We studied a paleoseismic trench excavated in 2017 across the Banning strand of the San Andreas fault and herein provide the first detailed record of ground-breaking earthquakes on this important fault in Southern California. The trench exposed an ~40-m-wide fault zone cutting through alluvial sand, gravel, silt, and clay deposits. We evaluated the paleoseismic record using a new metric that combines event indicator quality and stratigraphic uncertainty. The most recent paleoearthquake occurred between 950 and 730 calibrated years B.P. (cal yr B.P.), potentially contemporaneous with the last rupture of the San Gorgonio Pass fault zone. We interpret five surface-rupturing earthquakes since 3.3–2.5 ka and eight earthquakes since 7.1–5.7 ka. It is possible that additional events have occurred but were not recognized, especially in the deeper (older) section of the stratigraphy, which was not fully exposed across the fault zone. We calculated an average recurrence interval of 380–640 yr based on four complete earthquake cycles between earthquakes 1 and 5. The average recurrence interval is thus slightly less than the elapsed time since the most recent event on the Banning strand. The average recurrence interval on the Banning strand is thus intermediate between longer intervals published for the San Gorgonio Pass fault zone (~1600 yr) and shorter intervals on both the Mission Creek strand of the San Andreas fault (~215 yr) and the Coachella section (~125 yr) of the San Andreas fault.

The San Andreas fault is the longest fault (~1300 km) and has the highest slip rate of all faults in California. The northern section of the fault ruptured in 1906, and the south-central section ruptured in 1857. Only the southernmost section of the San Andreas fault has not ruptured during the historical record.

The Banning strand of the San Andreas fault is located within a complex portion of the southernmost section of the San Andreas fault zone (Fig. 1). Within the San Gorgonio Pass region, the fault bends to a west-northwest orientation, resulting in a complex zone of transpressional faults (Yule and Sieh, 2003). In addition, faults diverge northward from the southernmost section of the San Andreas fault toward the Eastern California shear zone (Fig. 1). This geometry triggers many questions about the southernmost section of the San Andreas fault system, such as: (1) How is plate-boundary slip accommodated within this region? (2) How have earthquake ruptures in this region made use of the various fault strands within this complex region?

The geometry of the southernmost San Andreas fault becomes more complex from southeast to northwest; the Coachella section is a single strand, but it splits as it approaches the San Bernardino Mountains (Fig. 1). Near Indio, California, the Banning strand diverges from the Mission Creek strand, and farther to the northwest, the Garnet Hill strand deviates from the Banning strand. In the San Gorgonio Pass region, the southernmost San Andreas fault comprises an intricate network of right-lateral, oblique thrust and reverse faults known as the San Gorgonio Pass fault zone (Allen, 1957; Matti et al., 1985; Matti and Morton, 1993; Yule and Sieh, 2003), which connects in turn with the San Bernardino strand of the southernmost San Andreas fault. None of the strands southeast of the San Bernardino section has ruptured in a major earthquake during the historical period, with the exception of the 1986 North Palm Springs earthquake of Mw 6.0, which produced minor fracturing, as discussed below. The Mill Creek strand, which diverges from the Mission Creek strand within the San Gorgonio Pass region, also connects to the San Bernardino strand, but its recency of motion is less clear (cf. Kendrick et al., 2015; Fosdick and Blisniuk, 2018). Given this complexity and limited slip rate data, it is has long been debated whether earthquake ruptures on the southernmost San Andreas fault have propagated through the faults within the San Gorgonio Pass region (Allen, 1957; Yule and Sieh, 2003).

The distribution of slip among various strands of the southernmost San Andreas fault is poorly known. The southeastern portion of the Mission Creek strand seems to accommodate the majority of the slip: The slip rate is 12–22 mm/yr (14–17 mm/yr preferred) at Biskra Palms (Fig. 1; Behr et al., 2010; Fletcher et al., 2010) and is 21.6 ± 2 mm/yr at Pushawalla Canyon (Blisniuk et al., 2021). The slip rate of the Banning strand has been proposed to be 2–6 mm/yr in the Indio Hills (Scharer et al., 2016), and it has been measured at 2–6 (4–6 preferred) mm/yr at Painted Hills, near Whitewater (Gold et al., 2015). No slip rate estimates are available for the Garnet Hill fault, but it is thought to be an active right-lateral fault based on the presence of uplifted late Quaternary deposits along a series of left-stepovers (Yule and Sieh, 2003; Cardona, 2016).

Although the Mission Creek strand appears to have the highest slip rate of the fault strands in the Coachella Valley, Holocene slip on this fault strand is not established more than a few kilometers northwest of Highway 62 (Fig. 1), and late Quaternary activity on the continuation of this strand is debated (e.g., Kendrick et al., 2015; Fosdick and Blisniuk, 2018). Instead, slip on the Mission Creek strand has been proposed to transfer northward to the Eastern California shear zone (Nur et al., 1993; Rymer, 1997; Gold et al., 2015), as also suggested by modeling of geodetic data (Meade and Hager, 2005; McCaffrey, 2005; Spinler et al., 2010; McGill et al., 2015).

To better characterize the seismic behavior of the southernmost section of the San Andreas fault, the temporal behavior of these six faults (the Mission Creek, Mill Creek, Banning, Garnet Hill and San Bernardino strands, and the San Gorgonio Pass fault zone) must be examined, including the rate of ground-rupturing prehistoric earthquakes. This paper contributes the first detailed record of dates of prehistoric earthquakes on the Banning strand.

Two significant historical earthquakes have occurred near the Banning strand of the San Andreas fault: the 1986 Mw 6.0 North Palm Springs earthquake and the 1948 Mw 6.3 Desert Hot Springs earthquake. Aftershocks of the 1986 event define a nearly planar surface that strikes N60–70°W, is ~15 km in length, dips northeast, and projects to the surface near the Banning strand trace (Nicholson, 1996). The first-motion focal mechanism for the main shock indicates essentially pure right-lateral slip (Nicholson, 1996). Aftershocks of the 1948 event define a plane that strikes N55°W, dips steeply 60° to 70°NE, is ~15 km long, and projects to the surface near the trace of the Banning strand near the northern end of the Indio Hills (Nicholson, 1996). This event was also predominantly right lateral in slip based on focal mechanisms (Nicholson, 1996). From this seismicity, Nicholson (1996) proposed that the Banning strand is nonvertical, is likely segmented according to fault dip as well as fault strike, and was the primary source of both of these recent, moderate-sized earthquakes in the Coachella Valley. A steeper northeastern dip for the Banning strand was modeled from seismic refraction and reflection lines (Fuis et al., 2017).

The 1986 earthquake triggered up to 9 mm of right-lateral slip 44–86 km southeast of the epicenter along the southernmost San Andreas fault (Sharp et al., 1986; Williams et al., 1988). It also produced some ground cracking on the Banning strand between Whitewater River and Highway 62 (Fig. 1), but this was considered to be caused by strong shaking rather than actual surface rupture (Sharp et al., 1986). The 1948 earthquake was very similar to the 1986 earthquake in that they were both initiated at depth, propagated bilaterally, and did not break the surface (Nicholson, 1996).

