The San Andreas and San Jacinto faults are the primary plate-boundary structures in southern California and present a large earthquake hazard for the region. They approach each other in the Cajon Pass area between the San Gabriel and San Bernardino Mountains, where the northern end of the San Jacinto fault forms a 2-km-wide releasing step with the San Andreas fault. In this study, we used paleoseismic data from sites on the San Jacinto and San Andreas faults near their juncture to evaluate spatial and temporal patterns of surface rupture between these major structures of the North American–Pacific transform plate boundary. We present a new 3700 yr paleoseismic record from the northern San Jacinto fault at Mystic Lake, where trench excavations exposed evidence for at least 16 surface ruptures. A sedimentary gap in our trench exposures separates three ruptures in the oldest part of the record from 13 ruptures during the past 2000 yr. For the past 2000 yr, the mean recurrence times varied from 86 to 312 yr, with a mean recurrence interval of 160 yr. This rate of surface rupture is roughly equal to that of the southern San Andreas fault south of the Cajon Pass juncture, but it is half that of the San Andreas fault north of the juncture, indicating that coseismic strain on the San Andreas fault is split between the southern San Andreas and San Jacinto faults south of Cajon Pass. Comparison of the past 2000 yr of the Mystic Lake record to similar paleoseismic records from nearby sections of the San Andreas fault suggests that: (1) the current open interval on these two faults in the study area is longer than their average recurrence intervals, but that similar intervals of quiescence have occurred in the past 2000 yr; (2) the San Andreas and San Jacinto faults have probably ruptured together multiple times in the past 2000 yr; and (3) a joint rupture of the San Jacinto fault with the Mojave section of the San Andreas fault may be a more likely source of a major earthquake in southern California than a rupture on the San Andreas fault from the Mojave segment to the southern end of the fault.

The interaction between major faults in the San Andreas fault system is a primary concern for earthquake hazards in California, and it is crucial to understanding the dynamics of the plate boundary. Continental plate boundaries typically consist of multiple faults that accommodate the relative motion between lithospheric plates. Individual faults in these plate-boundary systems can link to each other via steps or splays, and the size and lateral extent of earthquake ruptures depend on whether or not ruptures can propagate through these junctions. It is therefore critical to understand how neighboring faults interact in order to understand, and forecast, earthquake behavior within a fault system. The relative abundance of paleoseismic data from faults in the San Andreas fault system in California presents an opportunity to study the degree and nature of interaction between separate faults on time scales of hundreds to thousands of years. In this paper, we present new paleoseismic data from the northern San Jacinto fault in southern California that we compare with similar data from the adjacent San Andreas fault to evaluate the 2000 yr history and spatial pattern of earthquake ruptures on these two faults near their junction.

