide‐fault zone, and how (and if) it could be determined that strands were active in recent earthquakes (Fig. 2b). Participants also had lengthy discussions about bends in the channels near the fault and how these bends should be handled when reporting offset measurements (Fig. 2b,c).
All groups debated which feature(s) was appropriate to measure: the thalweg (channel centerline), channel walls or margins, or multiple geomorphic components of a single feature. The best element to measure often depended on the details of the site; some gullies were similar in form upstream and downstream while in other cases, a channel edge provided the most continuous and easily reconstructed feature. All groups deliberated over how to project the upstream and downstream features to the fault, how to handle sinuous channels, and how much feature degradation affected the measurements. Groups typically provided either (a) a thalweg measurement with channel margin half‐width providing the uncertainty or (b) a minimum and maximum range determined by the shape of the landform and the range of its acceptable reconstruction. Participants disagreed about the value of two mapping approaches: measuring offsets of multiple components of a single feature (to provide multiple independent estimates that likely incorporate different uncertainties) or, measuring the best‐preserved feature (to obtain the most accurate estimate, rather than swamping the results with poorly resolved features).
Larger offsets posed additional problems, as smaller preserved features within the offset increase complexity. For example, some groups measured deflections of smaller inset surfaces, assuming they reflected better preservation of a more recent channel form rather than, for example, the deflection produced from connectivity of older landforms that had been moved into the current position (Fig. 3d). Similarly, changes in the channel width at the time of the earthquake instead of the present day channel width (including postoffset geomorphic changes) were vigorously debated. Several groups measured short, very straight, fault parallel reaches of the channel right at the geomorphic trace of the fault as evidence for the most recent offset (Fig. 3b); other groups felt these measurements were not meaningful for tectonic slip, and resulted from the incision history of the drainage. In light of these differences in approach, a common conclusion of participants was that the reconstruction itself (i.e., how the mapper interpreted the preserved topography and how they reconstructed the postoffset changes to the terrain) was an epistemic uncertainty that could effectively determine the uncertainty in the offset (and aleatory uncertainty from the measuring itself; Fig. 3b).
The group discussed the meaning of the measurement uncertainties, and emphasized that the uncertainties typically do not provide a true standard deviation (see Gold et al., 2013 for an approach to this problem), but rather physical or plausible limits on the amount a feature could have been offset. Many agreed that given the uncertainties in feature reconstruction, most offsets should be recorded as a minimum and maximum allowable offset. The range permitted for an offset feature implies that there is an appropriate probability‐distribution function (PDF) for the offset; outside that range, the geologist finds no correlation of features. We discussed the forms of PDFs used in existing studies (uniform/boxcar, triangle, and truncated Gaussians) and generally agreed that a uniform PDF was appropriate to use for most offsets. We note, however, that most measurements assume a simple continuity of the feature across the fault (most often that the gully was straight before it was offset), and to our knowledge no study has considered the effect of this assumption on the resulting offset measurements. Approaches to stacking and correlating measurements along strike were outside the scope of our workshop, although several papers treat this issue quantitatively (McGill and Sieh, 1991; McGill and Rubin, 1999; Gold et al., 2013; Madden et al., 2013; Rockwell and Klinger, 2013).
ASSESSING QUALITY AND RANKING SYSTEMS
Participants were asked to provide two in‐field assessments of offset quality. The first was simply a subjective rank from 1 to 5 of the overall quality of the feature. The second was a bivariate that described (low, medium, or high) the definition of the fault zone and the obliquity of the offset feature with respect to the fault—two important and measurable controls on offset quality (Madden et al., 2013). Some sites received the same quality ranking by all groups while other sites received the entire range of quality rankings. Groups had differing opinions on whether or not to measure low quality or complex features, concerned that a large number of lower quality measurements could impact derivative studies in contradiction to the confidence of the original mapping. Vegetation, competence of the substrate, local geomorphic terrain, number and precise location of fault strands, and distinctiveness or geomorphic activity of the offset feature were discussed as impacts on offsets that could be quantitatively incorporated into future compilations, and which would impact along‐fault reconstructions.
