Paleoseismic Evidence of an Mw 7 Pre‐Hispanic Earthquake in the Peruvian Forearc

We present the results of a paleoseismic survey of the Incapuquio Fault System, a prominent transpressional fault system cutting the forearc of South Perú. High‐resolution Digital Elevation Models, optical satellite imagery, radiocarbon dating, and paleoseismic trenching indicate that at least 2–3 m of net slip occurred on the Incapuquio Fault generating a complex, ∼100‐km long set of segmented fault scarps in the early 15th century (∼1400–1440 CE). We interpret the consistent along‐strike pattern of fault scarp heights, geometries and kinematics to reflect a surface rupture generated by a single Mw 7.4–7.7 earthquake, suggesting that brittle failure of the forearc poses a significant, yet mostly overlooked, seismic hazard to the communities in coastal areas of Perú. The timing of this earthquake coincides with the collapse of the Chiribaya civilization in ∼1360–1400 CE, and we present evidence of damaged buildings along the fault trace that may be of Chiribayas age. Our surface faulting observations, when combined with observations of deformation in the forearc from geodesy and seismology, also demonstrate that the forearc in South Perú experiences a complex, time‐varying pattern of permanent strain, with evidence for trench‐parallel shortening, trench‐parallel extension, and trench‐perpendicular shortening all in close proximity but in different periods of the megathrust earthquake cycle. The kinematics of recent slip on the Incapuquio Fault are consistent with the sense of interseismic strain within the forearc measured by GPS, suggesting the fault is loaded toward failure between megathrust earthquakes.

ruptured in at least one earthquake that could have been up to M w 7.4-7.7 from observations of fault scarps in optical satellite imagery, high-resolution Digital Elevation Models (DEMs), field mapping, slickenline measurements, paleoseismic trenching, and radiocarbon dating. We discuss how the kinematics of this earthquake compares to other kinematic indicators of deformation in the forearc of South Perù. Finally, we present evidence that a large earthquake on the IFS may have contributed to the collapse of the Chiribaya civilization at the start of the 15th century.

Geological and Seismotectonic Setting
The Andes are the classic example of mountain building along a convergent ocean-continent plate boundary. Crustal thickening and uplift along the western margin of South America has been driven by Mesozoic-Cenozoic subduction of the Nazca Plate (Horton et al., 2001;Isacks, 1988;Jaillard et al., 2002;McQuarrie et al., 2005;Ramos, 2010). At the latitude of South Perú, the Central Andes can be divided into five different zones based on their morphology: (a) the forearc, (b) the Western Cordillera, (c) the Altiplano, (d) the Eastern Cordillera, and (e) the sub-Andes (Figure 1, inset). The forearc morphology is characterized by a gentle, westward-dipping topographic slope extending from the magmatic arc toward the coastline. Ephemeral rivers draining across the forearc from the Western Cordillera have incised deep, NE-SW trending valleys into its gentle surface, creating a prominent erosional fabric and depositing vast Quaternary alluvial fans ( Figure 1). Faulting within the forearc generates differential uplift, causing variations in the depth of incision within these valleys, abandonment of the alluvial fan surfaces, and deflections in the NE-SW drainage pattern (Audin et al., 2006;Hall et al., 2012). Thermochronological dating of exhumation within canyons near Arequipa suggests a period of enhanced incision in the Western Cordillera and forearc initiated around 14-10 Ma and ended 5-2.2 Ma (Schildgen et al., 2007), around the same time as paleoelevation estimates BENAVENTE ET AL.  (Audin et al., 2006;Benavente et al., 2017). The letters A, B, C, and D indicate the location of the Toquepala, Chintari, Tauja, and Santa Elena zones, respectively. The yellow zones indicate the areas where we built high-resolution DEMs using a drone. T1, T2, and T3 mark the locations of paleoseismological trenches. Blue lines represent the major rivers, squares indicate the main population centers and the black diamonds represent archeological sites (https://www.geoidep. gob.pe/ministerio-de-cultura). The inset shows the location of the study area in a yellow rectangle with the main morphological units of the Andes labeled (F, Forearc; WC, Western Cordillera; AP, Altiplano; EC, Eastern Cordillera; SA, Sub-Andes; NR: Nazca Ridge). Plate velocities are shown relative to stable South American (DeMets et al., 2010) and black triangles mark active volcanoes. imply significant uplift of the Western Cordillera (Garzione et al., 2017). Since 5-2.2 Ma, drainage has been focused into the canyon systems, and the extreme aridity has led to low denudation rates (<0.5 m/Ma) on the abandoned alluvial fan surfaces (e.g., Dunai et al., 2005;Hall et al., 2012). Therefore, fault-related Quaternary geomorphology is well preserved throughout the forearc.
The role that forearc faults play in uplifting the Western Cordillera relative to the coastline and in accommodating oblique subduction within South Perú remains unclear. Benavente et al. (2017) identified several NW-SE and E-W trending, active left-lateral transpressional faults in the forearc of South Perú that uplift the Western Cordillera and accommodate a component of the oblique convergence ( Figure 1). In contrast, Schildgen et al. (2009) proposed from field observations and thermochronology studies that between 14 and 2 Ma normal faulting in the forearc generated up to 2 km of relief in the Arequipa region. Norton and Schlunegger (2011) have argued from landscape modeling of channel morphologies that little deformation has occurred in the Perúvian forearc since ∼7 Ma. While, Hall et al. (2012) and Armijo et al. (2015) have proposed that recent deformation of the Western Cordillera and forearc is associated with monoclinal warping above west-vergent thrusts subparallel to the subduction megathrust. Differentiating between these contrasting structural interpretations of deformation in the forearc requires measurements of the geometry and kinematics of faults in the region.
The IFS is one of the major faults within the forearc of South Perú and trends subparallel to the topographic transition between the forearc coastal plain and the deeply incised Western Cordillera ( Figure 1). The IFS consists of multiple fault strands active during the Cenozoic that cut Jurassic sedimentary and late Cenozoic volcaniclastic rocks with transpressional kinematics, and which dip steeply toward the northeast beneath the Western Cordillera (Bellido, 1979;Jacay et al., 2002;Wilson & García, 1962). The earliest studies to map the IFS also identified near-vertical bedrock scarps, and fresh scarps in alluvial deposits, along the fault trace with an apparent normal sense of motion (Audin et al., 2006;Macharé et al., 2003;Sébrier et al., 1985). The reported scarps have an along-strike length of almost 100 km and may represent the surface rupture generated in a significant pre-historic earthquake; though no paleoseismic trenching, dating, or detailed scarp morphology data currently exist to test this hypothesis. In addition, several of the newly identified active transpressional fault systems in the forearc connect with the IFS (Benavente et al., 2017;Hall et al., 2012), raising the question of whether recent slip on the IFS was in fact normal in sense. In this study, we explore the recent kinematics and history of slip on the IFS using structural and geomorphological measurements of these fault scarps.