Geologic and geodetic data show that along the North American–Pacific plate boundary in central California, the San Andreas fault accommodates ∼70% of the total displacement across the boundary (e.g., Thatcher, 1979; Sieh and Jahns, 1984; Prescott et al., 2001). In southern California, however, the ∼35 mm/yr late Quaternary slip rate on the Mojave section of the San Andreas fault (e.g., UCERF3 appendix B inDawson and Weldon, 2013) decreases to the southeast though the Cajon Pass to only 7–16 mm/yr along the San Bernardino section of the San Andreas fault (Fig. 1; McGill et al., 2015, 2013; Spinler et al., 2010). The San Jacinto fault diverges southward from the San Andreas fault in the Cajon Pass area and makes up for most of the missing slip rate, with a slip rate of 13–18 mm/yr (McGill et al., 2015; Onderdonk et al., 2015; Blisniuk et al., 2013). Displacement must be transferred between the San Jacinto fault and San Andreas fault across a stepover between the two faults at the north end of the San Jacinto fault in the Cajon Pass (i.e., Morton and Matti, 1993; Machette et al., 2004; McGill et al., 2013), but it is not known whether this occurs only as interseismic strain accommodation, or if the two faults have ruptured together. Anderson et al. (2003) used both static and dynamic stress modeling to infer that a rupture on the northern San Jacinto fault will increase Coulomb stress on the Mojave segment of the San Andreas fault and possibly lead to simultaneous rupture. They also modeled a southward-propagating rupture on the Mojave section of the San Andreas fault and showed that resultant rupture on the San Jacinto fault was possible. Other studies (Sanders, 1993; Pollitz and Sacks, 1992) have suggested that historic earthquakes on the San Jacinto fault have been triggered by stress changes due to the A.D. 1857 event on the Mojave section of the San Andreas fault. Grant Ludwig et al. (2015) showed that the presence of precariously balanced rocks near the juncture of the San Jacinto fault and San Andreas fault is more consistent with the lower ground motions expected from joint rupture of the San Jacinto fault and Mojave section of the San Andreas fault than from a rupture passing through the area on the San Andreas fault. This led them to hypothesize that most ruptures in the area initiate at, terminate at, or pass through the stepover between the San Jacinto fault and San Andreas fault. Dynamic rupture modeling of the San Andreas fault–San Jacinto fault junction by Lozos (2016) supported this idea and showed that historic and paleoseismic evidence for the A.D. 1812 earthquake on the San Andreas fault is best explained by joint rupture of the San Jacinto fault and San Andreas fault during that event. Here, we show that paleoseismic data from the past 2000 yr strongly support the hypothesis of joint rupture and suggest that some ruptures during this time period have passed through the step between the San Andreas fault and San Jacinto fault. We propose that this scenario is more likely than a similarly large rupture involving the entire southern San Andreas fault, based on the fault zone geometries and earthquake histories of the two faults.

Our new paleoseismic data were collected from fault trenches excavated across the Claremont strand of the northern San Jacinto fault at Mystic Lake (Fig. 1). The Mystic Lake paleoseismic site is a 400-m-wide by 600-m-long sag that developed due to subsidence and periodic ponding of water against scarps that formed during surface ruptures (Fig. 2). Initial trenching at the site (trenches T1, T2, T3, T4) was done in 2009 to precisely locate the primary active fault strand (fault A at the southwestern side of the sag), and to evaluate the relative activity of three additional fault strands (B, C, and D) that form scarps bounding the northeastern side of the sag. Three additional trenches (T5, T6, T7) were excavated in 2010 to a depth of 1.5 m across fault A to provide additional exposures and identify the best stratigraphic location along the primary active fault strand. These shallow trenches exposed evidence for seven earthquakes since A.D. 500 (Onderdonk et al., 2013), and cone penetrometer testing across the sag documented the progressive growth of the sag over the past 7000 yr (Marliyani et al., 2013). Here, we present and discuss data from additional deeper trenches that were excavated in the same location as trench 6 (Fig. 2), where the stratigraphy was preserved the best along the fault. We excavated a 4-m-deep trench in 2012 (T8), and a 5.5-m-deep trench in 2013 (T9). Both trenches were benched with 1.5–1.75-m-tall bench walls (Fig. 3). The trench exposures allowed us to significantly refine the timing of the previously documented earthquakes with radiocarbon dating of additional samples, and extend the paleoseismic record farther back in time to ca. 1700 B.C.