We report here: (1) the evidence and ages for eight recent paleoearthquakes on the Banning strand; (2) the average recurrence interval between earthquakes on this strand; and (3) a comparison of the timing of paleoearthquakes on the Banning strand with those on the San Gorgonio Pass fault zone and Mission Creek strand of the San Andreas fault.

Petra Geosciences excavated a paleoseismic trench on the Banning strand at 18th Avenue, North Palm Springs, California (33.9172°N, 116.538°W). The purpose of the trench was to determine the precise location of Holocene fault strands for the development of the site as required by the Alquist-Priolo Act of 1972. The lead consultant on the trench invited us to conduct a more detailed paleoseismic study on the open trench. The four-tier, benched trench was ~250 m long, ~8 m deep, ~9 m wide at the bottom and ~22 m wide at the ground surface. Most of this study focused on the northern end of the trench (Fig. 2), where an ~40-m-wide fault zone was exposed in interbedded boulder, cobble, gravel, sand, silt, and clay deposits. The trench was excavated into the floodplain of Mission Creek, an ~5-km-wide, broad alluviated surface that slopes gently to the south (Fig. 2, inset).

We observed recent faulting in the northernmost 40 m section of the trench. Five fault strands at the north end of the trench exhibited down-to-the-south faulting, whereas the eight to nine prominent fault strands farther south exhibited down-to-the-north separation (Fig. 3; Plate 1). The fault geometries and thickened section between them are consistent with a pull-apart basin, which likely formed as a result of a small right step in the Banning strand that is evident in publicly available light detection and ranging (lidar) data (Fig. 2; Bevis and Hudnut, 2005; Bevis et al., 2005). The trench configuration produced an asymmetrical exposure of the pull-apart basin because the trench is deepest at the south side of the fault zone and thus reveals older section and more vertical separation across faults. Unfortunately, a fiber-optic line prevented us from extending or deepening the northern end to look for similar relationships there. Although additional secondary faults may exist farther north than at the area excavated, we infer that the exposed pull-apart basin (from about 8 to 28 m) represents the main fault zone because the faults bounding it had the largest vertical displacements (Fig. 3; Plate 1) and the strongest facies changes across these faults (Plate 1), suggesting large strike-slip displacements.

Within the pull-apart basin, 1- to 20-cm-thick layers of very fine sand, silt, and clay were interbedded with coarse sand to granule layers up to 1 m thick. The fine-grained layers were thickest near the center of the pull-apart basin and thinned away from the center. Most of the fine-grained layers either pinched out or gradually became coarse grained within 16–18 m from the center of the basin. A few fine-grained layers extended all the way to the southern end of the section of the trench that was logged (~38–40 m from the center of the basin).

South of the fault zone, the trench was dominated by coarse sand and gravel units ~10–30 cm thick; fine-grained layers were absent, and a few layers contained boulders up to 0.5 m in diameter. Overall, the stratigraphy showed distinct and abrupt contacts between layers and consistent lateral continuity along the trench wall. Locally, and often near fault terminations, upper contacts were scoured and overlain by younger deposits. Our ability to follow contacts through the fault zone was locally restricted by two factors: (1) In places, the vertical separation (and presumably large lateral separation) produced strong changes in the texture of units across the fault zones, and (2) the width of the trench, 9 m at the base and 22 m at the ground surface, made the detailed correlation of strata between the west and east walls of the trench impossible, except for a few prominent layers.

Field Work

Both the east and west walls of the trench had four vertical tiers ~1.5 m high separated by three horizontal benches (each ~1.5 m in width). We photographed the entire fault zone and used Structure-from-Motion photogrammetric techniques to produce a scaled photomosaic (Plate 1) that we used to log the trench (Fig. 3). Absolute reference frame and scale were determined with a total station survey of a grid of nails placed ~2 m apart along the top and base of each tier. The photomosaic provided a base for logging of stratigraphic layers and faulting in the field. The final orthomosaic (Plate 1) was rectified using the surveyed nails as control points.


Thirty-three samples of charcoal, typically collected from fine-grained layers, were radiocarbon dated at the Center for Accelerator Mass Spectrometry at Lawrence Livermore National Laboratory to constrain the ages of the prehistoric earthquake horizons. We used the online software OxCal (Bronk Ramsey, 2009) with the IntCal 13 calibration curve (Reimer et al., 2013) to calibrate radiocarbon measurements. The results are listed in Table 1.

In total, 17 samples were collected and dated using post-infrared infrared stimulated luminescence protocol (p-IR IRSL; Buylaert et al., 2009) at the University of California–Los Angeles Luminescence Laboratory. The equivalent dose (De) values for individual grains were measured using a modified single-aliquot regenerative (SAR) dose protocol (Wintle and Murray, 2006) and synthesized into sample De values using conventional age models (Galbraith et al., 1999). Post-IR IRSL225 ages were obtained by dividing De by the geologic dose rate on the DRAC 1.2 online calculator (Durcan et al., 2015). The mean age ± 1σ uncertainty are reported for all luminescence ages (Table 2). Details of the luminescence dating methods used are provided in the Supplemental Material.1

Recognition of Earthquake Horizons

A central challenge in paleoseismic investigations is to identify horizons that were at the ground surface when (prehistoric) earthquakes ruptured the fault. Upward termination of fault strands is a common type of evidence for an earthquake horizon. However, upward termination of fault strands alone does not always provide a reliable indicator of the stratigraphic position of an earthquake horizon because fault strands with small displacements may not necessarily have ruptured all the way to the ground surface at the time of an event (e.g., Bonilla and Lienkaemper, 1991). Earthquake horizons are considered more reliable if a sedimentary response to the displacement is preserved (e.g., Scharer et al., 2017; Onderdonk et al., 2018). For example, when a graben or uphill-facing fault scarp is formed, it may create a closed depression where water will pond and deposit fine-grained material in the depression. In this configuration, the fault scarp may be preserved and buried by the fine-grained sediments, which generally thin and pinch out at the edges of the depression that formed as a result of vertical separation at the surface. Thus, both the fault strand and the subsequent depositional record provide more robust evidence of the paleoearthquake horizon than simple upward termination alone.

At the 18th Avenue trench, most of the exposed fault strands accommodate a component of down-to-the-north displacement. Given the southward slope of the floodplain of Mission Creek, down-to-the-north displacement on individual fault strands created uphill-facing scarps and ponding of fine-grained sediments. Thinning of fine-grained layers across the scarps that filled fault-bounded depressions provided strong evidence for some of the earthquake horizons at the site.

Event Indicator Quality Ranking

At the 18th Avenue trench, we found evidence for eight paleoearthquake horizons with varying quality. To compare the strength of evidence for the different events, we used a ranking scale to classify each of the indicators based on the quality of structural and sedimentological evidence used to identify the stratigraphic level that was at the ground surface at the time of the event. Each event indicator was given a quality rank on a scale of 0 to 5, with higher numbers representing more reliable evidence (Table 3). Table 3 was developed based on a table previously published in Scharer et al. (2017), with several modifications that tailored the criteria to the 18th Avenue trench. Because of unique considerations related to local geology and fault zone architecture, there are always a certain number of site-specific decisions/adjustments that must be made.

