Paleoseismologic evidence for the relatively regular recurrence of infrequent, large-magnitude earthquakes on the eastern North Anatolian fault at Yaylabeli, Turkey

Over the past several decades, paleoseismologic data have provided key insights and raised provocative questions about the repeatability and regularity of earthquakes on many of the world's major continental fault systems (e.g., Sieh, 1978; Schwartz and Coppersmith, 1984; Sieh et al., 1989; Marco et al., 1996; Rockwell et al., 2009; Dawson et al., 2003; Friedrich et al., 2003; Okumura et al., 2003; Weldon et al., 2004; Hartleb et al., 2006; Dolan et al., 2007). For example, is earthquake occurrence in time and space a relatively regular process, or do earthquakes occur more randomly? Are earthquake clusters, which have recently been observed on both single faults and within regional fault networks (e.g., Barka, 1992; Marco et al., 1996; Stein et al., 1997; Rockwell et al., 2000; Friedrich et al., 2003; Dolan et al., 2007), a common phenomenon? Do different faults behave differently, perhaps in response to issues such as the relative structural complexity of the plate boundary (e.g., Dolan et al., 2007) or temporally variable loading histories (e.g., Friedrich et al., 2003)? While both fascinating and of fundamental importance for seismic hazard assessment and our understanding of the physics of the earthquake process, such discussions have been in general significantly limited by the availability of paleoseismologic observations, particularly from multi-event sites with well-constrained ages. The key to addressing these questions lies in obtaining the most complete paleoseismologic records possible, which in turn requires that we demonstrate repeatability of paleoseismologic observations.

Along a few faults, such as the Wasatch fault in Utah (e.g., Schwartz and Coppersmith, 1984; Friedrich et al., 2003) and the Mojave section of the San Andreas fault in California (Sieh et al., 1989; Weldon et al., 2004), several high-quality paleoseismologic sites occur in close proximity, allowing detailed analysis of spatial and temporal behavior of earthquakes and assessment of the completeness of the paleoearthquake catalog. However, such sites, particularly high-quality sites with peat stratigraphy and deep-time records, are rare along most faults, precluding repeatability of measurement and limiting our ability to assess the spatial and temporal occurrence of earthquakes and catalog completeness.

In this paper, we exploit a rare situation in which two high-quality, sedimentologically distinct peat sites occur in close proximity along the eastern part of the North Anatolian fault in Turkey. Herein, we report new data from one of these sites, which facilitate comparison of the earthquake record along this stretch of the North Anatolian fault. We use these new data, in combination with previously published paleoseismologic results, including those of a companion study at a nearby site (Hartleb et al., 2006), to generate an updated space-time diagram of earthquake occurrence along the North Anatolian fault over the past 2000 yr. We then discuss these results in light of their implications for long-term patterns of seismic strain release on the North Anatolian fault and for reproducibility and completeness of paleoseismologic records, both along the North Anatolian fault and more generally.

The North Anatolian fault is an arcuate, right-lateral strike-slip fault that extends for 1500 km from the Karlıova triple junction in eastern Turkey westward across northern Turkey and into the Aegean Sea (Fig. 1). Together with the left-lateral East Anatolian fault, the North Anatolian fault accommodates westward motion of the Anatolian block in response to subduction rollback along the Hellenic trench and collision of Arabia with Eurasia (McKenzie, 1972; Şengör, 1979; McClusky et al., 2000; Reilinger et al., 2006). Geologic studies indicate that the slip rate on the North Anatolian fault over the past 10³–10⁵ yr is ∼15–22 mm yr⁻¹ (Hubert et al., 1997; Hubert-Ferrari et al., 2002; Okumura et al., 2003; Kondo et al., 2004; Kozacı et al., 2007, 2009; Pucci et al., 2007; Rockwell et al., 2009), whereas geodetic studies reveal somewhat faster rates of strain accumulation of ∼24–29 mm yr⁻¹(McClusky et al., 2000; Reilinger et al., 2006). Both geologic and geodetic data indicate that almost all of the strain associated with the westward extrusion of the Anatolian block in northern Turkey is accommodated along the North Anatolian fault system. For example, the very small residual site velocities for the geodetically constrained block model of Reilinger et al. (2006) demonstrate that the velocity field of northern Turkey can be accounted for almost completely by elastic strain accumulation along the North Anatolian fault. These results indicate that strain accumulation associated with other faults (e.g., Ovacik fault, Northeast Anatolian fault; Fig. 2) is minor compared to the North Anatolian fault. Moreover, geodetic measurements indicate that internal deformation rates within central Anatolia are <2 mm yr⁻¹ (McClusky et al., 2000; Reilinger et al., 2006), attesting to the notably “block-like” behavior of the Anatolia-Eurasia plate boundary in northern Turkey. Thus, the North Anatolian fault is the dominant fault within a relatively simple plate boundary.

Between 1939 A.D. and 1999 A.D., a 1000 km length of the North Anatolian fault ruptured during a generally westward-propagating series of eight M ≥ 7 earthquakes (Ketin, 1948; Richter, 1958; Allen, 1969; Ambraseys, 1970; Toksöz, et al., 1979; Barka, 1992, 1996, 1999; Ambraseys and Finkel, 1995; Stein et al., 1997; Barka, 1999). This sequence included four very large earthquakes: the 1939 M_w 7.9 Erzincan earthquake, the 1943 M_w 7.7 Tosya earthquake, the 1944 M_w 7.5 Bolu earthquake, and the 1999 M_w 7.5 İzmit earthquake. Stein et al. (1997) have shown that, in general, the westward propagation of this sequence is consistent with a model of stress triggering of each earthquake by earlier events in the sequence.

These observations raise several fundamental questions. First, does the seemingly simple structure of the Anatolia-Eurasia boundary along the North Anatolian fault lead to simple patterns of earthquake occurrence, as witnessed during the twentieth-century sequence? In other words, is the normal mode of earthquake occurrence on the North Anatolian fault characterized by brief periods (decades to a century in duration) of intense earthquake activity during which most or the entire fault ruptures? Or, was the twentieth-century earthquake sequence an anomaly? Does the scarcity of other nearby, major seismic sources along the central and eastern North Anatolian fault lead to simpler patterns of stress evolution? If so, does this simpler loading history result in a more regular recurrence of large earthquakes?