Fault Scarp Mapping
The Geological Survey of Perú (INGEMMET) has been updating the national neotectonic map (e.g., Benavente et al., 2013Benavente et al., , 2018Macharé Ordoñez et al., 2009). As part of this project, we mapped the IFS fault scarps using 1-m resolution optical satellite imagery from Digital Globe (Google Earth) and 12-m resolution Tandem-X DEMs (Rizzoli et al., 2017), covering a surface area of ∼2,000 km 2 with detailed inspection. Within the optical satellite data, the IFS is delineated by a sharp, linear boundary caused by vegetation contrasts, contrasts in bedrock color, shading from topographic scarps, deflections in the drainage pattern, and topographic steps. From these satellite observations, we built a preliminary map of the 100 km long fault trace between Moquegua and Aguas Calientes. These preliminary fault maps were then used to guide subsequent field work.
In order to study the morphology of the fault scarps, we constructed high-resolution DEMs at three localities (sites B, C, and D in Figure 1) using the structure-from-motion technique. Photographs were collected from an eBee senseFly drone and processed with Agisoft Photoscan (Westoby et al., 2012). Ground control points used to guide the image matching were located using differential GPS. The resulting point clouds have at least 1 point per 10 cm 2 horizontal resolution, and are complemented with orthomosaic imagery at 5 cm 2 pixel resolution. The DEMs are available through OpenTopography (www.opentopography.org; see Acknowledgments).
Fault scarp heights were calculated using two methods: (a) estimating the vertical separation of correlatable surfaces either side of a scarp using two-dimensional topographic profiles perpendicular to the scarp strike (e.g., Wolfe et al., 2019), or (b) estimating vertical offsets from offset layers exposed in trench sections (e.g., Morell et al., 2018). Uncertainties in the scarp heights due to noise in the point cloud and the extrapolation of the alluvial surfaces across the scarp were typically an order of magnitude smaller (∼10-20 cm) than the along-strike deviations in the measured scarp heights. The scarp heights were converted into fault throw and net fault slip using the strike, dip, and rake measurements of slickenlines exposed on nearby bedrock sections of the fault plane, and the dip and aspect of the offset surfaces (e.g., Mackenzie & Elliott, 2017). Where multiple scarp heights could be measured on a single section of the fault, we used the distribution of scarp heights and differences in the scarp morphology to discriminate cumulative scarps from single-event scarps (e.g., Grützner et al., 2017).

Fault Kinematics
Many of the fault scarps along the IFS formed in consolidated alluvial and volcaniclastic rocks and preserved slickenlines on an exposed surface, typically a southwest-facing overhang. From herein, we refer to these types of scarps as "bedrock scarps." Given the protracted Cenozoic history of movement along the IFS, it is possible that sections of the bedrock scarps preserve slickenlines from ancient episodes of slip that do not reflect the fault's current kinematics. Therefore, we only collected slickenline measurements from bedrock scarps that either show evidence for offsetting late Quaternary sediments, or transition along-strike into a scarp that offsets late Quaternary deposits.

Paleoseismology
In order to determine the chronology of scarp formation, we studied the stratigraphy in three natural exposures of the fault using the standard protocol for cleaning, recording, and analysis proposed by Mc-Calpin (2009). Charcoal, sediment, and root samples from offset layers and colluvial wedges were collected for radiocarbon dating. Each of the samples was placed in an aluminum foil packet, and was dated by the Beta Analytic laboratory in Miami. All radiocarbon ages were modeled with the OxCal v4.3.2 online program, which uses a Bayesian statistical method (Bronk Ramsey, 2017) to propagate analytical errors and errors in the calibration curve SHCal13 (Hogg et al., 2013) through the age calculation, providing probabilistic estimates for the age of each sample.

Teleseismic Body-Waveform Modeling
The focal mechanisms of earthquakes can place additional constraints on the timing and kinematics of deformation in the forearc around the IFS. We used the long-period teleseismic body-waveform modeling program MT5 of Zwick et al. (1994), which is based on the algorithm of McCaffrey and Abers (1988), to determine point-source focal mechanisms and centroid depths for two moderate-magnitude earthquakes with epicenters near the IFS (Event 1: M w 5.7 August 26, 2003 and Event 2: M w 5.5 November 20, 2006) and one earthquake that ruptured the forearc to the southwest of the IFS 20 km offshore (Event 3: M w 6.2 May 06, 2010). The MT5 waveform-modeling method has been described extensively elsewhere (e.g., Devlin et al., 2012;Molnar & Lyon-Caent, 1989); therefore, we provide the details of the data processing, and the velocity models used to generate the synthetic seismograms, in Text S1.

Fault Scarps Along the Incapuquio Fault Zone
Fault scarps on the IFS can be traced discontinuously between Moquegua and Aguas Calientes using optical satellite imagery and high-resolution DEMs (Figure 1). The scarps can't be traced continuously, because rivers cross cutting the fault have incised deep and wide canyons, eroding the morphological evidence of the IFS in several areas. As a result, the morphological trace of the fault is partitioned into four different along-strike zones: The Toquepala, Chintari, Tauja, and Santa Elena Zones, which are labeled A, B, C, and D in Figure 1, respectively.