First, it is common for faults that slipped with minor displacement to terminate upward at different stratigraphic levels because some of the strands did not rupture all the way to the surface (Bonilla and Lienkaemper, 1991; Weldon et al., 2002). Because of this, Scharer et al. (2017) made a distinction between faults with minor offset versus faults with moderate offset, with the latter being viewed as more likely to have ruptured to the ground surface, thus meriting a higher quality ranking. We chose to classify faults as having “moderate” offset if there was ≥5 cm of vertical separation, or if the correlation of units across the fault was uncertain due to textural changes, indicating that the lateral slip was reasonably large. Previous studies have documented rupture with up to 20 cm of vertical separation that terminated at different stratigraphic levels during the same prehistoric earthquake (e.g., Weldon et al., 2002; Bonilla and Lienkaemper, 1991). In other studies, earthquakes with displacement of only a few millimeters or less have ruptured to the ground surface (e.g., McGill and Rubin, 1999). Our selection of 5 cm to define “moderate” offset falls between these values.

Second, we adjusted the ranking to take into account the effects of uncertainties in stratigraphic correlation that made attribution of an indicator to a specific horizon challenging. Specifically, in cases where the upward termination of a fault was not distinct, and there were multiple horizons that could reasonably be associated with the upward termination, we assigned a quality ranking of 0. If the upward termination was indistinct or unclear, but there was an unfaulted unit that was distinguishable below the next higher earthquake horizon, we assigned a quality ranking of 1 for minor offset and 2 for moderate offset, because this was clearly an event distinct from the next younger event, even if the precise location of the earthquake horizon could not be determined. It is worth noting that our treatment of upward fault terminations was designed to produce a minimum number of paleoearthquakes. In other words, in cases where a fault terminated below the stratigraphic level of a known earthquake horizon elsewhere in the trench, and if there were no unfaulted layers precluding the fault from extending up to the earthquake horizon, we assumed that the fault did extend up to the previously recognized earthquake horizon.

Third, Scharer et al. (2017) made a distinction between folding and thickness changes that are “small” versus those that are “substantial,” with the latter being viewed as more likely to have resulted from coseismic deformation of the ground surface, thus meriting a higher quality ranking. We specified thickness changes of 20 cm or more to be “substantial,” thus justifying a ranking of 3 if a causative fault was not clearly identifiable or 4 if a causative fault was identifiable. Our selection of 20 cm reflected our subjective judgement based on observations in unfaulted portions of the trench where thickness changes and undulations in contacts smaller than this may have resulted from nontectonic causes in some cases.

Stratigraphic Correlation Rankings

Uncertainties in stratigraphic correlation at the 18th Avenue trench led to difficulty in tracing the event horizons along the trench and between the two walls, thus producing additional uncertainty in the event recognition. Because of the broad width of the trench, only four distinctive layers could be confidently correlated between the east and west walls. Even along the same wall of the trench and in the absence of faulting, lateral facies changes, local channel scour, and areas of poor stratigraphy made it challenging to trace some stratigraphic units over long distances (>10 m). Most strata could be clearly correlated across faults with minor offset, but stratigraphic correlation was more difficult across a few of the faults with larger amounts of offset.

To address the uncertainty in our stratigraphic correlation of event indicators, we created a stratigraphic correlation ranking table (Table 4). This correlation ranking table required a “type locale” to be defined for each individual event. We selected the type locale from among the best-ranked event indicators for that specific event (i.e., where the event horizon could be narrowly assigned within a stratigraphic package) and located where it was possible to correlate layers at the type locale with other event indicators. The relative stratigraphic position of each event horizon was traced along the trench wall (i.e., within a particular stratigraphic package) from the type section to the other indicators that occurred at that stratigraphic level. A stratigraphic correlation rating between 1 and 5 was assigned based on the continuity of the package. A rating of 5 means an event indicator could be easily followed all the way to the type section, confirming it was located at the same stratigraphic level as the event horizon at the type section. A rating of 1 means that the correlation was uncertain enough to create ambiguity as to the event horizon to which this indicator should be associated. This situation could result from pinch out of marker beds, mismatch or changes in character of units across faults or along section, position of benches, or presence of bioturbated zones. It is worth noting that we attempted to assign each event indicator to an event that also had other indicators, rather than treating each indicator that had an uncertain correlation as a potential independent event. Our method, therefore, was biased toward producing the minimal number of events necessary to explain the observations.

Characterizing the Likelihood of Each Event

We qualified the likelihood of a paleoearthquake at each stratigraphic horizon based on the quality and the number of individual event indicators. Following the example of Scharer et al. (2017), we used the terms “probable,”“likely,” and “very likely” to denote horizons with increasing probability of representing a paleoearthquake horizon. Also, like Scharer et al. (2017), we used the label “probable” when the number and quality of event indicators suggested at least a 50% likelihood of a surface-rupturing event at that horizon. At this site, we considered events that had three or more indicators with at least one having a quality rating of 2 or higher to be probable earthquakes. Those with one or two indicators with quality rankings of 3 or higher were considered likely or very likely, respectively (Table 5). Horizons with isolated, weak evidence that did not meet these criteria were not given an event number. As noted above, our treatment of upward fault terminations and our approach to correlating earthquake indicators were both designed to produce the minimum number of events. It is quite possible that additional events occurred that were not identified due to these approaches. It is also possible that multiple events occurred at any of the identified horizons if no sedimentation occurred between events.

The orthorectified, georeferenced, and annotated photomosaics for the west and east walls of the trench are shown in Plate 1. A simplified representation of the contacts and faults extracted from Plate 1 is shown in Figure 3. Within Plate 1, strata that could be traced were assigned unit numbers, with unit numbers increasing with stratigraphic depth.

Four correlatable units were used to anchor the stratigraphic numbering of units between the trench walls. The uppermost unit is unit 290, which is a 30-cm-thick silt layer with a distinct thin, brown clay at its base. This unit contains more charcoal fragments than any other unit within the trench. The next lower anchor unit is unit 610. On the east wall, this unit is distinguished by the presence of many large boulders (~0.5 m diameter) within a 0.5- to 1-m-thick layer of coarse sand with pebbles and granules. Boulders this large are not present within any other unit. On the west wall, this unit is sandier, and the boulders are smaller (~0.25 m), but it is still the coarsest unit on the west wall, and it is a similar depth below the surface. Immediately below unit 610, there is unit 620, which is composed of coarse sand, pebbles, and small boulders (~0.25 m diameter) on the west wall and pebbly gravel on the east wall. These two units are separated by a sharp contact. The lowest unit is unit 850, a muddy sand and gravel bed with a sharp upper contact. The unit is brownish in color compared to overlying units, which are grayer in color.

All units on the west wall have a “W” prefix before the unit number, and those on the east wall have an “E” prefix. Unit numbers between 0 and 99 were assigned to strata above the E1 horizon, numbers between 100 and 199 were assigned to strata between the E1 and E2 horizons, and so on. Except for units 290, 610, 620, and 850, it was impossible to correlate units between the two walls. Thus, units that have the same number but a different prefix (E vs. W) should not be interpreted as correlating at all across the trench, nor should units with a lower number on one wall necessarily be assumed to be younger than a unit with a higher number on the opposite wall. Uncertainties in correlation of the event horizons were addressed through the correlation ranking system described above (Table 4).