To address these questions, we present results from a paleoseimologic study of the North Anatolian fault from the village of Yaylabeli in north-central Turkey along the 1939 M_w 7.9 surface trace. The Yaylabeli site is only 2 km west of the Çukurçimen trench site of Hartleb et al. (2006), providing an ideal opportunity to compare the results from these two high-quality peat sites and assess the reproducibility and completeness of the paleoseismologic record along this part of the North Anatolian fault.

The Yaylabeli paleoseismic trench site is located ∼65 km west of the city of Erzincan, along the eastern part of the North Anatolian fault (Figs. 1 and 2). The site is an ∼350-m-long, 40- to 60-m-wide, semi-enclosed marshy basin that has formed at a 25-m-wide releasing step in the fault (Figs. 3 and 4). The basin is bounded by a broad, ∼20-m-high shutter ridge to the north, and by 300-m-high peaks to the south. Silt, sand, and pebble gravel are transported off of the southern ridge and are deposited as a small alluvial fan that extends northward into the marsh (GSA Data Repository Fig. DR1¹). There is a gentle down-to-the-east surface gradient at the site, allowing limited water discharge at the east end of the marsh. In the lowest parts of the marshy depression, local areas with standing surface water occur throughout the year, whereas the surface of the rest of the marshy basin is damp during summer and exhibits standing water only during the winter rainy season.

To the east and west of the releasing step, the North Anatolian fault through the site is expressed as a single, geomorphically well-defined strand. The most recent surface rupture at Yaylabeli occurred during the 1939 M_w 7.9 Erzincan earthquake. This earthquake was generated by rupture of an ∼400-km-long section of the eastern North Anatolian fault, from the 5-km-wide extensional stepover in the Erzincan basin westward to the 10-km-wide extensional Niksar stepover, where the rupture extended southwestward along the Ezinepazarı fault (Barka, 1996) (Fig. 2). Surface displacements in the 1939 earthquake ranged up to ∼10 m near our study site (Hartleb et al., 2006). Although no significant scarps or mole track remnants from the 1939 earthquake are preserved at the site, interviews with local eyewitnesses confirmed the presence and location of the surface rupture within the marshy basin. Moreover, our easternmost trench was located ∼10 m west of a dirt road that was offset 9.5 ± 1.5 m during the 1939 earthquake (Hartleb et al., 2006, their fig. 5).

We excavated four fault-perpendicular trenches at Yaylabeli (T1–T4) (Fig. 5). The fault zone was exposed in all four trenches, allowing us to delineate the fault geometry across the site (Fig. DR2 [see ^{footnote 1}]). Trenches T3 and T4, in particular, revealed well-defined event stratigraphy, as well as numerous peat samples for radiocarbon dating.

In order to expose the deepest possible stratigraphic section, trenches T1, T2, and T3 were benched due to shallow groundwater and resulting problems with the stability of the trench walls. After encountering very shallow groundwater in our trench T1 (excavated during the summer of 2004), an ∼300-m-long, fault-parallel drainage trench was excavated to divert the groundwater flowing into the basin in the beginning of summer 2005. This drainage trench successfully reduced groundwater input from the mountains to the south, enabling us to excavate trench T3 to a depth of 4.5 m below ground surface.

Trench T1 was excavated as an exploratory trench to locate the fault and determine the width of the fault zone. We were not able to log this trench in detail due to unstable trench walls associated with a very shallow water table (<0.5 m depth). We did, however, locate the fault in the northern end of the trench, where it is expressed as a <1-m-wide zone of fault gauge and multiple subvertical fault strands (Fig. 6; Figs. DR3 and DR4 [see ^{footnote 1}]). We also recovered peat samples from T1 for stratigraphic correlation purposes.

Trench T2 was excavated during the summer of 2005, 45 m west of T1, and two months after we dug the drainage trench. In trench T2, the fault lies within the heavily altered schist bedrock (now a pale-gray clay) and bedrock-derived soil exposed at the northern end of the trench (Fig. 7; Figs. DR5, DR6, DR7, and DR8 [see ^{footnote 1}]). Because the fault is not located within the marsh deposits, we were not able to identify individual paleoearthquake event horizons, with the exception of an extensive extrusive sand sheet, which provides evidence of a paleoliquefaction event (Figs. DR5 and DR6 [see ^{footnote 1}]). As with T1, we located the fault in T2 and collected peat samples for stratigraphic correlation.

Trench T3 was excavated 28 m to the east of T1, where the fault extends across the active marsh. Trench T3 was a 4.5-m-deep, benched exposure with three 1.5-m-high faces. We were able to map the eastern wall of T3 in detail (Fig. 8; Figs. DR9, DR10, and DR11 [see ^{footnote 1}]). However, despite our drainage trench, groundwater still entered the trench from near the base of the west wall, causing it to cave repeatedly. Consequently, we were unable to log the west wall of T3 in detail.

Trench T4 was excavated in the summer of 2006, 8 m to the east of T3. Trench T4 was excavated as a 90-cm-wide slot trench designed to encompass the entire fault zone and provide reproducibility of the event horizons documented in T3. We exposed the fault zone in T4 and logged both walls in detail (Fig. 9; Figs. DR12 and DR13 [see ^{footnote 1}]).

All trenches were logged on a 1 m × 1 m grid system, which was set arbitrarily for each trench. Locations of specific features in the trenches are described in the following sections according to their trench coordinates, with a horizontal meter measurement starting at the north end of each trench, followed by a depth below an arbitrary “0” datum, which is different for each trench (e.g., m3.7,−2.5 corresponds to a point at station 3.7 meters horizontally and 2.5 meters below the arbitrary level line marked as “0”). Stratigraphic units are numbered to facilitate discussion, with numbers increasing with depth. Peat horizons that extend across the entire trench exposure are labeled as multiples of 10 (e.g., peat 10, 20, 30), whereas a peat unit that is limited in lateral extent is labeled as P15. Individual units that we can confidently correlate between different trenches are labeled with the same numbers in all trenches. All trench locations were surveyed with a laser-range-finding total station.