Toquepala Zone
The Toquepala Zone is the northern-most section of the IFS where we have been able to identify fresh fault scarps (A in Figure 1). In this region, the IFS is marked by a 20 km long, NW-SE trending topographic step in the west-dipping slope between Toquepala Village and the Rio Salado, with the fault trace being picked out by sharp changes in stream incision and rerouted drainage along the fault strike (Figures 2a and 2b). The fault is also distinct in geological maps of the region, as it separates late Cretaceous-early Paleogene granite porphyrys that outcrop northeast of the fault from Eocene-Miocene volcaniclastic rocks to the southwest (INGEMMET, 2017). The fault trace itself has a number of left steps, where the fault strike becomes oriented ∼E-W (Figure 2a). We mapped 8 km of fault scarps at Toquepala that were delineated by changes in the vegetation, shading or ground surface color in optical satellite images (Figure 2c). In between the fault scarps, rivers and mass movements have eroded morphological evidence of recent faulting.
During the field investigation, we identified a locally exposed, ∼2.5 m high bedrock scarp within a 1 km left step in the fault trace (Figures 2d and 2e). The scarp cuts through consolidated, poorly sorted debris BENAVENTE ET AL.   Figure 2d, and a stereonet showing fault slickenline data (black arrows indicating movement of hanging-wall with respect to the footwall) produced using FaultKinWin(R) program (Allmendinger, 2018). IFS, Incapuquio Fault System. flow deposits containing subangular to rounded clasts ranging in size from pebbles to boulders. Fault plane measurements from the exposed scarp had a consistent orientation 270°/40°. Slickenlines preserved on the surfaces of the clasts within the debris flow deposit also had a relatively consistent orientation, pitching between 58° and 67° toward the northeast (Figure 2d, inset). The slickenline orientations suggest the IFS accommodates left-lateral transpression near its northern tip (Figure 2e, inset).

Chintari Zone
The Chintari Zone stretches for ∼20 km along an NW-SE trend between the Rio Salado and Rio Sama (B in Figure 1). The fault trace is poorly exposed in this zone due to extensive incision from southwest draining rivers. However, fault scarps that follow the projected trace of the IFS are occasionally found within the upper reaches of tributaries. We mapped 6 km of continuous scarps cutting alluvial gravels and late Cretaceous-Paleogene volcaniclastic rocks (INGEMMET, 2017) that run along the southern margin of the Chintari river valley (Figure 3a). In many places, northward draining streams run parallel to shutter ridges, incising the fault footwall as drainage becomes rerouted along the fault, accentuating the topographic expression of the scarps (Figure 3b).
We extracted nine topographic profiles perpendicular to the fault from a high-resolution DEM of the lower 2 km of the Chintari river valley (Figure 3a). The profile locations are shown in Figure 3b and the distribution of vertical scarp heights calculated from these profiles are shown in Figure 3c. The raw topographic profiles are shown in Supporting Information. Within the Chintari zone, the scarp heights vary between 1.9 ± 0.1 and 12.5 ± 0.1 m, with a mean height of 3.9 ± 0.1 m and a median height of 2.  The uncertainty of each measurement is ∼10 cm, therefore is smaller than the symbol (see Table S1). The mean (orange line) and median (blue line) of all measurements are also shown.
Near the confluence between the Rio Curibaya and the Rio Chintari, we identified a section of ∼2 m high fault scarps cutting consolidated, poorly sorted alluvial gravels with pebble to boulder-sized clasts. The fault plane had a highly variable orientation of 249°-281°/46°-71° with the slickenlines pitching between 55° and 85° toward the north or northeast, suggesting the scarps were formed due to reverse slip with a significant left-lateral component. The variability in the fault plane measurements resulted from the irregular shape of the fault plane in the coarse alluvial deposits. Further along-strike, late Cretaceous volcaniclastic rocks to the northeast of the fault trace have been thrusted over the young alluvial gravels (Figures 4b  and 4c), confirming that the IFS has recently accommodated transpression with a left-lateral component of slip in the Chintari zone.

Tauja Zone
The Tauja Zone lies southeast of Chintari and spans a 30 km section of the IFS between the Rio Sama and the Rio Caplina (C in Figure 1). The fault trace in this region splits into two subparallel strands-a northern and southern strand (  In (b) yellow arrows indicate an old, single-track road that skirts the Western Cordillera from north to south and links small buildings and irrigation canals. The black-dashed circle highlights buildings that could correspond to the Chiribaya period, based on the type and ceramics found. As with previous figures, the light blue arrows indicate the direction of streamflow and the black bars indicate the extracted topography profiles with corresponding numbering. The camera symbols indicate the direction of the photos shown in Figure 6. (e) Scarp heights measured from the 28 faultperpendicular profiles taken across the high-resolution DEM, with mean and median highlighted. DEM, Digital Elevation Models. and Paleozoic metamorphic basement. In this region, the IFS had previously been mapped as a reverse fault dipping toward the southwest (INGEMMET, 2017).
We extracted 28 topographic profiles perpendicular to the fault scarps from a high-resolution DEM of the Tauja Zone. The profile locations are indicated in Figures 5b-5d. The scarp heights vary between 0.7 ± 0.1 and 12.0 ± 0.1 m (Figure 5e), with a mean height of 3.1 ± 0.1 m and a median height of 2.3 ± 0.1 m. Similar to the Chintari Zone, most of the scarp heights cluster between 1 and 3 m, while a subset of scarps (24, 25, 26, 27, 28, and 37) are up to 12-m high (Figure 5e). The sense of motion across the Tauja fault scarps is the same as at Chintari and Toquepala, with the north-eastern fault block being uplifted relative to the southwestern block.
Bedrock scarps within the Tauja Zone are typically formed in the dark-gray volcaniclastic rocks that have been thrusted over Holocene alluvial gravels. The ∼2-m high bedrock scarps have a strike of 251°-275°, a dip of 42°-45° (Figures 6c-6h) and contain slickenlines that pitch 57°-67° toward the northeast (Figures 6a  and 6b, inset), indicating that the most recent sense of slip in the Tauja Zone was reverse motion on a north-dipping to northeast-dipping fault with a significant component of left-lateral slip.
Notably the bedrock scarp along this section of the fault often lies at the foot of a taller, rounded scarp (Figures 6e and 6f). In many areas, the uplifted north-eastern block is covered with a thin layer of unconsolidated alluvial gravel and sand which has collapsed over the bedrock scarp, forming a southwest-facing talus slope (Figures 6e and 6g). The southwest-facing slope has previously been interpreted to reflect the fault dip, leading Macharé et al. (2003) to suggest that the most recent sense of slip on the IFS in this region was normal on a southwest-dipping fault. We found that, where the fault plane is well-exposed in stream gullies, it dips toward the northeast (e.g., Figure 6h), confirming that the most recent sense of slip on the IFS in the Tauja Zone was in fact reverse.
A second strand of the IFS has previously been identified in the Tauja Zone within the Rio Sama Valley, between Tarata and Chipispaya Villages (Macharé et al., 2003) (Figure 1, marked "Eastern Toquepala Segment"). The morphological expression of this northern strand is unclear due to extensive fluvial erosion within this high-relief region, but a cross-section through the fault is well-exposed within the banks of the Rio Sama (discussed further in Section 4.2). The exposed fault plane contains slickenlines pitching at 29° toward the ENE on a plane with orientation 280°-292°/37°-53°, indicating the northern strand of the IFS accommodates predominantly left-lateral shear with a minor component of shortening. Notably the sense of motion on this strand of the fault has a much larger strike-slip component than at Toquepala, Chintari, and the southern Tauja Zone, despite the strike and dip of the fault being similar to the other zones. The reasons for the difference in slip vector are unknown.