Some unit labels also include a suffix letter after the unit number. In regions where a layer could be traced or correlated with certainty, the suffix letter is the same. In cases where there was some uncertainty in the correlation of the layer across a fault, bench, or region of poor stratigraphy, a different suffix letter was used on each side of the feature interrupting the ability to continuously trace the layer. For example, unit W110A is a very fine sand layer that crosses a fault at meter 8 on tier 2 of the west wall (Plate 1). The plate shows a possible correlation of unit W110 across this fault, but because the correlation is not confirmed by other layers offset with a similar sense and amount, we call this unit W110B on the opposite side of this fault. Similarly, at meter 12 on tier 2 on the west wall, we make a proposed, but uncertain correlation of this unit across an area of poor stratigraphy, so the unit is labeled W110C.

A complete listing and description of all evidence for each event, including event quality and stratigraphic correlation ratings, may be found in Table S1 in the Supplemental Material. Figure 4 summarizes these data by illustrating the number of indicators for each event, as well as their quality and strati-graphic correlation rankings. The following sections explain key pieces of event evidence in detail. When referring to event indicators, we abbreviate their location by providing the meter number followed by the wall on which the event indicator is located and the tier number in parentheses; e.g., m36W (tier 1) refers to tier 1 on the west wall at meter 36. As will be shown below, applying Table 5, we distinguished one very likely event (E1), four likely events, and three probable events.

E1. Very Likely. The most recent paleoearthquake at the 18th Avenue paleo-seismic site has 16 event indicators. There are three pieces of evidence with a ranking of 3 and two with a ranking of 4 (Plate 1; Table S1), all of which involve sharp upward termination of faults with more than 5 cm of vertical separation and/or stratigraphic units that thin and pinch out against a scarp or folding horizon. An event indicator with a rank 3 is shown in Figure 5A. Two of these five strong pieces of evidence were on the west wall, and three were on the east wall. On each wall, the stratigraphic horizon associated with the strongest evidence for E1 could be traced with moderate to high certainty along the length of the trench. Although it was impossible to physically correlate individual layers in this stratigraphic range between the two walls due to the configuration of the excavation, the sediment packages were similar on both walls (discontinuous 2- to 10-cm-thick layers of coarse sand with granules, fine sand, and silt). Given the similarity in the style of deformation and units both deformed by and postdating the deformation, we infer that the paleoearth-quake closest to the present-day ground surface is the same on both walls. There were also 11 lower-rated indicators. Four were faults that terminated upward at the E1 horizon but that had very small amounts of vertical separation, and seven were faults that terminated 8–40 cm below the E1 horizon but that were assumed to have slipped during E1 because of the lack of any unfaulted stratigraphy that would preclude that interpretation (Table S1). It is possible that these seven faults that terminated 8–40 cm below the E1 horizon could represent one or more additional earthquakes between events 1 and 2, but we think it is more likely that they slipped during E1 and that the faults cannot be traced higher due to poor stratigraphy.

E2. Likely. There are six indicators for this event, including two with quality rank 3, one with rank 1, and three with rank 0 (Fig. 4). The type locale for this event has a quality ranking of 3 and is located at m28w (tier 1) (Fig. 5B). At that location, layer W220E is vertically separated ~40 cm across a zone of four faults. Layers W170E through W108E onlap across an erosive contact that appears to cap the faults. Alternate logging interpreted layer W170E as offset (correlating with W208E or W218E). While the stratigraphic units involved are not sufficiently distinct to determine which interpretation is correct, the event horizon in the alternate interpretation would definitely be below layer W108E, which thins and pinches out over the scarp. Given the location of samples that constrain the age of this event, the uncertainty in the exact paleo–ground surface does not impact the age estimate for the event. However, this uncertainty could allow for multiple events in the vicinity of the E2 horizon.

Another indicator for E2 is found at m17E (tier 2), where a fault terminates at the base of a gravel deposit (Fig. 5C). This fault has moderate vertical separation (12 cm), and layer E290 has been clearly offset. The fault can be traced to the base of a scour below which layer E210 is truncated by the fault. Layer E210 cannot be seen on the southwest side of the fault, suggesting at least moderate lateral offset. Because the earthquake horizon is represented by a scoured surface, it is possible that multiple events occurred at this horizon. This event indicator has a quality ranking of 3 but is on the opposite wall of the trench from the type locale, so we cannot say with certainty that these two indicators formed in the same earthquake. However, they are both located stratigraphically below E1 and above layer 290 (the base of which is the E3 horizon), which we consider to be correlated with certainty between the two walls, so this event indicator is given a stratigraphic correlation rank of 3.

Additional supporting evidence for E2 includes three event indicators that are found on the west wall (two with quality rank 1 and one with quality rank 0). At m7W (tier 3), there are distinct upward terminations of two faults with minor offset (rank 1). At m22W (tier 3), there are two faults with indistinct upward termini and minor offset (Plate 1). Both of these faults terminate at a strati-graphic level that is lower than E1, although the precise level of the event horizon is poorly constrained, and the fault cannot be traced downward to tier 2. Last, there is an event indicator with quality rank 0 that is found at m14E (tier 3). Here, there is a fault with moderate offset of layer E290 that stops at the bench level in between tiers 2 and 3. The exact location of the termination is unknown, but the termination lies stratigraphically above the event 3 horizon and below the event 1 horizon, suggesting it slipped during E2. The imprecise location of the E2 horizon at several indicators as well as uncertainty in correlations of indicators allow for the possibility for additional events between E1 and E3.

E3. Likely. There are six event indicators for E3, with ranks between 3 and 1. The strongest evidence for E3 (type locale) is the pinch out of a ≤40-cm-thick silt layer W290 on the west wall (between 8 and 20 m), centered within the main fault zone and underlying a ≤30-cm-thick package of laminated silty sand with gravel lenses. This ≤30-cm-thick package of laminated silty sand with gravel lenses thins to the north and south, and the underlying silt layer (W290) is absent from the remaining wedge-shaped section at meter 7 (Plate 1). We interpret this silt layer to have been deposited within a graben that formed during E3, resulting in a thick tabular layer of silt (W290) in the center of the graben and silty sand layers that later filled the depression and onlapped across the margins (layer W290 is at the base of the purple package in Fig. 3). Although the north end of the trench was not deep enough to reveal the faults that created this graben, it is clear on the west wall that a graben formed within the main fault zone (between 8 and 18 m) during E2. Thus, it is reasonable to infer that a graben may also have formed here during E3. Based on the stratigraphic evidence, we rank this event indicator with a quality rating of 3.

A similar charcoal-rich silt layer is also observed on the east wall at a similar stratigraphic level, so we correlate it with layer W290. On the east wall, layer E290 can be seen as far south as meter 20. Farther south, layer E290B is interpreted as a possible southward continuation of layer E290, but layer E290B is coarser and more poorly sorted than E290. We suspect that layer E290 thins and pinches out in the vicinity of meter 20, but this is obscured by the incision of a small gully into the trench wall that formed after excavation. Because of the uncertainty as to whether layer E290 pinches out near meter 20 or continues southward as layer E290B, we assign a quality rating of 1 for thickness changes in layer E290.