All four trenches exposed interbedded, fine-grained alluvial units (gravels, sands, and silts) and marsh deposits (peats and organic-rich clays) (Figs. DR14 and DR15 [see ^{footnote 1}]). These stratigraphic relations are continuous throughout the Yaylabeli trench site. We were able to correlate major peat horizons between trenches by their depth, thickness, color, texture, and the results of ¹⁴C dating. Layer thicknesses range from a few millimeters to a few decimeters, with generally sharp stratigraphic contacts associated with the near-complete absence of bioturbation in the marsh deposits due to the perennially wet conditions. Stratigraphic contacts, even where they are gradational, can be determined to within 5 cm. An increased proportion of distal fan deposits was observed in the southern ends of all trenches. These deposits become coarser grained southward toward the fan source. They are composed of coarse-grained sand, pebbly sand, and pebble gravel. Toward the center of the basin, the materials become finer grained. In this area, lacustrine deposits and peat horizons thicken into the trough at the axis of the marshy low. These deposits are composed of clays, silty clays, organic-rich clays, and peats. Toward the northern end of our trenches, the lacustrine deposits and peat horizons onlap and pinch out against the altered bedrock that makes up the shutter ridge to the north.

We radiocarbon (¹⁴C) dated 38 peat samples from the Yaylabeli trenches. All of these samples were pretreated and analyzed at the Center for Accelerator Mass Spectrometry (CAMS) at Lawrence Livermore National Laboratory. We examined the peat samples under a microscope and separated the “flattened,” partially degraded, leafy plant material, which we think represents the youngest, in situ component of each peat, from detrital charcoal (which may be significantly older than the deposit) and roots (which may be significantly younger than the deposit). Following this physical separation, we performed standard acid-alkali-acid (AAA) chemical pretreatments on these selected leafy components of each sample. In addition to radiocarbon dating of the resulting residual carbon, for some samples, we collected the alkali-treated washes in a separate container and dated the resulting humic fractions in order to obtain the youngest possible in situ ages for the peat samples, following the methodology established previously by others at stratigraphically similar peat-rich sites (e.g., Schoute et al., 1981, 1983; Johnson et al., 1990; Fumal et al., 2002). Accelerator mass spectrometry (AMS) radiocarbon dating of the paired AAA-treated fractions and associated humic fractions from the same peat samples yielded similar ages. Specifically, nine out of eleven of the humic fraction ages returned results that are very similar to those of the radiocarbon ages of the AAA-treated peat samples. Two samples (T4-07 and T4-08), however, returned ages for the AAA-treated fractions that are ∼100–200 yr older than those of the associated humic fractions from the same sample, as might be explained as a result of some detrital charcoal left in the residuals. In addition, one humic sample (T5-5-top-humic) yielded a clearly anomalous calibrated date range of 256–530 A.D. relative to the 1480–1651 A.D. date from the AAA-pretreatment of the same sample. Thirty-four out of 38 radiocarbon dating results of these samples revealed ages in correct strati graphic order (Table 1). The remaining four samples (T3-1-top, T3-1-base, T3-6, and T5-5-top-humic) apparently contained significant reworked detrital carbon.

In our descriptions of event ages in the following section, we used OxCal 4.0 (Bronk Ramsey, 1995; Bronk Ramsey, 2001; Reimer et al., 2004) to determine calibrated, calendric ages for each radiocarbon date. In addition to presenting the calendric date range of each sample, we also use OxCal 4.0 to model the distribution of sample ages that are in correct stratigraphic order. This program reweights the age distributions of adjacent layers that overlap by using stratigraphic ordering and incorporating the sample probability density functions to directly determine layer and event ages (e.g., Biasi and Weldon, 1994; Bronk-Ramsey, 1995; Hartleb et al., 2006). Finally, we compared these trimmed age ranges with historical earthquake records to determine if each of our paleoseismologically defined event horizons could plausibly be correlated with a major historical earthquake.

We identified evidence for five surface ruptures in our trenches, referred to as events 1 through 5, from youngest to the oldest. Evidence for these paleoearthquake event horizons includes upward fault terminations, fissure fills, the geometry of growth strata, and liquefaction features. Most of the event horizons were observed in multiple exposures. The predominance of sharp contacts and laterally continuous beds allows us to define the precise stratigraphic level of many of these event horizons. As noted already, most of our event evidence comes from trenches T3 and T4 (Figs. 8 and 9), with additional evidence from trenches T1 and T2 (Figs. 6 and 7). In the following sections, we present structural and stratigraphic evidence, age control, and pertinent historical data for each of these events.

The most recent surface rupture at the Yaylabeli site was the 1939 M ∼8 Erzincan earthquake. This event extends to within a few centimeters of the surface in T1, T2, and T4, and eyewitnesses to the earthquake corroborate the location of the surface rupture exposed in our trenches. Specifically, in trench T1, at least one, and possibly two, fault strands terminate ∼20 cm below the ground surface between m4 and m5 on the western wall, within unit P00 (Fig. 6). In trench T4, event 1 is expressed as a single, well-defined upward fault termination at m7, <5 cm below the ground surface on the eastern wall. On the western wall of trench T4, event 1 is marked by two fault strands that bound a 25-cm-wide, 30-cm-deep fissure fill at ∼m5.5. These fault strands terminate upward within the unit peat P00, as in trench T1. Interestingly, in trench T3, which was located between trenches T2 and T4, no fault strands extend to the surface. Instead, the upper meter of section in T3, encompassing peats P00 and P10, is folded into a syncline-anticline pair (Fig. 8). This young folding appears to reflect warping during the 1939 surface rupture within the releasing step between T1 and T4 (Fig. 5), although the fact that underlying layers do not mimic this near-surface folding is puzzling. Nevertheless, the syncline exposed at m25.4 in T3 was expressed as a topographic swale prior to opening the trench, attesting to both the recency of this deformation and the fact that this folding was not related to excavation of our trenches.

We do not have direct age constraints for the occurrence of event 1. However, based on the proximity to the ground surface of the multiple upward fault terminations, the topographic expression of the near-surface syncline in T3, and the eyewitness descriptions, we conclude that event 1 is the historical 1939 Erzincan earthquake.