Trench walls were scraped and cleaned by hand before a string grid was set up to divide the exposure into 1-m-wide by 0.5-m-tall panels. Stratigraphic units were identified, named, traced along the trench, and correlated with known units exposed in previous trenches. All trench panels were photographed and then digitally assembled into mosaics. Stratigraphic and structural relationships were then described and drawn onto the printed photo mosaics in the field. Datable material was collected and cataloged, and the locations were marked on the trench walls. Over 700 samples were collected for radiocarbon dating, and 118 of these were dated at the Keck Carbon Cycle Accelerator Mass Spectrometry Program at the University of California, Irvine (Table 1). Several models of the stratigraphic unit ages were developed and evaluated based on the dated samples and their relative stratigraphic position. Most of the samples collected were detrital charcoal, so there was an unknown amount of time between the formation of the charcoal during a brush fire and deposition of the charcoal into the stratigraphic section. Consequently, a stratigraphic layer is equal to or younger than the charcoal it contains. We therefore assumed that the youngest charcoal age from each layer provides the closest approximation of the age of that layer, and that radiocarbon samples with ages that are older than samples in underlying units do not represent the true age of the unit that contains them. Using these assumptions, we eliminated almost half of the dated samples, and our stratigraphic model is based on 70 samples (Table 1). An event history for the site was then determined using OxCal software version 4.3 (Bronk Ramsey, 2009), which calculates the probability density functions of radiocarbon sample ages and the event ages based on the dendrochronologically calibrated radiocarbon curve of Reimer et al. (2013).

The trenches exposed evidence for repeated surface ruptures along fault A (Fig. 2), which consisted of a main fault zone that showed northeast-side-down displacement, and a secondary fault zone 20 m to the southwest in the uplifted side (Fig. 4). Evidence of surface ruptures included upward terminations of faults, filled fissures, folding of strata across the main fault zone, and onlap of stratigraphic layers and angular unconformities associated with vertical movement across the main fault zone (Figs. 48). The earthquake record is most complete in the main fault zone at the northeast end of the trenches. There, the repeated subsidence of the sag during earthquakes creates accommodation space for additional sediment to be deposited, enabling us to distinguish between one earthquake rupture and the next. The stratigraphic level that represents the ground surface at the time of an earthquake and marks the upper limit of faulting at the site for that individual earthquake is defined as the “earthquake horizon.” Almost all of the earthquake horizons at the Mystic Lake site occur at the top of fine-grained, dark, organic layers. We interpret these layers to be paleosols that developed at the surface during periods of depositional quiescence and soil formation in between earthquakes. On top of the earthquake horizons, there are lighter-colored clay or silty-clay deposits that are less tilted than the underlying organic-rich layers, and that thin as they approach and cross the main fault zone (Figs. 48). We interpret clay-rich layers to represent filling of the sag after it experienced subsidence during an earthquake. The stratigraphic relationships show that each successive rupture faulted and/or folded the soil that existed at the surface at the time of the earthquake, and then the sag slowly filled with sediment. A soil later developed at the surface on the new sediment until the next earthquake caused renewed subsidence.

We documented evidence for 16 surface ruptures over the past 3700 yr. Paleoseismic event evidence collected from the five trenches across fault A is shown in Figures 4 through 11 and summarized in Table 2. Evidence of the three oldest events (between 1700 B.C. and 1000 B.C.) was only found along fault strands within the secondary fault zone (Figs. 911), where older sediments have been elevated by southwest-side-up slip along the main fault zone of fault A. Trenches 8 and 9 were not deep enough to expose sediments of this age within the main fault zone, so we have no record of events between ∼1000 B.C. and 100 B.C. (Fig. 12). We do believe we have a complete record, however, for the past 2000 yr, during which there were 12 or 13 ground-rupturing earthquakes. We generated probability density functions for the open intervals between each earthquake (we did not include the current open interval) and used the means of these individual distributions to estimate recurrence interval and the coefficient of variation. The mean recurrence time between earthquakes varied from 86 to 312 yr, with a mean recurrence interval for all 13 earthquakes of 160 yr and a standard deviation of 76. This results in a coefficient of variation of 0.48 for the past 2000 yr.