Three other event indicators associated with E3 are located on the east wall. First, at m18E and m19E (tier 3), there are two faults with significant lateral offset (indicated by the strong mismatch of layers across the faults) that are visible to the top of tier 3 but not visible in tier 2 (Plate 1). Layer E290 lies at the base of tier 2 and is unfaulted over the fault at 19 m, so this fault probably ruptured during E3, and we give this event indicator a quality ranking of 2. The event indicator at m18E (tier 3) has a quality ranking of 2 because layer E290 lies at the bench level and is not exposed over this fault at the base of tier 2. Last, at m27E (tier 2), a fault with minor displacement is capped by layer E290B, which is at the approximate stratigraphic level of layer E290 (and E3). We give this indicator a quality ranking of 1 because the fault has minor (<5 cm) displacement. The stratigraphic correlation ranking for this indicator is 1 (see Table S1). Overall, we rank E3 as likely because the evidence depends on our interpretation that the thickness changes in W290 at the type locale are a product of ground deformation, yet the potentially causative faults all extend higher than layer E290.

E4. Probable. There are six event indicators for E4, including two with rank 2, two with rank 1, and two with rank 0. The strongest evidence is found at m32W (tier 1) (type locale; Fig. 5D), where a fault produces moderate vertical separation of layer W405. This fault is truncated by an erosional contact (spray-painted green in Fig. 5D), which forms the base of a cobbly sand deposit that is unfaulted. The quality ranking is only 2 because it is not clear if the units on the southwest side of the fault were eroded or if the deposit is conformable there. This potentially raises questions as to whether the scoured surface represents the same stratigraphic horizon on both sides of the fault, or whether the fault could have ruptured higher than this scour surface. In either case, there is at least one event at this scoured horizon.

At m23W (tier 2), there is a fault with ~6 cm of vertical separation (moderate offset) that is capped by layer W390, which appears to be unfaulted (Plate 1). The contact at the base of W390 is sharp directly above the fault, but it lacks clear continuity on both sides of the fault, so we assign a quality rating of 2 instead of 3. The event indicators at m32W (tier 1) and m23W (tier 2) are both on the west wall, but they are separated by ~9 m laterally and a bench. Our correlation of stratigraphic units suggests that these two indicators are at the same stratigraphic level. We assign a stratigraphic correlation rating of 3 to this indicator.

At m22W (tier 3), there is a fault with moderate offset of layer W555C. The fault could be capped by layer W470C, near the top of tier 3, or it could continue onto the base of tier 2, and be capped by W390C. In the latter interpretation, this fault would have slipped in E4. In the former interpretation, this fault would have slipped in an earthquake between events 4 and 5. Due to the significant uncertainty in the location of the upward termination of the fault, we assign it a quality ranking of 0.

At m32E and m36E (tier 2), there are two faults that terminate upward with minor vertical separation (quality ranking 1). We cannot be certain that these faults slipped in the same event as E4 on the west wall, but this would be the simplest interpretation and is compatible with the relative stratigraphic position of this feature. Last, at m24E (tier 3), there is a fault with minor offset originally mapped as being capped by unit E480, but it may extend upward and correlate with a fault in tier 2, which could be E3 or possibly even E4. Thus, it has a quality ranking of 1. Overall, E4 is considered a probable earthquake because all of the indicators are upward terminations with less than 10 cm of separation, and we do not observe stratigraphic evidence of significant vertical ground deformation (e.g., pinch out of units across grabens).

E5. Probable. There are six indicators for this event, including one with rank 3, three with rank 1, and two with rank 0. The best evidence and type locale can be found at m50W (tier 1), where we observe four faults with minor vertical separation of layer W590D that are all capped by the unfaulted layer W490D (Fig. 6A). Total vertical separation across the zone is ~10 cm. Therefore, we assign this moderate offset rank 3. The faults may terminate at or up to 20 cm below layer W490D. The 20 cm interval within which the earthquake horizon lies is bracketed by radiocarbon samples A18 and A24, as well as by luminescence samples L13 and L07. Because there are no dated samples from within this 20 cm interval, the uncertainty in the exact event horizon does not affect the event age.

At m26E (tier 3), three faults with minor offset downdrop a debris-flow deposit (E520) into a small graben that is capped by the unfaulted clayey silt layer E490 (Fig. 6B). We give each of these faults a quality ranking of 1. At m23E (tier 3), we also observe a fracture in layer E520 that may connect downward to a fault (quality rank 0) (Fig. 6B). These indicators from the east wall are all clearly capped by the same horizon (E490). At m32E (tier 2), there is a fault splay with unknown but likely very minor offset that is capped by layer E440, assigned a quality ranking of 0. Because E490 pinches out before reaching this location, E440 likely represents the first sediment deposited after E5, and this fault splay likely slipped in E5. Correlation of evidence for E5 between the two walls is uncertain, but this represents the simplest interpretation. We qualify E5 as probable because all of the event indicators are upward terminations, and no unit thickness changes can be clearly followed above the terminations.

E6. Likely. There are five indicators for E6. There is one event indicator with a rank of 3 and four additional event indicators with a rank of 1. The type locale is located at m22–27E (tier 3). Here, the thickness of layers E520 and E525 decreases from 60 cm at 23 m down to 20 cm at 26 m (Fig. 6B). We interpret these layers as part of a postearthquake depositional sequence that filled a closed depression that was formed during E6, similar in character to basin formation in the youngest events. The large and rapid changes in unit thicknesses indicate filling of a depression, so we give it a quality ranking of 3. The earthquake horizon is interpreted to be at the top of layer E610, which is a thick, bouldery debris-flow deposit that has no noticeable thickness change across this region of the trench.

On the west wall, unit W610 consists of boulders within moderately well-sorted coarse sand. We interpret this layer to be the continuation of debris-flow deposit E610, on the opposite wall of the trench. At m23W (tier 4), two faults produce minor vertical separation within layer W610 and are capped by unfaulted layer W590 (Plate 1), and they are given a quality ranking of 1. Farther south on the west wall, we tentatively correlate the thick clay layer W590 with silt layers W590B, W590C, and W590D, which is the most laterally continuous silt layer to the south. At m38W and m39W (tier 2), we observe two faults that have produced minor vertical separation (~2 cm) of layer W604C and are capped by unfaulted layer W590C. If our correlation of layer W590 with W590C is correct, then these two faults provide additional indicators (quality rank 1) for E6. However, it is also possible that layer W590 at 23 m correlates with layer W604C (at 36–50 m). If this alternate correlation is correct, then the two minor faults at 38 and 39 m would be indicators for an event between events 5 and 6.

E7. Likely. There are three event indicators for E7, one with rank 3, one with rank 2, and one with rank 1. The strongest evidence can be seen at the type locale at m32E (tier 4). Here, we observe an 80-cm-thick package of very fine sand, silt, and clay (E670–E690) that thins to 10–15 cm (E690) as it approaches a fault at m34E (Fig. 7); we give a quality rank of 3 due to the substantial thickness change observed here. The causative fault (at 34–36 m) has reruptured in a younger earthquake.