Event 2 is defined by upward fault terminations in trench T3 and T4, at m27.47 and m6, respectively, and by contorted bedding of peat 20 in trenches T2 and T4, which we interpret as evidence for liquefaction of this stratigraphic interval. In trench T3, a clearly defined fault strand at m27.47 offsets peat P15 but terminates upward at the top of peat P15, beneath peat P10; P10 and the underlying clay layers above P15 are not offset. At m6 on the western wall of trench T4, a fault strand offsets peat P20, which exhibits liquefaction features, and terminates below peat P10. On the eastern wall of T4, however, a clear fault strand, distinct from the strand that ruptured in event 1 (1939), could not be observed. In trench T2, we interpret a laterally extensive sheet sand at this stratigraphic level as an extrusive sand blow deposit. At m6.5, this sand can be traced down into an apparent subvertical feeder dike that crosscuts stratigraphy. Although this extrusive sand deposit occurs at approximately the same stratigraphic level as the E2 event horizon observed in our other trenches, we cannot unequivocally correlate the extrusive sand with event E2 because peat P15 is not present in trench T2. Moreover, the sediment accumulation rate in T2 is slower (and probably less continuous) than in trenches T3 and T4, and we therefore place more emphasis on event stratigraphy observed in those trenches.

We dated samples T3-2, T4-06, and T4-11 from peat 20 to provide a maximum age constraint on the timing of event 2. Dating of peat sample T3-2 from trench T3 returned a calibrated, calendric age range of 995 A.D. to 1181 A.D. Dating of T4-06 and T4-11 from peat P20 in trench T4 returned calendric age ranges of 997 A.D. to 1156 A.D., and 986 A.D. to 1154 A.D., respectively. The similarity of these dates confirms our stratigraphic correlation of peat P20 between trenches T3 and T4.

Two calendric, calibrated dates from the lower part of peat 10 (sample T5-5-base), which is not cut by event 2 strands, provide a robust minimum age for event 2. These sample ages are in good agreement, with a 1318–1439 A.D. date from the AAA-pretreated component from sample T5-5 base, and a 1409–1487 A.D. date from the humic component of the same peat sample (T5-5 base-humic). Thus, event 2 occurred before the 1318–1487 A.D. age range of the lower part of peat P10. This is consistent with the observation that the event horizon for event 2 is only ∼25 cm above the top of peat unit P20 in trench T3, with ∼25 cm of unfaulted stratigraphic section below the base of peat P10. Given the lack of evidence for any major hiatuses in this stratigraphic interval exposed in trenches T3 and T4 (which had the most continuous sediment accumulation rate across the fault zone), this would suggest that E2 probably occurred soon after the deposition of P20 (i.e., soon after the 986–1181 A.D. age range of P20). Moreover, E2 occurred after event E3, which we interpret in the following section as the historical 1045 earthquake. Thus, event 2 occurred between 986 A.D. and 1487 A.D., and presumably between 1045 A.D. and 1487 A.D. We therefore interpret event 2 as the historical 1254 A.D. earthquake (Ergin et al., 1967; Ambraseys and Melville, 1995; Barka, 1996; Ambraseys and Jackson, 1998). This large-magnitude earthquake caused substantial damage and loss of life in Erzincan and apparently caused surface rupture along the North Anatolian fault over a distance of at least ∼150 km between Erzincan and Suşehri, as detailed in a remarkable eyewitness account reported in Ambraseys and Jackson (1998). The age limits of Yaylabeli event 2 (986–1487 A.D.) are similar to those of well-constrained event B in the nearby Hartleb et al. (2006) trench (890–1487 A.D.). Our results thus corroborate the findings of Hartleb et al. (2006) and confirm that there were no major surface ruptures on this stretch of the North Anatolian fault between the 1254 A.D. earthquake and the 1939 A.D. Erzincan earthquake.

Evidence for event 3 is best observed in trench T3 between m23 and m27.2. There, nine fault strands terminate upward at ∼75 cm depth, just below the base of P20. Liquefaction of peat P30 in trench T3 probably also occurred during this event. Moreover, the greater vertical separation of P30 relative to P20 at m7 in trench T4 is consistent with an event horizon at this depth. Specifically, P30 exhibits 35 cm of down-to-the-south separation on the fault at m6.25 in T4, whereas P20 exhibits only 14 cm of down-to-the-south separation. The absence of isolated, upward fault terminations at the same stratigraphic level in T4 suggests that event 3 ruptured only the main strand of the fault at m7, which subsequently reruptured in later events.

The ages of peat P20 and peat P30 bracket the occurrence of event 3. Three samples (T3-3, T4-05, and T4-10) from P30 provide a calibrated, calendric maximum age range for event 3 of 663 A.D. to 1153 A.D. As noted already, samples T3-2, T4-06, and T4-11 from peat P20, which was deposited after event 3, yield a calendric age range of 986 A.D. to 1181 A.D. A chronological model of layer and event ages based on OxCal 4.0 (Bronk Ramsey, 1995; Bronk Ramsey, 2001; Reimer et al., 2004) suggests that event 3 likely occurred between 914 A.D. and 1103 A.D. (Fig. 10). The historical, large-magnitude earthquake of 5 April 1045 A.D., which is thought to have ruptured a long section of the eastern part of the North Anatolian fault in the Erzincan-Suşehri region (Ergin et al., 1967; Ambraseys, 1970; Barka, 1996; Ambraseys and Jackson, 1998; Hartleb et al., 2006), occurs in the middle of the likely age range for event E3. We therefore interpret event 3 as the historical 1045 A.D. earthquake. At their Çukurçimen paleoseismic site, Hartleb et al. (2006) also saw evidence for a surface rupture that occurred soon after 880 A.D., which they also interpreted as the 1045 A.D. earthquake.

Event 4 is well recorded by fissure fills and upward fault terminations in trenches T3 and T4. Specifically, in trench T3, two fault strands terminate upward at m24.81 at 180 cm depth. These strands are ∼25 cm apart at ∼180 cm and merge at ∼230 cm depth, forming a wedge-shaped fissure. The fissure is filled with fine-grained sand derived from a thin sand layer between peat P40 and unit U45. The faults bounding the fissure terminate upward within this sand bed. Similarly, in trench T4, 20- and 30-cm-wide, wedge-shaped fissures are present on the east and west walls, respectively. These fissures are centered at m8.4 on the east wall and at m8.21 on the west wall. Faults bounding the fissure merge at m8.4,−1.74 on the east wall, whereas the faults bounding the fissure on the west wall merge downward at m8.39,−2.07.