Recurrence Intervals and the Current Open Interval

The Mystic Lake paleoseismic record (Fig. 12) shows no strong clustering of earthquakes during the past 2000 yr, and the 0.48 coefficient of variation suggests fairly regular ruptures. The most recent ground-rupturing earthquake occurred in the early 1800s. A rupture at the same time is also recorded at both of the two paleoseismic sites that have been studied farther to the northwest on the northern San Jacinto fault: the Quincy site (Onderdonk et al., 2015) and the Colton site (Kendrick and Fumal, 2005). This rupture may be one of two large historic earthquakes that caused damage at several Spanish missions in southern California on 22 November 1800 and 8 December 1812 (Toppozada et al., 2002). The 22 November 1800 earthquake, however, is interpreted to have occurred farther south along the San Jacinto fault, since it mainly caused damage at the San Juan Capistrano and San Diego missions, and it is preserved in paleoseismic records along the central San Jacinto fault (Salisbury et al., 2012; Rockwell et al., 2015). Grant Ludwig et al. (2015) hypothesized that the 8 December 1812 earthquake that ruptured the Mojave section of the San Andreas fault (Jacoby et al., 1988; Toppozada et al., 2002) also ruptured the San Jacinto fault, based on the low levels of ground motion required to preserve precariously balanced rocks in the area near the juncture of the two faults. Based on fitting dynamic rupture models to historic and paleoseismic records of this earthquake in the region, Lozos (2016) further hypothesized that this earthquake began on the northern San Jacinto fault and propagated onto the San Andreas fault. In light of these studies, we believe the most recent event at Mystic Lake was the 8 December 1812 earthquake. An alternative interpretation is that the most recent event on the northern San Jacinto fault was not substantial enough to be documented in historic records, but this seems unlikely because of the large size of the earthquake that should have occurred in a rupture that extended most of the length of the fault. The open interval that has occurred since the most recent earthquake is at least 167 yr, and possibly as long as 217 yr (206 yr if the earthquake occurred in A.D. 1812). This is equal to or longer than the average recurrence interval (160 ± 76 yr) for the northern San Jacinto fault during the past 2000 yr (Fig. 12), suggesting the fault may be near failure. There are three intervals between earthquakes, however, that may have been just as long, or longer, as the current open interval, given the uncertainties in earthquake ages. For example, the open interval between events 5 and 6 could have been as long as 400 yr if event 6 occurred at the oldest end of its possible time range and event 5 occurred at the young end of its time range. Open intervals between events 6 and 7, and between events 9 and 10 could have been as long as 250 and 300 yr, respectively.

Similar relationships exist between the average recurrence interval and the present open interval for sites on the southern San Andreas fault (Fig. 13). At the Wrightwood paleoseismic site, on the Mojave section of the San Andreas fault, the most recent earthquake was in A.D. 1857 (160 yr ago), but the recurrence interval for the past 2000 yr is only 95 ± 5 yr (Scharer et al., 2010; Field et al., 2013). On the San Bernardino section of the San Andreas fault, the most recent earthquake was in A.D. 1812 (206 yr ago), while the recurrence interval is only 147 ± 14 yr at the Pitman Canyon site, and 173 ± 8 yr at the Burro Flats site (Field et al., 2013). However, like the Mystic Lake record, each one of these sites has open intervals in their prehistoric records that may be equally as long as the current open interval, given the uncertainties in event ages. The same is true when looking at the San Jacinto and San Andreas faults combined; there appears to be an unusual lack of ruptures in the area during the past 150 yr, but there are multiple time spans in the past 1200 yr where 150 yr intervals with no earthquakes on either fault are possible within the limits of dating uncertainties (for example, A.D. 900 to A.D. 1050, A.D. 1080 to A.D. 1240, or A.D. 1520 to A.D. 1680). Therefore, it appears that the current open interval may not be unusual and does not indicate some fundamental change in the fault system behavior.

