SEISMOTECTONICS OF THE INNER TIENSHAN: SUUSAMYR BASIN AND ADJACENT AREAS

. The Ms=7.3 Suusamyr earthquake of August 19, 1992 occurred in an area reputedly aseismic. Because it was not expected there, this event attracted worldwide attention of researchers in seismology and seismotectonics, but their results have not been included in the most recent seismic zoning map of Kyrgyzstan. New studies of neotectonic structures and focal mechanisms of earthquakes in the Suusamyr area and adjacent areas give reason to revise the established notions about the seismicity of the region. The seismic hazard in Inner Tienshan appears important and M max are comparable to those of the Northern and Southern Tienshan, where numerous destructive events were documented in the XIX and XX centuries. For the southern parts of the study area, along Naryn River, where hydroelectric power stations are planned, the new data should be used. (IB7320-110694).

According to instrumental data, the epicenter of the Suusamyr earthquake is in the heads of the West Aramsu and East Aramsu rivers, which are spatially related to the E-W South-Aramsu strike-slip fault and the Kyzyloi thrust ( Fig.  2, a). The second, strongest shock came 68 minutes after the main shock on the N-W segment of the Suusamyr-Toluk Fault, where a surface rupture occurred. This rupture could be traced for more than 6 km lengthwise as a series of imbricate fractures cutting moraines and proluvial cones (Fig. 2, b). In kinematic terms, the rupture plane dipped <30° to the southwest and had a 0.9-1.4 m thrusting displacement with a small (10-15 cm) dextral component [Bo gachkin et al., 1997;Su Zongzheng et al., 1999]. The second seismic disruption appeared 25 km to the east, at the junction of the Aigyrzhal and East-Aramsu faults (Fig. 2,  a, b). It consisted of a 4 km long series of fractures, the most prominent of which in the flood plain of Suusamyr River, near km 162 of the Bishkek-Osh highway. It was also a thrust with dextral strike-slip component. The largest vertical dis placement was 2.7 m with a horizontal offset of 20-30 cm [Bogachkin et al., 1997;Ghose et al., 1997;Ainscoe et al., 2019].
The area of subsequent events such as landslides, rockslides, rockfalls, mud eruption and soil fracturing amounts to more than 4,000 km². Mapping of these features, together with the degree of infrastructure damage, formed the basis for the earthquake intensity assessment [Korjenkov, 2006]. Intensity isolines 9 and 10 (MSK64 units) define two separate ellipses elongated along the activated faults ( Fig. 2, a), one for the main shock along the E-W Aigyr djal Fault, the other for the second shock along the NW-SE Suusamyr-Toluk Fault. Westward propagation of the seismic rupture along the Suusamyr-Toluk Fault was inferred from the aftershock migration direction [Djanuzakov et al., 2003;Korjenkov, 2006]. As both rupture zones have thrust and dextral strike-slip components, the double shock induced a counterclockwise rotation of the Aramsu Block (Fig. 2, b). This interpretation was already stated after primary investi gations of the source area of the Suusamyr earthquake [Bogachkin et al., 1997]. The highest density of aftershocks was recorded within this Aramsu block. Mud eruptions shortly after the seismic event along the borders of this block are consistent with the block rotations and subsequent openings of favorably oriented fracture zones (Fig. 2, b).
Suusamyr river after (Morozov et.al.,1988) Suusamyr basin m SHAMSI GROUP occur at an interval of 3-5 ka [Ainscoe et al., 2019]. These data, along with the decoding of the neotectonic structure of the area and information about the present-day stress field [Rebetsky et al., 2016] probably are sufficient to create a tectonophysical model of the earthquake source based on the dextral rotation of the Aramsu block. Structures in the area to the south of the Suusamyr earthquake source zone are completely different. The EW-striking conjugated Northern Kavak and Southern Kavak Thrusts brought Paleozoic rock units on the up to 3500-4000 m thick Jurassic to Cenozoic sediments of the Minkush-Kokomeren depression [Bachmanov et al., 2008;Sadybakasov, 1990] (Fig. 3), the stratigraphy of which is similar to that of the main basins, such as the Chu, Issykul and Naryn Basins of Central Tien Shan (see Fig. 1b). The Northern Kavak and Southern Kavak Faults join near the eastern end of the Tokotgul reservoir, and the westward continuation of the Minkush-Kokomeren depression is a suture [Burg, Miko laichuk, 2009;Mikolaichuk et al., 2008]. The Minkush-Kokomeren depression formed during Late Pliocene -Pleistocene transpression with shortening (evidenced by folds, reverse and thrust faults) combined with sinistral strikeslip offsets [Bachmanov et al., 2008].