The ages of peat P40 and unit 45 (an organic-rich clay layer) bracket the timing of event 4. Three calibrated, calendric dates from P40 (T3-5, T4-04, T4-09) yielded an age range of 689–991 A.D., providing a minimum age for event 4. Sample T4-07 from unit 45 yielded a calendric age range of 665–866 A.D., providing a maximum age. Thus, event 4 occurred between 665 A.D. and 991 A.D. Additional layer- and event-age modeling of these age constraints in OxCal v. 4.0 (Reimer et al., 2004; Bronk-Ramsey, 2001) allows us to refine and bracket the likely age of occurrence of E4 to between 717 and 844 A.D (Fig. 10).

This eighth- to ninth-century surface rupture at Yaylabeli was not observed by Hartleb et al. (2006) at their nearby Çukurçimen site, which had no recorded events between 540 A.D. and 930 A.D. Our reevaluation of the Çukurçimen trench data reveals that this interval in their trenches is stratigraphically and structurally complex. In addition, this stratigraphic interval was located at a horizontal bench at the fault zone in their key trenches. These observations lead us to suspect that this event horizon may exist, but it went unrecognized at Çukurçimen.

In marked contrast to the other four earthquakes that we identify (E1, E2, E3, and E5), to the best of our knowledge, there are no known historical accounts for a large-magnitude earthquake in this part of Anatolia during the 717–844 A.D. period when event 4 likely occurred. This was a period of unrest in eastern Anatolia, with ongoing wars between Byzantines and Arabs (e.g., Haldon, 2002), and it is possible that earthquake catalogs may be incomplete for this time period. These observations emphasize that the historical record may be incomplete, even in settings such as Turkey, which has a remarkably long and detailed historical earthquake catalog (e.g., Ergin et al., 1967; Ambraseys, 1970; Barka, 1992, 1996; Ambraseys and Finkel, 1995; Ambraseys and Jackson, 1998).

Event 5 occurred after deposition of peat P60, and before the deposition of P50. The thickness of peat P50 suggests that it represents a period of prolonged marsh development, probably due to a period of decreased alluvial input into the Yaylabeli site. In trench T3, the event horizon is marked by two fault strands that terminate upward at m22.66 and m22.98, just below peat P50 (Fig. 8). In trench T4, a fault strand at m6.09 on the east wall terminates at the base of peat P50, at the same stratigraphic level as in T3. In trench T1, however, we could not observe a distinct upward fault termination for this event. Instead, the fault is marked by an ∼30-cm-wide gouge zone, suggesting repeated rupture of the same strand. On the other hand, a fine-grained sand unit that exhibits liquefaction features is exposed between peats P50 and P60 in trenches T1 (Fig. 5; m3,−1), T2 (Fig. 6; m10,−2), T3 (Fig. 7; m23,−2), and T4 (Fig. 8; east wall; m6.5,−2), providing additional evidence for event E5.

Calibrated, calendric radiocarbon dates from five peat samples (T1-6A, T1-6B, T3-7, T4-03, and T4-08) from peat P50 in trenches T1, T3, and T4, indicate that event 5 occurred before deposition of P50 at 648–965 A.D. Sample T4-02, from peat 60 in trench T4, yielded an age of 253 A.D. to 415 A.D., providing a maximum age limit. Event 5 therefore occurred between 253 A.D. and 965 A.D. Our stratigraphic layer modeling narrows the likely age range for event 5 to between 302 A.D. and 724 A.D.

It is possible that event 5 is the historical 499 A.D. earthquake, which caused substantial damage in the Niksar-Suşehri region ∼75–175 km to the west of the Yaylabeli site (Ergin et al., 1967; Ambraseys and Jackson, 1998) (Fig. 1). This inference is supported by interpolation of our sediment accumulation rate curve, which suggests that event 5 occurred at ca. 500 A.D. (Fig. 11).

One of the most basic rationales for our work at the Yaylabeli site was the opportunity to test the completeness of the paleoseismologic record along this part of the North Anatolian fault by direct comparison of the new results with the results of the earlier companion study at the nearby Çukurçimen site (Hartleb et al., 2006). Such opportunities are exceedingly rare in paleoseismologic research, and the scarcity of such sedimentologically independent “paired” sites has hindered our ability to assess the completeness of paleoseismologic catalogs along most of the world's continental fault systems. Our comparison of the paleoearthquake records from these two sites provides insights into both the record of large-magnitude earthquakes along the North Anatolian fault and, more generally, into the completeness and repeatability of paleoseismologic records. As discussed in the following, comparison of the paleoseismologic records of these two sites yields generally close agreement, with one notable exception.

The Yaylabeli trenches provide a record of five surface-rupturing earthquakes on the eastern part of the North Anatolian fault during the past 2000 yr. Relatively rapid, continuous sediment accumulation at the site of 1–3 mm yr⁻¹, and wide stratigraphic separation of the different earthquake event horizons, suggest that this record is likely to be complete. Our results are generally similar to those of Hartleb et al. (2006) from the nearby (2 km to the east) Çukurçimen trench site, which was also located in a semi-enclosed, peat-rich marsh fed by small alluvial fans. The reproducibility of the results from these two sedimentologically independent trench sites (Fig. 12), coupled with the generally close correspondence between our paleoseismologic observations and historical earthquake records, lends confidence to our results.

Specifically, both the Yaylabeli and Çukurçimen trench sites contain evidence for: (1) the historical 1939 M_w 7.9 Erzincan earthquake (event 1 at Yaylabeli; “event A” at Çukurçimen), (2) an event (event 2) that occurred soon after the 986–1181 A.D. age of a well-dated peat, and after the historical 1045 A.D. earthquake at Yaylabeli, and between 980 A.D. and 1420 A.D. at Çukurçimen (event B; we follow Hartleb et al. [2006] in interpreting this event as the historical 1254 earthquake); (3) a surface rupture dated to between 914 A.D. and 1103 A.D. at Yaylabeli (event 3) and 930 A.D. to 1070 A.D. at Çukurçimen (event C; we concur with Hartleb et al. [2006] that this event is likely the historical 1045 A.D. earthquake); and (4) an event that likely occurred between 302 A.D. and 724 A.D. at Yaylabeli (event 5) and between 250 A.D. and 540 A.D. at Çukurçimen (event D). This event probably represents the historical 499 A.D. earthquake. The addition of the new Yaylabeli paleoseismic data allows us to update the space-time history of the earthquake occurrence along the North Anatolian fault of Hartleb et al. (2006) (Figs. 13 and 14).