Santa Elena Zone
The Santa Elena Zone is located at the southeastern end of the traceable fault scarps along the IFS, between the Rio Caplina and the village of Aguas Calientes at the foot of Cerro Tacora (D in Figure 1). In this region, we identified 9 km of subparallel fault scarps that run along the southern side of an E-W striking valley, with north-draining streams having scoured the scarp hanging-walls making the fault clear in optical satellite imagery and DEM relief maps (Figure 7). The IFS is also clear in geological maps of the Santa Elena Zone, where it juxtaposes late Eocene granodiorites northeast of the fault trace against a conformable sequence of Jurassic carbonates southwest of the fault. Near Cerro Santa Elena, the IFS cuts through younger late Cretaceous to Miocene volcaniclastic rocks (INGEMMET, 2017). Throughout this region, the IFS is mapped as a southwest-dipping reverse fault.  scarps have a mean height of 2.3 m, and they mostly cluster between 2 and 4-m high (Figure 8). Some localities preserve taller scarps that reach up to ∼14-m high (Figure 8). The taller scarps typically have rounded crests with a fresh, steeply dipping bedrock scarp at their base that offsets the young, unconsolidated alluvial deposits at the surface.
All of the bedrock scarps preserve a sense of slip in the last scarp-forming event of left-lateral transpression on a 40°-70° dipping fault that dips toward the north or northeast beneath the Western Cordillera. We found no evidence for the southwest-dipping fault that appears on most geological maps of the region. The mean azimuth of the slickenlines on the bedrock scarps is 038°.