Rupture across Stepovers

The northern San Jacinto fault (Claremont strand) forms a 2-km-wide stepover with the central San Jacinto fault (Clark strand), with ∼24 km of overlap between the two faults (Fig. 1). The Mystic Lake site is located on the northern San Jacinto fault at the north end of this stepover, making it a good location at which to evaluate the possibility of rupture across the step. Both the Mystic Lake site on the northern San Jacinto fault and the Hog Lake site on the central San Jacinto fault (Rockwell et al., 2015) record at least 12 surface ruptures in the past 2000 yr (Fig. 13), but visual comparison of the two records shows no clear patterns of coincidence between ruptures on the two fault strands. Relative quiet on one fault strand during some time periods while the other is more active suggests these fault strands are not rupturing together or triggering each other most of the time. There are, however, three events for which the age ranges are almost identical at the two sites (ML3, ML6, and ML11), which tempts the interpretation that these events may have ruptured both fault strands. However, uncertainties in the event dates (typically tens to hundreds of years) make it impossible to correlate events from two different paleoseismic sites. Statistical analysis of overlapping probability density functions for earthquakes does not provide any benefit to correlation confidence, since even fully overlapping probability density functions only provide a range of time during which two completely unrelated earthquakes could have occurred decades or even years apart. For this reason, additional constraints, such as event size and/or rupture length, are needed to evaluate the probability of event correlation between two sites (e.g., Biasi and Weldon, 2009). For the rupture histories considered here, measurements of slip-per-earthquake on the northern San Jacinto fault provide information about the average earthquake size that supports the idea of occasional joint rupture of fault strands separated by steps at the north and south ends of the Claremont strand. The age of offset streams and timing of earthquakes at the Quincy site (Fig. 1) indicate that the average slip during ruptures in the past 11 or 12 earthquakes was 2.5 m, and greater than 3 m for the last three earthquakes (Onderdonk et al., 2015). If we assume that slip measurements at the site are representative of the average surface slip along the rupture length of these earthquakes, then the observed value of 2.5 m or more of slip in these events suggests rupture lengths of ∼110 km or more (Wells and Coppersmith, 1994), which is longer than the 75 km total length of the Claremont strand. The earthquakes that were larger than the average, especially one or more of the last three, likely extended beyond the length of the fault and continued through one of the steps at either end of the Claremont strand. The strongest evidence for this exists for event 3 at Mystic Lake, which we interpret to have been a very large earthquake based on the large amount of tilting observed in the Mystic Lake trenches at the stratigraphic level of event 3, and evidence of 3.1 m to 3.6 m of lateral slip during this event at the Quincy site (Onderdonk et al., 2013, 2015).

Cajon Pass “Earthquake Router”

During the past 1000 yr, there were nine earthquakes recorded on the Mojave section of the San Andreas fault at Wrightwood, but only six recorded at paleoseismic sites on the San Bernardino section of the San Andreas fault to the southeast (Fig. 13). An additional nine earthquakes were recorded at Wrightwood during the preceding millennium, while only three are recorded in the record from Burro Flats. No data exist for this earlier time period from the Pitman Canyon site, but unless there was a big change in recurrence interval around A.D. 1000, we can assume that there were a larger number of earthquakes on the San Andreas fault northwest of Cajon Pass than to the southeast for the past 2000 yr. This suggests that some ruptures on the Mojave section of the San Andreas fault stop in Cajon Pass, as did the Fort Tejon earthquake in 1857, with Cajon Pass acting as an “earthquake gate” (term from Oskin et al., 2015), which can inhibit or facilitate a through-going rupture. However, the structure and tectonic geomorphology of the San Andreas fault through Cajon Pass are relatively continuous and show no obvious indication of a barrier to rupture propagation at the surface (e.g., Sedki, 2013; USGS Q-faults database: Machette et al., 2004) or in three-dimensional fault models (Plesch et al., 2007). If large ruptures on the Mojave segment of the San Andreas fault are occasionally starting or stopping in Cajon Pass, work is needed to identify the type of geologic, structural, or stress conditions that are responsible for this.