However, the Yaylabeli trenches also yielded multiple exposures of an event marked by well-defined fissure fills that occurred between 717 A.D. and 844 A.D. This event was not observed at the earlier Çukurçimen site, despite the very favorable conditions and multiple trench exposures at that site. Our reexamination of the Çukurçimen trench logs reveals that the stratigraphic position of the event horizon for the missing event was located at a particularly complicated part of the exposure in the most informative trench at Çukurçimen (trench T1; Hartleb et al., 2006). Specifically, based on the radiocarbon age constraints from Çukurçimen trench T1, the event horizon for the eighth- to ninth-century surface rupture was located in an area of extremely complex, disturbed stratigraphy that masked evidence for the event. In addition, this stratigraphic section happened to coincide with the exact depth at the fault crossing of a horizontal bench that separated the upper vertical face of the trench from the lower vertical face (Fig. 12). These observations reinforce the need for multiple trench exposures at multiple sites and demonstrate that even under the most favorable trench conditions (i.e., fine-grained, well-bedded stratigraphy, a narrow fault zone, rapid sediment accumulation rate, and abundant dateable material), events may be missed.

The issue of the regularity of earthquake occurrence is of basic importance for probabilistic seismic hazard assessments (PSHA), because the ages of past major earthquakes are a fundamental input into such analyses (e.g., Cornell, 1968; WGCEP, 2007). Inasmuch as many PSHA studies rely on paleoseismologic data, the accuracy and especially the completeness of the paleoseismologic record become basic concerns. The ability to assess catalog completeness is thus of critical importance for accurate probabilistic forecasting. Yet, studies such as this one, in which high-quality, sedimentologically independent paleoearthquake sites in close proximity can be compared directly to assess the completeness of the paleoearthquake catalog, are rare. Because of the scarcity of such opportunities, it is imperative that both researchers and funding agencies realize the importance of such situations for studies of seismic hazard.

One of the basic issues in PSHA is the mode of earthquake recurrence. For example, is the recurrence of large-magnitude earthquakes relatively regular, or do these events occur more randomly in space and time? Turkey is an ideal place to test such questions because the North Anatolian fault, in marked contrast to some other major faults that traverse more structurally complicated plate boundaries (e.g., San Andreas and Garlock faults in California), is the dominant fault in a relatively structurally simple plate boundary (e.g., Barka, 1992; McClusky et al., 2000; Hubert-Ferrari et al., 2002; Hartleb et al., 2006; Reilinger et al., 2006). The relative structural simplicity of the Anatolia-Eurasia plate boundary in northern Turkey suggests that the North Anatolian fault is largely isolated from complexities in stress evolution related to earthquakes generated by other faults as well as from complicated patterns of elastic strain accumulation due to the relative absence of potential regional fault interactions such as those inferred for more complicated boundaries (e.g., Dolan et al., 2007). These factors suggest that the North Anatolian fault might be expected to exhibit relatively simple, perhaps even quasi-periodic earthquake recurrence.

The detailed, repeatable analysis of the paleoearthquake record afforded by the companion Yaylabeli and Çukurçimen sites, coupled with the historic record, demonstrates that while earthquake recurrence is usually relatively regular on at least the eastern part of the North Anatolian fault, this is not always the case; some intervals are of much longer duration than the “typical” average recurrence time. Specifically, the Yaylabeli and Çukurçimen paleoseismologic data indicate that only five surface ruptures have occurred on the eastern part of the North Anatolian fault over the past ∼1800 yr. Assuming that our correlations with the historical records are correct, three of the four intervals between these five earthquakes are 200–350 yr long (1045–1254 = 209 yr; ca. 770? ± ∼70 A.D. to 1045 ≅ 200–350 yr; 499? to ca. 770? ± 70 = ∼200–350 yr). This suggests that the timing of earthquake occurrence on the eastern part of the North Anatolian fault is commonly relatively regular. The fourth interval, however, is 685 yr long (1254–1939 A.D.), i.e., two to three times longer than the others. The repeatable measurements between the two sites are a key consideration in these statements, because our Yaylabeli data both: (1) revealed a newly identified surface rupture that was not recognized at the earlier Çukurçimen site, which occurred during the middle of one of the previously identified “long” interevent periods of Hartleb et al. (2006); and (2) allowed us to confirm the earlier measurements of the long interevent period between the historical 1254 A.D. and 1939 A.D. earthquakes. This temporal gap is real, and it demonstrates conclusively that earthquake recurrence on this part of the North Anatolian fault is not always temporally regular.

As noted by Hartleb et al. (2006), the historical 1579 A.D. and 1583 A.D. earthquakes, which caused widespread damage in parts of northern Anatolia, occurred during the long 1254–1939 A.D. interval. The 1579 A.D. earthquake, in particular, may have caused damage that may have extended over a 400-km-long region of northern Turkey, from Çorum to Erzincan (Ambraseys and Finkel, 1995), reminiscent of the damage pattern that might be expected from a large-magnitude North Anatolian fault earthquake. Thus, it is a potentially attractive idea to assume that either the 1579 A.D. or 1583 A.D. earthquakes (or both) occurred on the North Anatolian fault. Yet, like Hartleb et al. (2006), who found no evidence for a sixteenth-century surface rupture in their Çukurçimen trenches, we find no evidence for a surface rupture during this period in our trenches at Yaylabeli. Given the absence of any other sources large enough to generate an earthquake capable of causing damage over such a wide area, the lack of evidence for either of these earthquakes in the Yaylabeli and Çukurçimen trenches is puzzling. It is, of course, possible that an event may have gone unrecognized in the paleoseismologic trenches, as was the case for our Yaylabeli event 3 at the nearby Çukurçimen site. However, in light of the large number of trench exposures, the rapid sediment accumulation rates, absence of bioturbation, and the excellent stratigraphy at these sedimentologically independent sites, this seems rather unlikely. In Yaylabeli, for example, the sediment accumulation between events E1 (1939) and E2 (1254?) appears to be continuous throughout this interval in both T3 and T4, without any evidence for erosional features suggestive of significant hiatuses during this interval. In addition, the E1 and E2 event horizons are separated by a meter of section characterized by continuous, thin bedding with sharp depositional contacts. It is important to note that in our Yaylabeli trenches, the stratigraphic interval encompassing the E1 (1939 A.D.) and E2 (1254 A.D.) event horizons was entirely within the vertical wall of the uppermost tier of trench T3, and that there are no benches to potentially obscure stratigraphic and structural relationships within trench T4. At the Çukurçimen site, Hartleb et al. (2006) carefully assessed the possibility that 1579 A.D. and/or 1583 A.D. earthquake(s) were missed. As at Yaylabeli, their Çukurçimen trenches showed no evidence for any of these earthquakes, despite exposure of laterally continuous, thin-bedded strata across the fault zone within this stratigraphic interval (Fig. DR16 [see ^{footnote 1}]). As at Yaylabeli, the entire stratigraphic interval encompassing the event A (1939 A.D.) and event B (1254 A.D) surfaces ruptures at Çukurçimen was exposed continuously in the vertical wall of the top tier of the Hartleb et al. (2006) trench. These observations indicate that the 1579 A.D., 1583 A.D., and 1668 A.D. earthquakes did not cause surface rupture on the stretch of the North Anatolian fault that we examined west of the Erzincan basin.