MICROSTRUCTURE -FAULT DATA
The structural work consisted in a systematic investigation of outcrops exposing mapped faults to determine relative movements from fault surface structures such as striations and sense of shear criteria. Striae are mostly minerals fibres, ridges and grooves produced by hard objects driven along the fault surface (Fig. 4). Local senses of shear were defined from steps, calcite fibre growth, Riedel shears and half-moon features (e.g. [Twiss and Moore, 1992]). All fault planes were very sharp. Those are features characterizing brittle faults formed under very low temperatures near surface conditions. Taking into consideration this observation and working on or close to active faults with locally documented surface ruptures, we assumed that measured striae and movement features recorded recent movements. More than 12 mesoscopic fault planes with attitudes as various as possible were measured in all sites, which are few tens of metres long and usually of uniform lithology. Such field precautions are required to allow a reliable kinematic analysis.
Fault data were systematically recorded with the aim of computer-aided paleostress tensor calculations. We used the Program FSA 28.3 [Célérier, 2009], which is based on a Monte Carlo search calculation, using random stress tensors to evaluate the tensor best-fitting the measured fault planes and their striations. The detailed assumptions and working procedure are described in [Burg et al., 2005] and further developed in [Célérier et al., 2012]. The best-fitting solution is a reduced stress tensor that can be graphically visualized as stereographic, lower projection of fault data and calculated principal stress. Robustness of the result can be checked with Mohr circles of calculated states of stress and histograms of angular error [Burg et al. 2005]. Since such calculations were aimed at supporting instrumental, seismic information from focal mechanism, we will not develop further this routine technique of structural geology (e.g. [Angelier, 1994;Twiss Unruh, 1998]). At variance with seismological information, it is accepted that such calculations provide a longer term assessment of 'stress' directions than the instrumental time.
The results are summarized as stereographic, lower hemi sphere projections of principal stress directions and the shape ratio of the corresponding stress ellipsoid r=(σ 1 −σ 2 )/ (σ 1 −σ 3 ), where σ 1³ σ 2 ≥σ 3 , all positive in compression. r»0 means that σ 1 » σ 2 , r≈1 if σ 2 »σ 3 . The latter case concerns the two easternmost sites, all other calculated tensors being >0.7 (Fig. 5). Altogether, these values indicate that principal stresses have magnitudes very close to each other, which can be expected for near-surface faulting and which makes easy swaps between σ 1 and σ 2 or σ 2 and σ 3 , thus a variable stress field combining compression and extensionregimes. Only two sites (Muztor River and Suusamyr 1.3) yielded ' Andersonian' states of stress, with one of the principal stress nearly vertical. However, horizontal σ 1 in Muztor River indicates compression, while subvertical σ 1  in Suusamyr 1.3 indicates extension. All other calculated tensors yielded inclined principal stress directions, most of them with a rather, yet ill-defined WSW-ENE compression. Two reasons are envisioned to explain this variability in the orientation and the shape ratio of the calculated tensor.
(1) Measured fault planes were reactivate while stresses vary or the rock mass rotates so that each fault datum records a different stress tensor relative to its refe rence frame; (2) Measured fault planes were reactivated under a stable configuration of stress and rock mass orientation, but local reactivation matching local heterogeneities yields a regionally imperfect solution. Since in most tensors, none of the principal axes is either vertical or horizontal, a most likely hypothesis is tilting due to block rotations after faulting, possibly linked to free topographic effects.
The complexity of the results may also reflect the fact that measurements lump fault information related to major faulting events, which are responses to regionally significant stress fields, with fault data recording 'aftershock' events, which relax local stresses and deformation. Comparison with modern seismicity suggests that calculated tensors like focal mechanisms document movements responding to local post-seismic relaxation of stresses along different faults under possibly different states of stress. Calculations are therefore significant in interpreting long-term accumulation of stresses and strain in the deforming upper crust.