The quality rank 2 indicator can be seen at m36E (tier 3) (Fig. 7). Here, we observe a fault with 5 cm of vertical separation in the upper part of tier 4 (pebble layer E710), capped by an unfaulted fine sand layer (E690) at the base of tier 3. Because the fault and the capping layer are not visible on the same tier, this event indicator is given a quality ranking of 2 instead of 3. Correlation of units between 32 and 36 m on the east wall is clear enough that we are confident that these two indicators represent the same earthquake horizon. The indicator at m36E (tier 3) has thus been assigned a stratigraphic correlation ranking of 4 with respect to the type locale at 32 m.

The rank 1 indicator can be seen at m28W (tier 4), where layer W690 thins to the south and pinches out (Plate 1). The causative fault is likely the fault at 28 m. Although we observe moderate thickness change, which suggests filling of a closed depression, layer W690 is only exposed over a short distance (~2 m), so the full extent of the closed depression is not visible. Therefore, we assign a quality ranking of 1. It is not clear whether layer W690 is the same as layer E690 on the opposite wall of the trench, but it is clear that on both the east and west walls, there is evidence for two events (events E7 and E8) between layers 610 and 850, both of which can be correlated between the two walls of the trench with a high degree of confidence.

Overall, E7 is considered likely because the mismatch of unit thickness at the type locale (Fig. 7) occurs across a fault and may be the product of later lateral motion (the fault strand extends almost to the surface and last ruptured in E1) rather than local graben formation and filling.

E8. Probable. There are three event indicators (all rank 2) for the oldest event horizon observed at this paleoseismic trench site. Two event indicators with a quality rank of 2 are located at m31W and m36W (tier 4), with the latter being the type locale. At m31W, there is a fault with moderate vertical separation (22 cm) of layer W815-W815B, which is then capped by a layer of fine sand (layer W790) (Plate 1). The basal contact of layer W790 is sharp directly above the fault, but it lacks clear definition to the north; therefore, we assign a quality rating of 2 rather than 3. At the 36 m location (type locale), we observe a fine sand layer (W790C) that thins to the south and pinches out against a fault at 36 m. We infer that layer W790 filled a depression that formed during E8. Only a small section of the pinching layer W790 is visible just below bench level. Therefore, we assign a quality ranking of 2 rather than 3. We tentatively correlate layer W790 with W790B and W790C, which would imply that these two event indicators were caused by the same event. However, is it possible that W790B correlates with W815B instead of with W790, and that W790 has no correlative unit south of the fault strand at 32 m, so we assign a correlation ranking of 3. In addition, it is not clear whether layer W790 is the same as layer E790 on the opposite wall of the trench. Nonetheless, it is clear that on both the east and west walls, there is evidence for two events (events 7 and 8) below layer 610 and above layer 850, and both units 610 and 850 can be correlated between the two walls of the trench with a relatively high degree of confidence.

At m36E (tier 4), layers E780 and E790 thin and pinch out to the south, and we infer these layers represent growth strata filling in a depression produced by an earthquake (Fig. 7). Movement on the fault at 38 m may have caused the depression in which these growth strata were deposited. Although the thickness changes are moderate, the fault at 38 m is not clearly linked to the growth strata, leading to quality rank of 2 for this event indicator.

Radiocarbon and Luminescence Dating

Layer ages were obtained from 34 radiocarbon-dated detrital charcoal samples and 17 sediment samples dated using post-IR IRSL techniques (Tables 1 and 2). All dates are plotted in Figure 8 as a function of stratigraphic depth measured at m24W down to layer W620, and then at m34 for layers older than W620 (Fig. 3). These sections were selected because they are within a wide, fault-bounded block with clear stratigraphy and are close to the area where layers are thickest. On Figure 8, all samples from the east wall, and samples from the west wall that are not from the type section are shown with vertical error bars that indicate the uncertainties in the stratigraphic positions of these samples relative to the type section.

The stratigraphic thickness of 9 m exposed in the trench was deposited within the past ~7000 yr, indicating an average depositional rate of ~0.13 cm/ yr. Based on the youngest samples at each depth, the sedimentation rate appears to have been relatively constant across the section, yet it could have been variable from unit to unit given the relatively sparse dating. The longest possible depositional hiatus can be no more than the 1–1.5 k.y. interval between samples L01 and L17 (~4–5 m depth). There is no clear indication of a hiatus at this time, and a hiatus this long is only possible if deposition stopped immediately after the layer containing L17 formed, and the overlying 1 m of sediment between sample L01 and L17 was deposited very rapidly. Based on texture and bedding, the layers between these two samples could have been deposited in a few depositional events. Episodes of rapid sedimentation may have also occurred, such as between samples L17/L18 and L20/L04/L02, which all have nearly the same age within the 1σ uncertainties, despite spanning ~1.5 m of stratigraphic depth. This depth range is dominated by the two thickest units in the trench, the boulder units 610 and 620, which were likely deposited very rapidly, potentially during the course of one or a few storms. The presence of scouring within some areas (e.g., Figs. 5A, 5B, and 5C) suggests that parts of the record are locally missing.

Paleoearthquake Model Ages

We used OxCal (Bronk Ramsey, 2009) to estimate the earthquake ages. OxCal uses Bayesian statistics to model posterior ages for paleoearthquakes based on all chronological constraints such as relative stratigraphic position and layer groups. Dates from the same layer, or from layers that we infer correlate with each other, were modeled using the “Phase” command such that relative stratigraphic position within the layer group was not applied (Bronk Ramsey, 2009).

The range of ages for several intervals shown in Figure 8 presents a challenge for constructing a straightforward age model. For example, samples A44 and C10 at 2 m stratigraphic depth have ages that differ by 2000 yr, as do samples A18 and A39 at 2.75 m stratigraphic depth. Many studies have shown that it is common for detrital charcoal samples to be older than the age of the layer from which they were collected by an amount that depends on the length of time between when the plant material stopped growing and when the sample was deposited in the location from which it was collected (e.g., Philibosian et al., 2011; McGill et al., 2002; Fumal et al., 2002). Aside from contamination of the sample, the only way for charcoal to underestimate the depositional age of a layer is if it was a root or was brought into the section by bioturbation. We were careful in the field to avoid collecting roots or any charcoal samples from obvious filled burrows or areas lacking clear stratigraphy, which could potentially be bioturbated zones, and we reviewed photographs of the sample locations to evaluate the potential for unrecognized bioturbation at each sample location. To construct our preferred age model, we therefore omitted any charcoal samples with dates that were older than other charcoal or luminescence samples from the same layer or lower layers. In Figure 8, dated samples that were included in our preferred age model are plotted with filled symbols, and those that were excluded are plotted with open symbols.