Participants disagreed about whether or not such factors should be individually ranked or lumped into a single‐quality ranking. A particular challenge is to separate those aspects of the quality of the offset and its reconstruction that are accounted for in the measurement uncertainty (ideally the aleatory aspects) from those that control the epistemic interpretation of the offset. All agreed that field notes and documentation of what was measured were fundamental to such work, and careful consideration of how to utilize the quality ranks when such offset measurements are incorporated into larger projects was needed. A standardized template for such metadata collection would be valuable.
Most groups investigated a 4–5 m deflection of an ephemeral channel in the Pearblossom area (Fig. 2). At this site, many groups identified two main traces (Fig. 2a; black arrows) and measured the diversion on the northern trace at the confluence of two small gullies. The observer noted that a few groups did not measure this feature due to time limitations and concern over the relative activity of the fault strands.
The channel bends slightly as it approaches the northern fault trace (Fig. 2b,c). Most groups evaluated the channel diversion by projecting the thalweg of the upstream channel to the fault (location A) and measuring the distance to the confluence (location C). All of the groups’ measurements overlapped within error, with a total range of best estimate values from 4 to 5.2 m (Fig. 2d). Small differences in the results largely derived from slight variations in the projection angle of the upstream gully or the location of the thalweg, resulting in lower and upper limits of the deflection from 3.6 to 6 m across the groups. One group provided a lower limit by allowing that the curve of the upstream channel (between locations A and B, Fig. 2b,c) predated any offset, and thus estimated a lower limit of 2.7 m based on the distance between locations B and C. In a slightly different approach, which interpreted that the bend and straight section were created in successive earthquakes, Zielke et al. (2012) produced two separate measurements for this gully, of 2.6 and 4.7 m, measuring the location from B to C and A to C remotely using LiDAR data.
Fieldshop participants ranked the quality of this feature contradictorily; obliquity values ranged from low to high (which likely reflected their concern over the curve in the gully between A and B); assessment of fault width ranged from moderate to high (narrow), and the general rank covered values of 2–5 (good to very poor). Several groups discussed the deflection downstream (location D in Fig. 2b) as a potential contributor to the recent offset, although no groups measured the deflection.
Following the Fieldshop, we returned to the deflection to consider the bend in the gully more carefully. A small excavation revealed that recent sediments (the upper 20–30 cm) on the low surface just west of the gully at location A (Fig. 2b) were not faulted at the mapped trace of the fault. No faulting was observed in the bedrock thalweg at the bend of the channel (although exposure was limited); faulting of the bedrock was observed in the channel just downstream of the confluence of the two gullies. Given these field relationships, faulting may not have produced the bend in the channel, rather it is a deflection. It is possible, however, that the channel incised into a post‐1857 storm deposit that mantled the fan and covered the fault zone. If the channel, pinned by upstream topography, reincised at the prestorm deposit location, then an offset may be recorded in this feature (for example as shown in Liu‐Zheng et al., 2006). Further excavations are needed to clarify these relationships, but the observation serves to underscore the epistemic uncertainties in reconstructing these features that are not included in the formal uncertainties of the measurement when more detailed fieldwork is not conducted. This small exercise also illustrates the value of field studies and the additional effort such studies require.
A larger, sinuous, deeply incised channel presented in Figure 3 produced several different interpretations by five groups. Here the fault strands are clearer, and most groups identified two main fault strands bounding a broad saddle (Fig. 3a). The observer watched as groups readily identified the southwest fault trace in the gully wall (Fig. 3b), whereas the northeast fault trace was largely inferred by the geomorphology of the saddle east of the gully. Groups measured one or two offsets from this feature. Three groups evaluated the straight section of the incised gully (Fig. 3a,b), and interpreted that this represented a discrete offset from one (or more) earthquakes based on a combination of the fault exposure and the appearance of the straight section in the topographic map. In the field, this feature is more sinuous than it appears in the high‐resolution topography and is obscured by vegetation (Fig. 3b). Consequently, the field measurements for this offset were variable and in some cases did not overlap (Fig. 3d), ranging from 7 to 9.5 m, with uncertainties from ±0.5 to 1 m (one group developed asymmetric uncertainties). In later discussions, many participants felt strongly that straight, fault‐parallel gullies were likely the product of recent, postearthquake geomorphic processes (e.g., headward erosion and incision) rather than a signature of the most recent offset(s), and thus the measurement was not a valid quantification of earthquake displacement. At this channel, for example, the left deflection in the stream along the northeast fault trace provides an example where the geomorphology and tectonics combine to produce a straight section but with an unlikely geometry, which presumably has little to do with fault displacement.