An alternative to ruptures stopping in Cajon Pass is the possibility of rupture being directed onto the adjacent San Jacinto fault, and several lines of evidence support this interpretation. First, the differences between the number of earthquakes recorded on the San Andreas fault north and south of Cajon Pass, and the recurrence intervals for the three fault sections show that the Mojave section of the San Andreas fault ruptures almost twice as frequently as the San Bernardino section of the San Andreas fault or the San Jacinto fault (Fig. 13). Second, the spatial distributions of slip rates on the San Andreas fault and San Jacinto fault indicate that displacement is being transferred between the two faults. The ∼34 mm/yr slip rate of the Mojave section of the San Andreas fault (e.g., Dawson and Weldon, 2013) is divided between the San Bernardino section of the San Andreas fault (7–15.7 mm/yr; McGill et al., 2013) and the northern San Jacinto fault (12.8–18.3 mm/yr; Onderdonk et al., 2015), with 2–5 mm/yr possibly distributed to the reverse faults at the margins of the San Gabriel and San Bernardino Mountains (McGill et al., 2013, 2015). A correlative distribution of coseismic slip on the Mojave section of the San Andreas fault between these two fault sections to the southeast and south would be a simple way of accommodating the division of accumulated slip from the Mojave segment of the San Andreas fault southward. Third, the average slip-per-event on the northern San Jacinto fault indicates that some ruptures on the fault may have extended beyond one (or both) of the stepovers at the ends of the fault. The stepover between the northern end of the San Jacinto fault and the San Andreas fault in Cajon Pass is only 1.6 km wide (Fig. 1), which is narrow enough to allow rupture propagation between the two faults, based on observations of historic ruptures across stepovers (Wesnousky, 2006). These characteristics of the three fault sections lead us to infer that some of the San Andreas fault ruptures that are “missing” from the San Bernardino segment of the San Andreas fault paleoseismic sites are present in the Mystic Lake record and represent joint rupture of the Mojave segment of the San Andreas fault and the northern San Jacinto fault. We note that if the Burro Flats paleoseismic record is complete (or mostly complete) for the past 2000 yr, then joint rupture of the San Andreas fault and San Jacinto fault may have been more common between A.D. 0 and A.D. 1000 than during the past 1000 yr. Activity at the Wrightwood site has been relatively constant during the past 2000 yr, with nine earthquakes recorded in each of the past two millennia (Fig. 13). However, the San Bernardino section of the San Andreas fault appears to have been more active between A.D. 1000 and the present, with six earthquakes recorded at Burro Flats, compared to only three between A.D. 0 and A.D. 1000. The Mystic Lake site shows an opposite pattern, with five earthquakes between A.D. 1000 and the present, compared to eight earthquakes between A.D. 0 and A.D. 1000.

These observations of San Jacinto fault and San Andreas fault kinematics support previous interpretations based on modeling and inferred strong ground motion patterns that inferred that ruptures can propagate from one fault to the other in the Cajon Pass area (Anderson et al., 2003; Grant Ludwig et al., 2015; Lozos, 2016). Grant Ludwig et al. (2015) also suggested that ruptures that did not jump across the step most likely started or stopped in Cajon Pass, and that through-going rupture on the San Andreas fault is rare. Our comparison of paleoseismic data from the past 2000 yr presented here cannot directly test the idea that through-going ruptures on the San Andreas fault are rare. However, the paleoseismic data do provide another line of evidence that supports the idea that the Cajon Pass area acts as an “earthquake router,” where ruptures have an end point in the Cajon Pass area, propagate across the step from one fault to another (or to one of the reverse faults of the Transverse Ranges), or continue through on the San Andreas fault.