Age Constraints on the Incapuquio Fault Scarps
We identified three sites along the IFS suitable for paleoseismological analysis (Figure 1; T1, T2, and T3, respectively). At all of these localities, the fault trace was difficult to access; therefore, we studied natural exposures of the fault with additional hand-dug sections.
The first site (Trench T1-17°34′0.81″S/70°8′50.15ʺW; Figure 5b) consisted of an exposure within the banks of an incised stream that crosses the fault, exposing a 0.5-m wide, 1.5 m deep cross-section of the fault (Figures 9a and 9b). Within the section, we identified a colluvial wedge consisting of a poorly sorted mixture of light-gray sand and silt with angular pebbles. The wedge fill was distinct from the reddish-brown volcaniclastic bedrock in the fault hanging-wall, and the light-gray, coarse-grained alluvial deposits containing rounded pebble to cobble-sized clasts in the fault footwall. We found no evidence of foliation or clast alignment within the colluvial wedge that might indicate the material had been sheared since its formation. A ∼40 cm thick horizontal, light-gray unconsolidated deposit covered the top of the colluvial wedge, and was not offset or deformed by the fault (Figure 9b We collected three samples from the exposed section for dating (sample details are given in Table 1). A dead root from 20 cm above the base of the colluvial wedge returned a radiocarbon age of 1338-1391 CE (sample: GA50-18-05). Root material from the base and top of the unconsolidated deposit that caps the colluvial wedge returned radiocarbon ages of 1726-1806 CE (sample: GA50-18-07) and 1806 CE-today (sample: GA50-18-06), respectively (Figure 9b). Tephra found within the shallower parts of the colluvial wedge may be from the VEI 6 Plinian eruption of the nearby Huaynaputina volcano in 1600 CE, which generated widespread ash fall within the region and is a common feature of the shallow stratigraphy (Thouret et al., 1999).
When combined together, the radiocarbon dates and the tephra chronology indicate that the colluvial wedge at T1 formed prior to ∼1600 CE. However, the lower bound on the timing of scarp formation is less BENAVENTE ET AL.  (c) Photograph of dark-gray Paleozoic rocks thrust over lighter Holocene deposits at trench site T2. Dead roots used to date the timing of the last slip event in this location were collected from the region of the black-dashed rectangle, which is annotated in Figure 9D. The inset shows detail of the slickenlines on the fault plane in red (see Table S2 and Figure S2). (d) Detailed view of Paleozoic rocks on Holocene gravel with sample locations and with their respective ages (see Table 1). (e) Sketch of the stratigraphy in trench T3. The location and ages of the radiocarbon samples are also shown. The detailed description of the stratigraphic units can be found in Figure S3, as well as the original photomosaic of the trench wall.
certain. The root material at the base of the colluvial wedge was presumably transported along the surface and into the wedge sometime after the surface rupture formed. The root itself may also have had some residence time on the surface, prior to being transported into the wedge, which may have been large given the hyperarid conditions. However, the residence time can't be greater than ∼200-300 years if the base of the wedge must predate 1600 CEand the plant died in ∼1338-1391 CE. Therefore, we suggest the surface ruptures formed around ∼1200-1500 CE, and were infilled with colluvium and tephra between ∼1200-1500 CEand 1600 CE, placing a bound on the timing of the scarp-forming earthquake at this locality.
The second trench site (Trench T2-17°34′22.64″S/70°8′14.78″W; Figure 5b) is another natural exposure in a stream gulley that crosses the fault. The fault trace is marked at the surface by thrusting of pale-brown volcaniclastic bedrock onto light-gray alluvial deposits and recent soils (Figure 9c). Along the margins of the gulley, we identified a 10 cm thick gouge that separates the overthrusting bedrock from the underlying alluvial deposit. The gouge was predominantly formed of material from the underlying alluvium. We collected three dead plant roots from within the gouge (Figures 9d and S1). Radiocarbon dating of these roots yielded relatively consistent ages of 1220-1296 CE (GA50-18-08), 1394-1446 CE (GA50-18-09), and 1310-1424 CE (GA50-18-10) (see Table 1).
If the roots were originally part of the alluvial fan surface, and were incorporated into the fault gouge through shearing during a surface-rupturing earthquake, then the roots should predate the earthquake, and the earthquake must have occurred after 1220-1446 CE. The bound on the timing of the surface-rupturing earthquake at trench site T2 (>1220-1446 CE) overlaps with the bound placed on the colluvial wedge formation at trench site T1 (1200-1500 CE). However, we treat these inferences with caution, as the relative ages of the roots and the fault gouge remain unconstrained.
The third site (Trench T3-17°29′53.77″S/70°11′41.12″W; Figure 1) is located southwest of the town of Tarata, and corresponds to the northern strand of the IFS in the Tauja Zone that was originally mapped by Macharé et al. (2003). At this locality the stratigraphy beneath the fault scarp is exceptionally exposed in a river-cut eroded by the Sama River ( Figure S2). We excavated and cleaned the outcrop on the northwest bank forming a 9 m long and 3 m deep exposure (Figure 9e). The deposits in the river bank contain at least five continuous, subhorizontal peat units between 1 and 10 cm thick, some of which are clearly truncated against the fault trace. The peat layers are rich in organic carbon and permit accurate dating of the stratigraphy at this locality.
Radiocarbon ages from five peat samples, three from the footwall, and two from the hanging-wall, place a clear chronology on the stratigraphy and faulting in the trench (Figures 9e and S3). The P-I and P-II peat samples at the lowest stratigraphic level in the trench footwall yielded ages of 761-882 CE (GA50-19-45-T) BENAVENTE ET AL.  and 982-1051 CE (GA50-19-33-T). These dates correlate with two major flood events within the region dated around ∼700 CE (Magilligan & Goldstein, 2001) and ∼1000 CE (Ortlieb & Macharé, 1993). The P-III and P-IV levels sit higher in the stratigraphy (Figure 9e), and both can be seen to truncate against the fault plane. These two layers have similar stratigraphic characteristics and returned similar, but not equivalent, radiocarbon ages of 1225-1310 CE(GA50-19-34-T) and 1290-1392 CE(GA50-19-36-T), which requires that the fault formed sometime after 1290-1392 CE. If the peat layers P-III and P-IV represent the same stratigraphic level, then they are offset vertically by 0.9 m. Given the highly oblique 29° pitch of slickenlines on the N280°/45°E fault plane exposed nearby, the net slip required to generate this vertical offset is 2.8 m, similar to the net slip required to form the 2-4 m scarp heights seen all along the fault trace described in Section 4.1. A final peat layer P-V that caps the fault, and therefore postdates faulting, yielded a radiocarbon age of 1434-1506 CE (GA50-19-37-T). The high-fidelity chronology afforded by the peat stratigraphy at trench site T3 indicates that slip occurred on this strand of the IFS after 1290-1390 CE, but before 1434-1506 CE.
Using the methodology proposed by Lienkaemper and Ramsey (2009), we built an OxCal stratigraphic model (Bronk Ramsey, 2009) to derive the most probable age of the fault slip event preserved in the stratigraphy at T3 based on the probability density functions of the peat layer radiocarbon ages ( Figure 10). We find that fault slip most likely occurred between 1400 and 1440 CE, presumably in a surface-rupturing earthquake. The timing of this earthquake predates the written record of seismicity in the region, which began in ∼1500 CE following the Spanish colonization of Perú (Silgado, 1978).

Recent Seismicity Near the Incapuquio Fault System
Two moderate-magnitude earthquakes with epicenters near the IFS have been recorded by the Global Seismic Network: one event on August 26, 2003 and a second event on the November 20, 2006. A third event occurred offshore the forearc on May 06, 2010. We used teleseismic body-waveform modeling to study the mechanisms and centroid depths of these earthquakes, and compare their kinematics with the recent sense of slip on the IFS.
The August 2003 M w 5.7 earthquake (Event 1) has compressional P wave first motions on vertical-component seismograms at all back-azimuths around the focal sphere suggesting a reverse-faulting mechanism ( Figure S4a). The orientation of the nodal planes is well-constrained by changes in the polarity of the S and sS waves on the transverse seismograms requiring a strike of 220°-250°, a dip of 42°-52°, and a rake of 100°-120°. Notably the strike of both nodal planes is significantly different to the strike of the IFS (∼280°-290°) and the possible earthquake slip vectors (NNW or SE) are strongly oblique to the slip vector of slickenlines we have measured from the IFS (NE). The earthquake centroid depth is well-constrained to be 30-34 km by the relative travel time between the pP, sP, and sS phases with the direct phases.
The November 2006 M w 5.5 earthquake (Event 2) has clear compressional arrivals on the broadband vertical-component seismograms at all back-azimuths around the focal sphere, indicating this was also a reverse-faulting event ( Figure S4b). Amplitude variations in the direct P-waves require that one of the nodal planes crosses reasonably close on the focal sphere to the stations with azimuths south of the earthquake epicenter. The transverse seismograms are particularly noisy due to the size of the earthquake, but still provide some constraint on the nodal plane orientations due to polarity changes in the direct S wave, requiring the strike to be 220°-260°, the dip to be 25°-40°, and the rake to be 40°-90°. As with the 2003 event, both nodal planes are distinctly different to the strike of the IFS (∼280°-290°) and the ∼N-S slip vector is oblique to that of the IFS (NE). The centroid depth for this earthquake is well-constrained by the differential travel time between the direct and depth phases to be 37-41 km.
The final event in May 2010 (Event 3) ruptured the forearc offshore. This earthquake was an M w 6.2 reverse-faulting earthquake and was well-recorded by stations surrounding the earthquake epicenter (Sup-portingInformation Figure S4c). Compressional P-wave arrivals at stations of all back-azimuths, and a clear change in the S-wave polarity around the focal sphere, require a fault with a strike of 5°-30°, a dip of 35°-45°, and a rake of 95°-125°. Notably the east-dipping nodal plane has a steep dip compared to the east-dipping subduction megathrust earthquakes that commonly occur beneath the forearc. The centroid depth is well-constrained to be 38-42 km. Figure 10. Compilation of the normalized probability density functions of radiocarbon ages from the three trenches discussed in the text and from sediments that date the major flood events in the region based on radiocarbon dating by Magilligan and Goldstein (2001) and Ortlieb and Macharé et al. (1993). The peat layers P-III and P-IV that were offset in trench T3 must predate the scarp-forming earthquake, while the layer P-V must postdate the scarp-forming earthquake, placing the most likely earthquake timing at ∼1400-1440 CE (shown by the vertical light-gray band).