Three groups considered the larger offset expressed by the channel, determining offsets from 16 to 23 m, with symmetric uncertainties that ranged from ±3 to 4.5 m (Fig. 3d). A weakly cemented colluvial cover, particularly well preserved on the downstream reach, was recognized on many of the hillslopes (Fig. 3c). Remnants of the colluvial contact on the downstream channel banks were used by most groups to estimate a channel form that projected to the southwest fault trace; the contribution of the northeast fault trace was not considered (although the high angle between the piercing line and the fault trace minimizes the effect in this case). The upstream piercing line was more ambiguous, but its location was typically informed by a low terrace preserved at the confluence of two channels upstream and a few remnants of the colluvium on the nearby hillslopes.
We later used the B4 LiDAR (Bevis et al., 2005) data to construct profiles across the channel at key locations where the channel form was preserved and identifiable (Fig. 3e). The broad form of the reconstructed channel permits a wide range of possible correlations, from ∼13 to 25 m (Fig. 3e, lower panel, assuming orthogonal projection to the fault); the largest values reflect the consideration that incision predated the colluvial cover, so this feature may have started as a narrow channel. If the modern thalweg, inset about 0.5 m into the broader channel form, is used for the reconstruction (Fig. 3e, upper panel), the value (19 m, ±2) is within the range of our reconstructions, but the uncertainties are significantly smaller, illustrating again how geomorphic reconstructions can control the formal error estimates.
Most of the Fieldshop participants are actively engaged in projects that utilize field and/or high‐resolution topography or other imagery‐derived measurements to understand the recent behavior of faults in southern California and elsewhere, leading to a multifaceted discussion about mapping and measuring geomorphic offsets. The Fieldshop and development of the report (Scharer et al., 2012) has led us to conclude that it would be beneficial to address several aspects of measurement of geomorphic offsets in future field and high‐resolution topography studies:
Field mapping was integral to understanding the offset features and obtaining an accurate measurement. In the field, participants recognized features that were not apparent in the LiDAR‐derived topography and concluded that some features apparent in the LiDAR‐derived maps were not offsets. In‐field sketch maps and annotated field photos provide important documentation that should be provided for each feature and included in a data repository. However, as the confidence in fault location or in a particular reconstruction increases when the along‐strike continuity of the fault geomorphology can be discerned, conducting initial mapping on high‐resolution topographic base maps was considered an important step in fully documenting the fault zone and making field time more efficient. Further, LiDAR datasets, particularly if they have higher shot density, permit more careful measurement of features, especially when paired with the field observations. Pairing the techniques was considered the best approach for developing datasets of geomorphic offsets.
Measurement errors should not be conflated with the quality rating. Whereas the ideal feature would have small errors and high quality, it is possible to have features with small measurement errors but poor quality because the geologist has doubts about the geomorphic reconstruction. It is important that aspects of the offset that can be measured (width of channel, angle between piercing line and fault zone, width of fault zone, distance between feature and fault, amount a bend contributes to a perceived offset, etc.) be documented as quantities and reported so that they can be evaluated for their contribution to the measurement. More interpretive aspects of the reconstruction (confidence that the offset is the product of faulting, pairing of features across the fault, age of geomorphic features) should, as much as possible, be qualified as well.
Most offsets should be recorded simply as a range that includes the measurement uncertainties. The range should capture the extent of the reconstruction (i.e., there is no evidence that the feature can be reconstructed outside of that range). In cases where the terrain permits, it may be appropriate to provide a best estimate, uncertainties, and associated PDF (e.g., triangle, truncated Gaussian) to convey more (and less) likely reconstructions within the range. If this is done, we suggest that authors provide clear guidance on the reasons for choosing a nonuniform PDF and emphasize that use of a PDF does not provide a statistical estimate of the variance.