Implications for a Large Southern San Andreas Earthquake

The possibility of joint rupture of the northern San Jacinto fault and the Mojave segment of the San Andreas fault has implications for probabilities of major earthquakes in the southern California fault system. To explore the effects on the population of southern California, Jones et al. (2008) modeled a large rupture on the southern San Andreas fault, nucleating on the Coachella Valley section of the San Andreas fault and propagating northward through the San Gorgonio Pass onto the Mojave segment of the San Andreas fault (called “The ShakeOut Scenario”). We propose that a more likely scenario for such a large event is the joint rupture of the San Jacinto fault and the Mojave section of the San Andreas fault, because this scenario involves a rupture path that has less structural complexity than the southern San Andreas fault. The southern San Andreas fault includes a structurally complex restraining bend in the San Gorgonio Pass, composed of an oblique-thrust system with strike-slip tear faults that link the San Andreas fault in Coachella Valley to the San Bernardino area (e.g., Morton and Matti, 1993; Yule and Sieh, 2003). Conversely, the San Jacinto fault is straighter and contains only narrow dilational steps that modeling suggests are mechanically easier for ruptures to pass through than similar-sized restraining bends or steps (Oglesby, 2005). In a large southern San Andreas fault rupture (e.g., Bombay Beach to the Mojave segment of the San Andreas fault, as modeled by Jones et al., 2008), average displacement would be expected to be 4 m or more, and it is more likely to be near the maximum in the middle of the rupture along the San Bernardino segment. Average slip-per-event on the San Bernardino segment of the San Andreas fault during the past 1000 yr, calculated from the number of earthquakes at Pitman Canyon and Burro Flats and the latest Pleistocene slip rate (McGill et al., 2013), is only 1.2–2.7 m/event. This is not as high as would be expected if a full southern San Andreas fault rupture had occurred during this time period, and it is lower than the average slip-per-event values for the past 1000 yr on the northern San Jacinto fault (2.5–3.5 m/event: Onderdonk et al., 2015) and the Mojave segment of the San Andreas fault (3.9 m/event: calculated here from the number of earthquakes recorded at the Wrightwood site and the slip rate). This suggests a “ShakeOut Scenario” event has not occurred in the past 1000 yr. Heermance and Yule (2017) showed that the thrust system in the San Gorgonio Pass has experienced offsets of 4–8 m in single events that may involve the San Andreas fault on both sides of the pass, but they noted that these are infrequent events (once every 1000–1500 yr). The long intervals between ruptures in the San Gorgonio Pass area also indicate that some of the hypothesized correlations between earthquakes recorded at paleoseismic sites on the San Andreas fault in Coachella Valley and the San Bernardino area (e.g., Philibosian et al., 2011; Rockwell et al., 2016) are incorrect and that the “ShakeOut Scenario” is not as common as these correlations would suggest. This is supported by observations that precariously balanced rocks exist in the San Bernardino Mountains near the San Andreas fault, where strong ground motion would be expected from a large San Andreas fault rupture that extended from Coachella Valley through the Cajon Pass (Grant Ludwig et al., 2015). Such a rupture would have toppled the precariously balanced rocks, indicating that a large rupture on the San Andreas fault through this area has not occurred in several thousand years (Grant Ludwig et al., 2015). The available paleoseismic data, average slip-per-event data, distribution of precariously balanced rocks, and the structural geometry of the two fault zones summarized here suggest to us that very large earthquakes in the southern San Andreas fault system may be more likely to occur from joint rupture of the Mojave segment of the San Andreas fault and the San Jacinto fault, rather than a San Andreas fault–only rupture from the Mojave Desert to Coachella Valley.

We thank the following former students and Southern California Earthquake Center (SCEC) interns for assistance in the field: Paul Alessio, Matt Cole, Ryan Danielson, Eric Gordon, Eric Haaker, Julian Lozos, Rainer Luptowitz, Gayatri Marliyani, Walter Nelson, Elizabeth Niespolo, Matthew Warbritton, and Neta Wechsler. We also thank Scott Sewell of the San Jacinto Wildlife Area for his assistance with this project and allowing us to access the Mystic Lake site. We thank Doug Yule, Glen Biasi, and three anonymous reviewers for their excellent reviews of earlier versions of this paper. This work was funded by U.S. Geological Survey (USGS) Earthquake Hazard Program grants (G11AP20136, G11AP20138, G11AP20142), and several grants from the SCEC. SCEC is funded by National Science Foundation Cooperative Agreement EAR-0106924 and USGS Cooperative Agreement 02HQAG0008. This is SCEC contribution 6238. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the U.S. Geological Survey. Mention of trade names or commercial products does not constitute their endorsement by the U.S. Geological Survey.

Science Editor: Raymond M. Russo