Interpretation of the Fault Scarp Observations
Fault scarps typically form in surface-rupturing earthquakes, with the along-strike length of the scarps, and the magnitude of the displacement, being related to the earthquake magnitude through empirical scaling relationships derived from modern earthquakes (e.g., Wells & Coppersmith, 1994). A major challenge in applying these scaling relationships to ancient fault scarps is differentiating between scarps that were formed in several different earthquakes, and those formed in a single earthquake.
One way to test whether the IFS scarps were formed in a single earthquake is to study the along-strike distribution of the scarp morphologies and kinematics. We found that the scarps have a consistent ∼E-W to ∼NW-SE strike, with occasional <1 km left steps in the fault trace. Our scarp height measurements range between 1 and 14 m, though 80% of the scarp heights are between 2 and 4 m. The highest fault scarps are typically heavily rounded along their crests and have a ∼2-m high, fresh "scarplet" at their foot. These scarplets are formed in consolidated bedrock, offset the most recent alluvial deposits at the surface, and are of similar height to the mean scarp height (2.3 m) along the IFS.
Slickenlines measured along exposed sections of the fault have similar geometry with a fault plane strike of 245°-290°, a dip of 40°-70°, and a slip vector pitching at 50°-70° toward the northeast. The mean azimuth of the fault slip vector is 038°. Notably, the macroscopic trace of the fault in the satellite data (∼300°) has a slightly different strike to our measurements on exposed fault surfaces, as three out of the four measurement sites are along left steps in the fault trace. If the fault slip vector azimuth is consistent along-strike, then left steps in the fault should have a slightly smaller dip-slip component to the slip vector, creating smaller, less prominent scarps. However, we find that the ∼30° variability in the fault strike does not contribute significantly to along-strike changes in the scarp heights ( Figure 8).
The simplest interpretation of the scarp morphologies and height distribution is that the highest, most rounded scarps are "compound scarps" formed in several surface-rupturing earthquakes. Similar structures have been observed along other faults within the forearc (e.g., the Purgatorio Fault; see Benavente et al., 2017). In some areas, the footwall of the compound scarps has been incised by streams that drain along the fault trace, accentuating their height. The 2-4-m high, fresh scarplet. along the majority of the IFS may therefore have been formed in the most recent surface-rupturing earthquake(s). If the scarps were formed in a single earthquake, we might expect the scarp heights to increase toward the center of the scarp trace (e.g., Middleton et al., 2017). However, we did not find any along-strike trend in the scarp heights ( Figure 8).
A second test of whether the IFS scarps formed in a single earthquake, or several earthquakes, is to date the most recent scarp-forming event at different points along-strike. Natural exposures of the stratigraphy at three localities along a 40 km stretch of the IFS point to the most recent surface rupture occurring within a ∼300-year period between ∼1200 CEand ∼1500 CE. At one locality (trench site T3), the timing of most recent surface rupture is well-constrained to be between ∼1400 and ∼1440 CE. Therefore, it is within the constraints of our dating that the IFS fault scarps formed in a single earthquake, with the most probable timing being between 1400 and 1440 CE. Dating the fault scarps near the ends of the mapped fault trace at Toquepala and Santa Elena would provide an additional test of this hypothesis.
In order to form the 2-4 m high scarps, the net surface slip in the earthquake would have been at least 2.2-4.5 m given the fault dip of 40°-70° and rake of 50°-70°. Using a mean scarp height of 2.3 m, a mean dip of 50°, and a mean rake of 60°, would require a mean fault slip of 3.4 m. It is possible that the true fault dip is toward the larger end of our measurements, as the fault trace is highly linear and there is the possibility for a component of collapse of the fault plane following scarp formation. If we have underestimated the fault dip, then net slip required to form the scarps would be smaller, with a minimum of ∼2 m. The cumulative length of the proposed fault rupture is ∼100 km, based on the total length of the fault scarps traced between Moquegua and Aguas Calientes. The corresponding fault slip-to-length ratio is ∼3 × 10 −5 , and the estimated moment magnitude of the last scarp-forming earthquake would be M w 7.5 and M w 7.4 according to the empirical scaling laws of Wells and Coppersmith (1994) and Stirling et al. (2013), respectively. Adding in the 30 km long northern segment of IFS in the Tauja Zone (Macharé et al., 2003) with 2.9 m of slip would increase the magnitude of this earthquake to M w 7.7 and M w 7.5 using the same empirical scaling relationships. Scatter in the scaling relationships would add an uncertainty of 0.1-0.2 magnitude units to these values.
The paleo-magnitude calculations presented above assume that all scarps that have been roughly dated to ∼1200-1500 CE ruptured in a single earthquake. Although we consider a single earthquake to be the simplest explanation of the along-strike similarity in scarp heights, scarp morphology, and the fault slipto-length ratio (Wells & Coppersmith 1994), it is also within the constraints of our dating and fault scarp observations that the fault ruptured in a series of smaller subevents within a ∼300-years time period. The magnitudes of these subevents would of course be smaller than ∼M w 7.4-7.7, with their size depending on the level of segmentation. However, modern earthquakes that generate scarps of at least ∼2 m are consistently M w 6.5 or larger (Wells & Coppersmith 1994). Earthquakes of this size still pose a significant hazard to communities that live within ∼50-100 km of the fault.
We cannot estimate any recurrence period or long-term slip rate on the IFS from our observations, as there are no correlatable surfaces offset across the taller cumulative scarps that could be reliably dated. Therefore, the regularity of large earthquakes on the IFS remains unknown. Although, given the limited size of the cumulative scarps (10s of meters), the extremely slow erosion rates in the forearc, and the lack of significant kilometer-scale relief change across the IFS, it is likely ruptures on the forearc faults are relatively irregular, and certainly much less frequent than the offshore megathrust earthquakes, which repeat every 150-300 years.