Alternatively, it is possible that the 1579 A.D. and 1583 A.D. earthquakes were local, smaller-magnitude events. Historical data indicate that the 1579 A.D. earthquake caused widespread damage in and around the cities of Çorum and Amasya, in north-central Anatolia; a separate, near-contemporary account suggests that damage may also have occurred in Erzincan, ∼350 km east of Amasya (Ambraseys and Finkel, 1995) (Fig. 1). If the report of damage in Erzincan relates to a different earthquake, the 1579 A.D. earthquake may not have caused damage throughout northern Anatolia. Rather, the well-documented damage in the Çorum-Amasya area could have been caused by a moderate-magnitude earthquake on the Ezinepazarı fault, part of which ruptured as the westernmost section of the 1939 A.D. rupture between Ilgaz and Niksar (Figs. 1 and 11), or one of the reverse faults associated with uplift of the Pontide Mountains north of Ilgaz. Similarly, historical evidence for the 1583 A.D. earthquake comes mainly from the Erzincan area, which was heavily damaged in this earthquake with great loss of life. The 1583 A.D. earthquake could have been a moderate-magnitude earthquake that occurred near Erzincan, similar to the 1992 M_s 6.8 Erzincan earthquake, which caused 2000 fatalities and was generated by rupture of an ∼30-km-long section of the North Anatolian fault that did not cause any surface rupture (Barka and Eyidoğan, 1993; Fuenzalida et al., 1997).

Nevertheless, the sources of the 1579 A.D. and 1583 A.D. earthquakes remain unknown, and if future studies unearth unequivocal evidence for one or both of these earthquake(s) on the eastern North Anatolian fault, this would cut the 685-yr-long gap between the 1254 A.D. and 1939 A.D. earthquakes nearly in half. The resulting intervals would be 325 yr (1254–1579 A.D.) and 360 yr (1579–1939). Thus, if the 1579 A.D. did occur on the eastern North Anatolian fault, the recurrence intervals between all five of the most recent earthquakes would range from 200 to 350 yr, indicating very regular recurrence of large-magnitude earthquakes on this section of the North Anatolian fault. Even if the 1579 A.D. (or 1583 A.D.) earthquake(s) did not occur on the North Anatolian fault, as indicated by our paleoseismic results, the occurrence of earthquakes on the eastern part of North Anatolian fault appears to be relatively regular, with at least three of the four intervals between the last five events varying by less than a factor of two, from ∼200 to 350 yr.

The Yaylabeli and Çukurçimen paleoseismologic data demonstrate that surface-rupturing earthquakes on the eastern part of the North Anatolian fault have been rare events during at least the past two millennia, with recurrence intervals ranging from ∼200 to almost 700 yr. Geologic data from the central part of the North Anatolian fault indicate a slip rate over this time interval of ∼18–22 mm yr⁻¹ (Kozacı et al., 2007, 2009). These observations indicate that large amounts of elastic strain accumulate during interseismic periods on this part of the North Anatolian fault. This accumulated strain energy is released by large, infrequent earthquakes, such as the 1939 M_w 7.9 Erzincan earthquake, which generated ∼10 m of surface displacement in the Yaylabeli-Çukurçimen area (Hartleb et al., 2006).

The North Anatolian fault is well known for the occurrence of a mainly westward-propagating sequence of moderately large to great earthquakes (Mw 7.0–8.0) during the twentieth century (Ketin, 1948; Richter, 1958; Allen, 1969; Ambraseys, 1970; Toksöz et al., 1979; Barka, 1992, 1996, 1999; Ambraseys and Finkel, 1995; Stein et al., 1997). Historical and paleoseismological evidence shows that the twentieth-century sequence is not the only such cluster generated by the North Anatolian fault. For example, historical and paleoseismic studies indicate that the entire central and eastern North Anatolian fault ruptured during a relatively brief cluster in the tenth to eleventh century, including the historical 967 A.D., 1035 A.D., 1045 A.D., and 1050 A.D. earthquakes (Barka, 1996; Ambraseys and Melville, 1995; Stein et al., 1997; Sugai et al., 1998; Yoshioka et al., 2000; Pantosti et al., 2008; Kondo et al., 2004; Okumura et al., 2003; Hartleb et al., 2006) (Fig. 11). Similarly, the entire central and western sections of the North Anatolian fault ruptured during a sequence of eight moderately large to great earthquakes between 1666 A.D. and 1766 A.D., including the great M ∼8 1668 A.D. earthquake, which probably ruptured an ∼400-km-long section of the central North Anatolian fault from near Bolu to the 10-km-wide Niksar extensional stepover (Barka, 1996; Ambraseys and Finkel, 1995; Stein et al., 1997; Hubert-Ferrari et al., 2000) (Fig. 1). As discussed already, our paleoseismologic data suggest that this cluster did not extend along the eastern North Anatolian fault, which had ruptured earlier in the 1254 A.D. earthquake. The Niksar stepover and the 5-km-wide extensional stepover in the Erzincan basin appear to represent fundamental barriers to rupture propagation (and therefore maximum magnitude), consistent with observations demonstrating that seismic ruptures appear to be incapable of propagating through stepovers that are >3–4 km wide (Wesnousky, 2006). For example, the Niksar step controlled the western and eastern boundaries, respectively, of the 1939 and 1942 North Anatolian fault ruptures (Ketin, 1948; Barka, 1992; Stein et al., 1997; Wesnousky, 2006). As noted previously, this major structural complexity also probably controlled the eastern boundary of the 1668 A.D. rupture. Similarly, the 5-km-wide Erzincan Basin stepover bounded the eastern end of the 1939 rupture (Barka, 1992; Wesnousky, 2006), leading us to speculate that this fundamental barrier may have also controlled the eastern end of the 1254 rupture.