The overall approach of this type of study, to collect as many measurements as possible and look for populations of offset within the results, yields an inherent tension between collection of observations at as many locations as possible (assuming there will be signal in the noise) versus inclusion of only data that are clearly offsets (rather than deflections) and have good quality ranking. The challenge increases along multistranded sections where offsets may be smaller and less well defined. Categorization of a deflection as an offset can produce two‐sided errors, where measurements can be too large (because the measured distance is due in part or wholly to deflection), or too small (because additional, smaller offsets were not mappable), so both possibilities should be evaluated. Recognizing that (a) preservation of small offsets along a fault can be sparse, thus making each data point of high value; and (b) the uncertainties inherent in the reconstruction and measurement of these features, we recommend that to address epistemic uncertainties, future studies strive to determine whether each offset is the product of tectonic rather than geomorphic activity. Subsurface investigations to show tectonic offset are warranted when deflection or postoffset modification rather than offset could have produced the feature (e.g., Lienkaemper and Strum, 1989; Liu‐Zheng et al., 2006). Other approaches include direct dating of offsets, or development of relative age through geomorphic metrics or modeling to improve the categorization of offsets.
In most settings, very few modern channels create straight, perpendicular piercing lines across the fault (even after 153 years of postearthquake storms at the field area, for example), which weakens the assumption that channels originally crossed the fault with no deflection. The development of methods for recognizing pre‐existing deflections and postoffset modification of the channels are needed, and techniques for incorporating uncertainties resulting from projection of the channel across curved segments should be explained in future studies. Similarly, good evidence for interpretation of straight, fault‐parallel sections of the offset as the most recent slip should be provided if it is considered evidence of the most recent rupture.
It would be valuable for studies to provide guidance on the subsequent use of offset measurements, including such considerations as thresholds of data quality, certainty of the origin of the offset feature, shape of offset PDF, and relative weighting compared to other offsets in the study. This guidance would be valuable to the subsequent use of offset measurements in other projects (e.g., Biasi and Weldon, 2009; Madden et al., 2013).
Measurement of successive displacements along a fault are critical data for estimating magnitudes of past earthquakes; when combined along strike, such data inform on the shape of past slip distributions and thus future earthquake potential of active faults (e.g., Sieh, 1978; Zielke et al., 2012; Salisbury, Rockwell, et al., 2012; Hecker et al., 2013). Critical examination of individual offsets, both in this study and previous (e.g., Lienkaemper and Strum, 1989; Gold et al., 2013) show that these measurements include epistemic and aleatory uncertainties that are often not reported. The key recommendations of this study are that field mapping should be combined with analysis of high‐resolution topographic data and supplemented with subsurface investigations when critical; geometric aspects of the measured features should be quantified and interpreted aspects of the feature should be qualified; and guidance on future use of the dataset should be included. The recommended approaches should lead to improvements in the development of such datasets, and ultimately, more robust understanding of past earthquake ruptures, given the challenges in measuring such offsets.
We are grateful for the exceptional participation by our colleagues during the SCEC 2012 Fieldshop, who worked through a 97° F day to contribute to this effort, and a return visit by many in April 2013, including: Austin Elliott, Beth Haddon, Brian Olson, Colin Amos, David E. Haddad, David Lynch, Frank Sousa, Gayatri Marliyani, Glenn Biasi, Guangfu Shao, James Dolan, James Hollingsworth, Janis Hernandez, Jason De Cristofaro, Jerry Treiman, Justin R. Brown, Ken Hudnut, Kim Blisniuk, Koji Okumura, Lisa Grant Ludwig, Margaret Gooding, Mike Oskin, Mitchell Prante, Nate Onderdonk, Orlando Teran, Peter Gold, Richard Heermance, Sally McGill, Scott Kenyon, Sean Bemis, Ting Lin, Tracy Compton, Tsurue Sato, Wendy Bohon, and Whitney Behr. We also thank Stephen DeLong, Ken Hudnut, and one anonymous reviewer for suggestions that improved the paper. This work would not be possible without permission of the landowners and was enhanced with the B4 LiDAR (Bevis et al., 2005) and TLS data provided by colleagues Tracy Compton, Peter Gold and Eric Cowgill. The base maps were built using processing services provided by the Open Topography Facility with support from the National Science Foundation under NSF Award Numbers 0930731 and 0930643. Internal and external funding from the USGS NEHRP has supported much of the research. The Fieldshop was funded by the Southern California Earthquake Center (SCEC). SCEC is funded by NSF Cooperative Agreement EAR‐0529922 and USGS Cooperative Agreement 07HQAG0008. The SCEC contribution number for this paper is 1802.