Deformation in the South Peruvian Forearc
There is growing evidence that Quaternary faulting within the forearc and Western Cordillera of South Perú has variable kinematics in space and time (Audin et al., 2006;Benavente et al., 2017;Binnie et al., 2020;Hall et al., 2012;Loveless et al., 2009;this study). Our observations from the IFS demonstrate that the boundary between the forearc and Western Cordillera failed in at least one large earthquake accommodating left-lateral transpression around 600-years ago (Figure 11a). The ESE-WNW striking Purgatorio and Sama Calientes Faults that lie within the center of the forearc have different kinematics to the IFS. Slickenlines on the near-vertical exposures of the Purgatorio fault plunge 50°-60°, and streams and ridge crests are offset laterally across the fault, indicating late Quaternary right-lateral transpression on this structure (Benavente et al., 2017). Cosmogenic radionuclide dating of surfaces offset by the Purgatorio Fault demonstrate that it has ruptured in the last thousand years (Benavente et al., 2017;Hall et al., 2012). No slickenline measurements have been made on Sama Calientes Fault, though Neogene rocks to the north of the fault trace have been heavily folded and uplifted suggesting it accommodates a component of ∼NNE-SSW shortening. Together the Incapuquio, Purgatorio, and Sama Calientes Faults form a conjugate array that may accommodate shortening of the forearc roughly perpendicular to the Perú-Chile trench (Figure 11a).
In contrast, NE-SW striking normal faults (Audin et al., 2006) and surface cracks (Loveless et al., 2009) that lie along the coast at Ilo have accommodated NW-SE trench-parallel extension in the Quaternary (Figure 11a). While south of the Arica Bend, near Pisagua, E-W striking reverse faults accommodate ∼N-S trench-parallel shortening (Allmendinger et al., 2005) and ruptured in foreshocks preceding the 2014 Pisagua earthquake (González et al., 2015).
The recent seismicity around the IFS consists of reverse-faulting mechanisms with ∼E-W to ∼NE-SW striking nodal planes with ∼N-S to ∼NW-SE trending slip vectors (see Figures 11b and Figures S4a and S4b). Given the mechanisms and depths of the earthquakes, they probably ruptured the base of the brittle part of the forearc wedge above the subduction interface, as moderate-magnitude earthquakes within the subducting slab have depths >50 km east of the coastline and uniformly accommodate ∼E-W extension throughout the Arica Bend (Figure 11b), while earthquakes on the subduction interface have low-angle thrust mechanisms with ENE-WSW slip vectors parallel to Nazca-South America relative plate motion. Notably, the possible slip vectors of the two forearc earthquakes studied in Section 4.3 (EV1 and EV2 in Figure 11b) are strongly oblique to the slip vectors measured on the bedrock fault scarps of the Incapuquio Fault, and are reversed relative to the sense of Quaternary extension across the normal faults at Ilo.
The variability in the strikes and slip vectors of the faults and earthquakes near the IFS demonstrate that the kinematics of permanent deformation in the forearc is complex, with evidence for trench-parallel extension, trench-parallel shortening, and trench-perpendicular shortening in close proximity. Similar variability occurs in the pattern of incremental strain within the forearc captured by GPS during the interseismic and coplus postseismic period of the 2001 M w 8.4 Arequipa earthquake (Figures 11c and 11d). Prior to 2001, the forearc was subject to predominantly ENE-WSW shortening with a minor component of NNW-SSE extension (Allmendinger et al., 2007)  earthquake the incremental strain field reversed, with coseismic and postseismic slip on the megathrust leading to NE-SW extension and NW-SE shortening of the forearc (Figure 11d). Over the same time period, normal-faulting earthquakes in the Altiplano led to NE-SW extension parallel to the principle axis of the coseismic extension (Figure 11b), and two earthquakes ruptured deep in the forearc leading to shortening subparallel to the principle axis of coseismic shortening (Figure 11b).
The GPS and faulting observations are consistent with the paradigm that permanent deformation in the overriding plate is a result of a small fraction of the interseismic strain becoming "sealed in" by slip on forearc faults between megathrust earthquakes (e.g., Allmendinger et al., 2005;Loveless et al., 2010), but also requires that a small fraction of the coseismic and postseismic strain captured by GPS is accommodated by forearc faulting (e.g., Hicks & Rietbrock, 2015). What fraction of the GPS-measured incremental strain will be preserved as permanent deformation in the forearc remains poorly understood, and may well be different between the interseismic, coseismic, and postseismic periods, but could be up to 10% (e.g., Baker et al., 2013).