One of the key results of this study is the identification of a well-defined surface rupture that occurred between 717 A.D. and 844 A.D. at Yaylabeli. As noted herein, this event was not identified at the earlier Çukurçimen site, clearly demonstrating the value of repeatability of paleoseismologic measurements at multiple sites, wherever possible. Identification of the 717–844 A.D. surface rupture, combined with other paleoseismologic data (Sugai et al., 1998; Yoshioka et al., 2000; Okumura et al., 2003), suggests the occurrence of yet another cluster during the eight to ninth centuries. This previously unrecognized earthquake sequence possibly ruptured more than 600 km of the central and eastern North Anatolian fault. However, the behavior of the western North Anatolian fault during this period is difficult to discern because the historical earthquake record before the sixth century becomes difficult to interpret, particularly in terms of magnitude and exact location of the earthquakes (e.g., Ambraseys, 2002, 2006). This problem is exacerbated by the absence of any paleoseismologic constraints for earthquakes from the western North Anatolian fault prior to the ninth to tenth centuries (e.g., Rockwell et al., 2001a, 2009).

Although these observations emphasize that relatively brief clusters of earthquakes are common along the North Anatolian fault, the earthquakes that occur during these clusters are not the same from sequence to sequence, in terms of location, magnitude, and sequence of rupture. For example, the largest event in the twentieth-century cluster was the M ∼8 1939 A.D. Erzincan earthquake, and events in this sequence occurred in a predominantly westward-propagating sequence. The sequence of individual earthquakes in the cluster has been explained as resulting from a combination of the stress evolution of the fault system, as modeled by changes in Coulomb failure function (Stein et al., 1997), and by fault-zone material contrasts, which may play a major role in controlling rupture propagation and arrest (e.g., Dor et al., 2008). Likewise, the seventeenth- to eighteenth-century sequence propagated mainly westward, but consisted of earthquakes with magnitudes and locations that were significantly different from those of the twentieth-century sequence (e.g., the great 1668 earthquake appears to have ruptured approximately the 1943 and 1944 rupture segments). In contrast to these mainly westward-propagating sequences, the tenth- to eleventh-century sequence appears to have propagated both to east (the 967 A.D., 1035 A.D., and 1050 A.D. earthquakes, which ruptured in sequence eastward from the Bolu area) and west (the 1045 A.D. earthquake occurred along the eastern North Anatolian fault, prior to the 1050 A.D. earthquake on the central part of the fault) (Fig. 1).

Some events, however, such as the 1254 A.D. earthquake on the eastern North Anatolian fault and the 1509 A.D. and 1912 A.D. earthquakes on the western North Anatolian fault, do not fit into readily recognizable clusters. However, it is interesting to note that in the context of the 1254 A.D. earthquake, two paleoseismic sites on the central North Anatolian fault identify evidence for a thirteenth- to fifteenth-century surface rupture on that part of the fault (Okumura et al., 2003; Kondo et al., 2004). Moreover, a series of earthquakes (1343 A.D. through 1354 A.D.) ruptured the westernmost part of the North Anatolian fault (Ambraseys and Finkel, 1995; Rockwell et al., 2009), and thus it is possible that there was a “twentieth century-type” cluster of earthquakes on the North Anatolian fault during the thirteenth and fourteenth centuries. Arguing against this is the absence of evidence for such an event from the high-quality paleoseismic site of Sugai et al. (1998) near Ilgaz on the central North Anatolian fault. Similarly, the 1509 A.D. earthquake in the Marmara region near İstanbul (Ambraseys and Finkel, 1995) does not fit neatly into any well-documented cluster, and the relationship of the 1912 A.D. M 7.4 (Pondard et al., 2007) Gulf of Saros earthquake to the twentieth-century cluster likewise remains unclear. Even if these events were not part of short-lived sequences of earthquakes, the paleoseismologic and historical data demonstrate that the North Anatolian fault commonly ruptures in a systemwide sequence of moderate- to large-magnitude earthquakes.

Our paleoseismologic trenches at Yaylabeli, on the eastern North Anatolian fault, provide evidence for five paleosurface ruptures during the past two millennia, including a previously unknown eighth- to ninth-century surface rupture. The timing of these earthquakes indicates the relatively regular recurrence of large-magnitude events for this part of the North Anatolian fault, with three of four intervals between events varying between 200 and 350 yr. The recognition of one long (685 yr) interval, however, indicates that recurrence is not truly quasi-periodic on this stretch of the North Anatolian fault. These results, in combination with other paleoseismologic data, indicate that the North Anatolian fault is characterized by the occurrence of infrequent, large-magnitude earthquakes. Moreover, our results, particularly the identification of a previously unrecognized eighth- to ninth-century cluster on the central and eastern North Anatolian fault, strengthen the idea that the North Anatolian fault commonly, but not always, ruptures in clusters lasting on the order of a century.

We attribute this relatively simple behavior to: (1) the relative structural simplicity of the plate boundary in northern Turkey, in which >90% of plate boundary motion is accommodated by slip along the North Anatolian fault, with few other significant, high-slip rate faults; and (2) the mechanical efficiency of the structurally mature North Anatolian fault. Thus, the stress evolution of the North Anatolian fault is generally not complicated by stress interactions from earthquakes generated by other faults. These results emphasize the importance of the degree of tectonic complexity in controlling earthquake occurrence and suggest that earthquake occurrence on mature, structurally isolated strike-slip faults will be dominated by the relatively regular recurrence of large-magnitude events.

We thank Serdar Akyüz for helpful discussions, Robert Finkel, Michaele Kashgarian, and Paola Zermano for help with dating, Metin Gürcan, Kurt Frankel, and İsmail Gümüş for their help in the trenches, and the villagers of Yaylabeli for their support and kind hospitality. We thank Aurelia Hubert-Ferrari and Sally McGill for their thoughtful and constructive reviews. This research was funded by the U.S. National Science Foundation (grants EAR-9980564 and EAR-0409767).