In total, 29 out of 34 radiocarbon dates were omitted from our preferred age model, which thus relied heavily on the IRSL ages to constrain the timing of past earthquakes. The omission of this many radiocarbon samples is supported by previous studies that have shown wide ranges of radiocarbon ages for charcoal samples from the same layer (e.g., McGill et al., 2002; Fumal et al., 2002; Philibosian et al., 2011). In the 18th Avenue trench, we also collected multiple samples from the same depths that have mean radiocarbon ages separated by as much as 2000 yr (e.g., samples A44 and C10 at ~2 m depth and A18 and A39 at ~2.6 m depth), suggesting that detrital lag times can be at least 2000 yr long on the Mission Creek fan. Fumal et al. (2002) also found lag times of up to ~600–1000 yr for charcoal samples from the same layer at a site on the Mission Creek strand at the Thousand Palms site, located ~20 km to the east, and that site, like the 18th Avenue trench, also has a large source catchment.

One IRSL sample from the east wall (L05) was also excluded from our preferred age model because it was older than most other samples at that stratigraphic depth, and it had large uncertainty in its stratigraphic depth relative to the type section on the west wall. The results of the preferred OxCal model are shown in Figure 9 and Table 6.

While it is common practice in paleoseismic studies to omit charcoal samples that are older than other charcoal samples from the same or underlying layers (McGill et al., 2002; Fumal et al., 2002; Philibosian et al., 2011), a factor that may be of concern in this study is that we also omitted charcoal samples that were older than luminescence samples from the same or underlying layers, resulting in an almost exclusive reliance on the IRSL ages. Therefore, we also created an alternate model that took a different approach based on the observation that the 2000 yr lag was based on only two radiocarbon samples. In contrast to the preferred model, the alternate model (Fig. A1) starts with the assumptions that those samples (A44 and A18) were contaminated by younger material (which can occur during sample collection and preparation), and that radiocarbon samples C10 and A39 were accurate. Samples A44 and A18 were thus removed from the OxCal model, along with sample A31, because it was much older than the five IRSL samples near the base of the section, which we assume are correct based on their reproducibility. To achieve an OxCal agreement index >50%, samples L16, L06, L07, and L01 were also removed because they were notably younger than both radiocarbon and IRSL samples in the same phase. Given that the IRSL signal measured for these samples is observed not to fade through time (Fig. A2), the only reasonable explanation for an apparent IRSL age that is too young is postdepositional mixing. Considering that it is unlikely that postdepositional mixing of grains would systematically shift ages in a way that is stratigraphically consistent, these were not removed in the preferred model. However, this alternative model has some advantage of retaining more of the original radiocarbon data, as it used 20 of the original 34 radiocarbon and 13 of 17 IRSL samples. In comparison to the preferred model, this approach produced much older earthquake ages for E2–E6 (Fig. 10; Table 6). Although both models are feasible, we chose the preferred model because it is common for detrital carbon samples to be older than true depositional ages, and the fading tests (Fig. A2) indicated that the IRSL samples are unlikely to be biased too young.

Recurrence Intervals

As shown in Figure 4, the overall number of observations and quality rank of event indicators decrease with depth. There are several reasons why the evidence for older events is weaker than that for the younger events. First, the older strata are only exposed within a small portion of the fault zone. Second, evidence for older events has been overprinted by younger earthquakes, making interpretation more difficult. Stratigraphic correlation of event indicators to their respective type locales was also a major challenge in this trench at all stratigraphic levels, but it was compounded for older events due to increased difficulty in correlating layers across faults in which the cumulative lateral offset in multiple events was large.

The earthquake record at the site includes evidence for five events in the upper 3 m, which occurred in the past 2.4–3.3 k.y. (age range for E5). The mean recurrence interval for these five events on the Banning strand is 490 yr (95% range 390–600 yr). The individual intervals between each of these five events are very similar to the mean (Fig. 10). As noted above, we cannot rule out the possibility that additional earthquakes may have occurred and not been recognized. If the alternate age model is correct, ages for E2 through E6 are 600–1800 yr older than in the preferred model, producing a mean recurrence interval of 940 yr, i.e., nearly twice as long as that in the preferred model (Fig. 10).

Only three events are recognized in the lower 6 m, during a 2.4–4.7 k.y. period between events E5 and E8 (Table 6). This suggests that either we are missing events in the lower section, or the recurrence pattern at the 18th Avenue site is variable. We cannot rule out either possibility. However, it is likely that the older part of the earthquake record at this site is incomplete due to limited exposure in which to make observations of the older stratigraphy. Specifically, the trench does not expose stratigraphy older than E3 north of m16, nor does it expose stratigraphy older than E5 north of m20. Faults with large vertical separations are present in the trench north of m20, yet we had no way to investigate paleoearthquakes older than E5 on these faults. In addition, if there was a 1–1.5-k.y.-long hiatus between the deposition of samples L17 and L01, as permitted in the dating, it is possible that we missed the detection of events during this period.

If the full record of up to eight earthquakes is complete, the average recurrence interval during the past ~7000 yr would be ~790 yr (95% range 680–910 yr; Table 7). If the alternate age model is correct, the average recurrence interval for all eight events would be 770 yr (95% range 680–870 yr; Table 7).

Comparison to Other Paleoseismic Sites

The mean recurrence interval between the past five surface-rupturing earthquakes at the 18th Avenue site is 490 yr (95% range 390–600 yr), which is longer than it is for the three sites on the Coachella section and Mission Creek strand (116–300 yr; Philibosian et al., 2011; Sieh and Williams, 1990; Fumal et al., 2002) and shorter than or overlaps with the measured recurrence intervals at the Cabazon trench site on the San Gorgonio Pass fault zone (two intervals between three paleoseismic events have been identified: ~450 and ~1600 yr; Wolff, 2018; Scharer and Yule, 2020). Thus, earthquakes at the 18th Avenue site on the Banning strand occur less frequently than those on the Coachella section and Mission Creek strand but more frequently than those on the San Gorgonio Pass fault zone.

Figure 11 shows the ages of paleoearthquakes at the 18th Avenue site in comparison with prehistoric earthquakes at other sites on the southernmost San Andreas fault. One of the most interesting features shown in Figure 11 is that the age of the two most recent paleoearthquakes at the Cabazon site and 18th Avenue site are contemporaneous. To illustrate how such observations can be modeled as ruptures, we provide two possible rupture histories for the southernmost San Andreas fault near San Gorgonio Pass (Figs. 11B and 11C). Both models are consistent with the dates of paleoearthquakes on the southern San Andreas fault, but they show different correlations that could reflect different geometric segmentation; many other scenarios are possible given the input data (Scharer and Yule, 2020). The first scenario (Fig. 11B) shows an event ca. 1200 CE that connects contemporaneous earthquake ages at Millard Canyon (Mi1), Cabazon (Ca1), and 18th Avenue (E1). Because no earthquakes at Burro Flats (Bu) or Coachella (Co) overlap in age with the limits of Ca1 and E1, this rupture was confined to the San Gorgonio Pass fault zone and Banning strand. This is also consistent with no rupture observed in the past 600 yr at the East Whitewater site on the Garnet Hill strand (Cardona, 2016). The second scenario shows an alternative that separates this rupture into two earthquakes between ca. 1100 and 1300 CE. In Figure 11C, E1 and Co5 are connected for a rupture along the Banning strand and Coachella section, followed by a rupture that is centered in San Gorgonio Pass fault zone, connecting Mi1 and Ca1. Although the San Gorgonio Pass fault zone rupture is relatively short, it could have continued onto other mapped faults in the region; for example, the San Gorgonio Pass fault zone extends for another 40 km to the west (Fig. 11; Yule and Sieh, 2003).