Implications for Seismic Hazard
The similarity between the shortening direction across the Incapuquio, Purgatorio, and Sama Calientes Faults, and the principle axis of interseismic shortening in the forearc, suggests that these faults are being loaded toward failure by the compression and torqueing of the overriding plate between megathrust earthquakes (e.g., Loveless et al., 2010). As forearc faults rupture less frequently than the megathrust, elastic strain accumulates within the forearc over several megathrust earthquake cycles. The mechanisms that allow enough strain to accumulate to rupture forearc faults without rupturing the megathrust, presumably the weakest component of subduction system, remain unknown. Nevertheless, faults such as the IFS are more likely to fail after a long-period without a megathrust earthquake, as late in the megathrust earthquake cycle the forearc will be highly compressed, promoting slip on the reverse faults in the forearc (e.g., Loveless et al., 2010). Forearc faults that accommodate deformation with different kinematics to the interseismic strain field, such as the trench-perpendicular reverse faults near the IFS, are not being loaded toward failure between megathrust earthquakes. These faults are more likely to rupture during and immediately after megathrust earthquakes, due to the incremental strain field change related to slip on the subduction interface.
An important implication of this transient compression and stretching of the forearc by the megathrust earthquake cycle is that the seismic hazard posed by any particular forearc fault varies over decadal time scales. Linking the decadal deformation measured by surface GPS to the incremental stress changes on forearc faults could offer a quantitative insight into changes in the possibility of fault rupture (e.g., Loveless et al., 2010;Mildon et al., 2019), but requires improved constraints on the rheological structure of the forearc and its faults. Our waveform-modeling results, in addition to microseismic surveys along the Andean coastline (e.g., Gonzalez et al., 2015), suggest that the forearc can accumulate significant elastic strain and generate earthquakes at depths of at least 40 km, far deeper than the seismicity within the Altiplano, presumably because of the cooling effect of the underthrusting oceanic plate. However, whether the region below 40 km and above the subducting slab deforms seismically or aseismically remains unclear, but will have significant implications for how permanent strain may accumulate in the forearc (e.g., Luo et al., 2019). A thick elastic crust within the forearc may also lead to large but infrequent earthquakes, as the available seismogenic area of fault planes is large. Given how intermittently preserved the fault scarps are along the IFS, despite the scarps occurring in an extremely arid and sparsely populated region, then infrequent but large magnitude earthquakes within the thick brittle crust of other subduction zone forearcs where erosion rates are far higher may be rarely preserved, and therefore overlooked as a source of seismic hazard (e.g., Astudillo-Sotomayor et al., 2021).

Collapse of the Chiribaya Civilization Around the 1400-1440 CE Earthquake
A number of different civilizations have thrived and collapsed in the hyperarid forearc around the Incapuquio Fault over the last ∼4 kyr (Goldstein & Magilligan, 2011). These civilizations based their fresh water supply on sophisticated irrigation systems and canals that drained water from the Western Cordillera into the arid forearc (e.g., Zaro et al., 2013). Consequently, it is often argued that exceptional rainfall, or destruction of the irrigation systems, was the main driver of the decline of ancient populations in the region. For example, Keefer et al. (2003) and Wells (1990) suggested that large floods associated with El Niño Southern Oscillation in ∼1330 CE (known as the Miraflores flood) may have contributed to a ∼70-80% decline in the population of the Chiribaya civilization of the Moquegua Valley (Figure 1) between 1360 and 1400 CE. However, Clement and Moseley (1991) argue that a famine in the nearby Carrizal valley around 1360-1400 CE is incompatible with a mega-Niño event. In addition, Zaro et al. (2013) argue that the Chiribayas adapted to the variable hydrological specificities of the Andean rivers, and that the Miraflores event alone may not be the trigger of the Chiribayas collapse. As no evidence for shallow-crustal earthquakes have been previously reported in this region, earthquake-related damage or deaths have not been considered as a potential influence on the collapse of the Chiribaya civilization.
During our field reconnaissance, we found collapsed structures such as terraces, storehouses, roads, aqueducts, and corrals in close proximity to the fault scarps on the IFS. In the Tauja Zone, we also identified buildings cut by the main scarp of the IFS with a net displacement of 1.6 m ( Figure 12). Although we do not know the exact age of these buildings, we hypothesize that they are from the Chiribaya period, based on the type of buildings, ceramics (Reycraft, 1998) Figure 5b). Given the style of the buildings, and the fact they are cut by the fault scarp, we hypothesize that they correspond to the Chiribaya civilization (Reycraft, 1998). (b) Field photograph of the buildings that cut by the fault scarp, which are now destroyed. (c) Damaged ceramics found within the confines of the buildings. Detailed archeological surveys of this site may be able to more accurately distinguish its provenance. IFS, Incapuquio Fault System. these structures so must postdate them. Although a provocative proposal, it is possible that infrastructure damage generated by a major earthquake along the IFS around 1400-1440 CE contributed to the decline of the Chiribayas. We stress that further archeological and geological investigation is required to test this hypothesis.

Conclusions
We have presented evidence that 2-4 m high fault scarps were formed by a transpressional earthquake that ruptured 100 km of the southern segment of the Incapuquio Fault System in the early 15th century (∼1400-1440 CE). Empirical scaling relationships suggest the magnitude of this earthquake could have been M w 7.4-7.7 with at least 2-3 m of net slip. Recurrence of similar events pose a significant hazard to the population living in the forearc region of Perú. The timing of this earthquake coincides with the collapse of the population of the Chiribaya civilization between ∼1360 and 1400 CE, and we present evidence of damaged and offset buildings along the fault trace that may be of Chiribayas age. The kinematics of the Incapuquio Fault suggest that it ruptures following the accumulation of compressional elastic strain in the forearc during the interseismic period of the megathrust earthquake cycle. Many other active faults within the forearc with different strikes and slip vectors to the Incapuquio Fault imply that permanent deformation in the region is complex, probably due to the time-varying stress field within the overriding plate caused by megathrust earthquakes.

Data Availability Statement
The drone DEMs are freely available through OpenTopography (www.opentopography.org) at https://doi. org/10.5069/G9JW8C3S (Chintari Zone), https://doi.org/10.5069/G9TD9VH0 (Tauja Zone) and https://doi. org/10.5069/G9PN93T3 (Santa Elena Zone). All other data used in this study are provided in the text. Clare College, Cambridge. X. Robert was supported by the Institut de Recherche pour le Développement (IRD) through BQR-SUD 2018 funds from ISTerre Lab and expatriation program. The supporting information contains figures of the topography profiles across the fault scarp and further information about the body-waveform modeling. We acknowledge the work of the Associate Editor Laura Giambiagi, and two anonymous reviewers for their constructive comments.