In contrast to the more segmented ruptures ca. 1200 CE, it is possible that the entire model domain in Figure 11 ruptured at 600 CE, as ages permit a San Bernardino–San Gorgonio Pass fault zone–Banning–Coachella rupture, potentially similar to the “Shakeout scenario”(Jones et al. 2008). Scharer and Yule (2020) showed that this rupture had the potential to extend onto the Mojave section of the San Andreas fault. As shown here, this single through-going rupture at 600 CE would have been at least 150 km long, equivalent to M 7.5 or greater using regression equations in Wesnousky (2008). It is worth noting that the event at 600 CE could also be modeled as separate ruptures. Only the two most recent paleoearthquakes at the 18th Avenue site occurred within the time period represented by the Burro Flats record. Although the most recent event at the 18th Avenue site may correlate with Mi1 and Ca1 on the San Gorgonio Pass fault zone, there is no record of an earthquake at this time on the San Bernardino strand at Burro Flats. The 600 CE event is the only Banning strand event that could potentially correlate with an earthquake on the San Bernardino strand within the past 1600 yr, assuming all the paleoseismic records are complete.

A general observation is that in the past 1600 yr, there have been eight earthquakes each at the Burro Flats (San Bernardino strand) and Coachella (Coachella section) paleoseismic sites, while there have only been two at the 18th Avenue site on the Banning strand and at the Cabazon site on the San Gorgonio Pass fault zone, and five on the Mission Creek strand at Thousand Palms. The more frequent earthquakes outside of the sites along the Banning strand and the San Gorgonio Pass fault zone contrast with the fewer earthquakes at those locations and control potential rupture histories in the area. Interestingly, a maximum of two earthquakes connecting the Coachella and northern Banning strands is permitted in the past 1600 yr. This requires that six ruptures on the Coachella section either terminated on the southern Banning strand, extended onto the Mission Creek strand, or ended at the splay (see also Scharer and Yule, 2020). These data generally support the dynamic rupture models of Douilly et al. (2020), which found that rupture on the Banning strand was less likely to occur given the geometry of the fault splay. Only the most recent earthquake at the 18th Avenue site occurred during the period observed at the Thousand Palms site on the Mission Creek strand, and it may have occurred at the same time as TP4 (Figs. 11B and 11C), allowing the possibility of rupture jumping from the Mission Creek to the Banning strand, as seen in some dynamic rupture model configurations (Douilly et al., 2020).

Prior to 400 CE, paleoearthquake data from nearby sites are sparse (Fig. 11D); only the Cabazon site has a record as long as the 18th Avenue site (Fig. 11D). In this time period, two ruptures connecting 18th Avenue and Cabazon are permitted, ca. 3600 BCE and 1000 BCE. While the E3 and Mi3 ruptures were contemporaneous, there is no correlative event at Cabazon if Ca2 ruptured with Mi2, as indicated. Wolff (2018) noted that missing events were likely at Cabazon, which is shown with an X on the speculative rupture ca. 200 CE that would span the Banning and San Gorgonio Pass fault zone.

Implications for Average Slip per Earthquake

The Banning strand of the San Andreas fault has a Holocene slip rate of 2.3–6.2 mm/yr (Gold et al., 2015), based on a fan offset at the Painted Hills site, ~8 km northwest of the 18th Avenue site (Fig. 1). Given the elapsed time since the most recent event (945–690 calibrated years B.P. or cal yr B.P.) and the slip rate at Painted Hill, we calculated that the fault could produce slip of 1.6–5.9 m in the next earthquake, using a slip-predictable model, and assuming a relatively constant strain release. Using the average recurrence interval of 390–610 yr between events E1 and E5 and the slip rate, we estimated that the average slip in these past five events was 1–4 m. These estimates of average slip per earthquake are simple because they do not include uncertainties related to variability in strain release, and they assume on-fault displacement is comparable at each location. Alternatively, as the fan age at Painted Hill overlaps with the section dated at 18th Avenue, we can infer that the fan has experienced 6–8 earthquakes if all the ruptures spanned the distance between both sites. This would produce an average slip per event of 2.5–5 m for the northern Banning strand, assuming the 18th Avenue site record is complete. We note that this range is similar to that predicted from the accrued slip since the most recent event, which may indicate the trench record is complete, or nearly so.

The 18th Avenue site provides the first dated paleoseismic record constructed on the Banning strand of the southernmost section of the San Andreas fault. This 7000-yr-long record can now be compared to paleoseismic records on the neighboring strands that make up the complex network of faults in the southern section of the San Andreas fault. Eight horizons contained evidence of paleoearthquakes based on sedimentological responses to deformation and fault terminations at or below each horizon. These earthquakes were qualified based on the record of deformation and our ability to correlate across this wide trench; E1 is considered very likely; E2, E3, E6, and E7 are considered likely; and E4, E5, and E8 are considered probable. Based on the stratigraphy, dating uncertainties, and physical limits of the trench exposure, the likelihood of missed events is greatest in the older parts of the section, below E5. The most recent event occurred 945–690 cal yr B.P. The open interval is longer than the average interval between events E1–E5 (390–600 yr), but it is shorter than the interval between events E5 and E6 and between events E6 and E7. The longer average interval for the entire section (680–910 yr for events E1–E8) may indicate temporal variations in the rate of strain release on the Banning strand, or it may indicate missing events in the older section of the trench. In the past 1600 yr, eight paleoearthquakes are documented at both the Burro Flats and Coachella sites, five are documented on the Mission Creek strand at Thousand Palms, and only two are documented on the Banning strand and San Gorgonio Pass fault zone. Given the earthquake ages, a maximum of two earthquakes connecting the Coachella and northern Banning strands is permitted in the past 1600 yr, and only one of these events (ca. 600 CE) could have potentially ruptured onto the San Bernardino strand as well. The interpretation of fewer earthquakes on the Banning strand is consistent with its lower slip rate compared to other sections of the southernmost San Andreas fault and would translate to average slip per event of ~1–5 m.

Figure A1 shows the radiocarbon and luminescence samples used in the alternate age model, and Figure A2 shows the OxCal model for those samples.

This research was funded by U.S. Geological Survey (USGS) National Earthquake Hazards Reduction Program grants G18AP00040 and G18AP00041, and internal funding from California State University–San Bernardino (CSUSB) and the USGS. We are extremely grateful to John Rogers for granting access to work on his property and Alan Pace from Petra Geosciences for accommodating our study. We want to thank all the students from CSUSB and California State University–Northridge who assisted with the hard labor. Reviews by Belle Philobosian, Ray Weldon, Peter Gold, and C.H. Jones improved the clarity of the manuscript. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the U.S. government. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. government.

1Supplemental Material. Details of luminescence methods and results, and table of event indicators. Please visit https://doi.org/10.1130/GEOS.S.14098589 to access the supplemental material, and contact editing@geosociety.org with any questions.
Science Editor: Andrea Hampel
Associate Editor: Craig H. Jones
Gold Open Access: This paper is published under the terms of the CC-BY-NC license.