MOTION TYPES IN THE EARTHQUAKE SOURCES
The existing network of seismic stations allows comparing the types of motion at the earthquake source with the movements from fault surface structures. There was no seismic event in the 20 years preceding the Ms=7.3 Suusamyr earthquake in its source area. For this reason, we could only use the fault plane solutions for the main shock and the aftershocks recorded by the Kyrgyz analogue network. The aftershock solutions within the first five hours after the main shock are not defined because of record overlays. The Suusamyr main shock is consistent with structural observations, i.e. reverse with a dextral component (Fig. 6). Focal mechanisms of M≥4 aftershocks vary regionally. In the southern part of the aftershock area, the thrust component is similar to the motion of the main shock. To the north, thrust movements that occurred on shallower dipping planes whilst strike-slip components are important. Normal faulting took place in the peripheral part of the after shock area. Several normal faults are located to the north-west of the main M=6.7 aftershock, and others at the south-eastern edge of the aftershock area (Fig. 6).
These solutions correlate well with the variety of motions and stress tensor calculations defined from microstructural studies, for example, for the Tuyashuu Pass, Alabel Pass and Karakol. However, thrusting is the prevailing motion of active faults of the Suusamyr earthquake source area. Earthquakes to the south of the Suusamyr earthquake epicenter were not strong (M<4.5). Only 4 earthquakes with М>4 were recorded in the last 15 years along the middle segment of the Talas-Fergana Fault (Fig. 7). Their focal mechanisms (two of them are normal faults, and two are thrusts) yield no prevailing motion type. Again, this difference in stress regimes matches microstructural calculations, normal faults likely accommodating local tension, for example in releasing bends in the general transpressional tectonic system under a sub-spherical stress tensor. To com plement information of so few earthquakes, the fault plane solutions of events with М<3.5 (Fig. 7) have also been considered in 'Earthquakes of Northern Eurasia ', Yearbook, 1992', Yearbook, -2007. There are 7 strike-slip, 5 thrust and 3 normal fault events. This suggests prevalence of strikeslip motion along the Talas-Fergana Fault segment traced on the map (Fig. 7).

MAXIMAL MAGNITUDE OF THE EARTHQUAKES
Historical and instrumental data provide no record of any ancient earthquake comparable in magnitude with the 1992 Suusamyr earthquake in the study area over the last 200-300 years (see Fig. 1, c). This period, however, is very short compared to the recurrence intervals of large intraplate earthquakes, which usually range from several centuries to several millennia [Djanuzakov et al., 1980[Djanuzakov et al., , 1997. That is why more reliable M max estimates require using paleoseismologic data [McCalpin, 1996[McCalpin, , 2009Solonenko, 1974;Yeats et al, 1997]. Paleoseismologic studies in the study area identified numerous surface ruptures, large rockslides and caldera-like cavities associated with large prehistoric earthquakes [Strom, 2000[Strom, , 2009[Strom, , 2013Korjenkov, 2006;Korjenkov et al. 2012;Mamyrov et al. 2009]. The largest possible М max M=7.3 for the South-Aramsu and M=6.9 for the Suusamyr-Toluk Faults (see Fig. 2, a) were estimated in accordance with the magnitudes of the main shock and largest aftershock of the Suusamyr earthquake. The М max for the Kyzyloi thrust was estimated from the Late Pleistocene Kokomeren rockslide, which is located in the western part of the fault and is one of the biggest paleoseismologic events in Kyrgyzstan [Strom, Stepanchikova, 2008;Strom, 2010]. This gigantic rockslide (41.93 °N, 74.23 °E) is approximately 1.0 km 3 . The 400-m thick rockslide is exposed on the left bank of Kokomeren River, where it covers a high fluvial terrace  that was about 100 m above the riverbed at the time of the rockslide. The most important indication in favour of seismically-triggered landslide is the presence of an active fault that displaced the river terraces few hund red meters downstream from the site [Strom, Stepanchikova, 2008]. According to its morphological parameters, this landslide is bigger than all the known surface damages caused by the Suusamyr earthquake. Consequently, the M max value along the Kyzyloi Fault is likely >7.3. The M max of the Talas-Fergana Fault is taken from the Chatkal earthquake (1946; M=7.5, I0=9-10 of MSK64) [Djanuzakov et al., 2003]. Moreover, the earthquake source zone with M=7.1-8 is aligned along the Talas-Fergana Fault Djanuzakov et al., 1980]. This zone is based on 17 seismic events documented between 6120±170 and 250±50 years, with an average return period of 300 years [Korjenkov et al. 2012;Mamyrov et al., 2009]. These events were dated by the radiocarbon method (Table; Fig. 8). Traces of the largest events allow concluding that M max for these events might reach 8 units Djanuzakov et al., 1980].
T a l a s -F e r g a n a F .

Radiocarbon dates of samples collected from displaced gullies along the Talas-Fergana Fault
Радиоуглеродные датировки образцов (палеопочв), отобранных из смещенных русел вдоль Талас-Ферганского разлома by a 34 km² prehistoric landslide that left the 6.5 km long headscarp in Devonian sedimentary rock. The dam volume is >6 km 3 [Strom, 2010]. Another paleoseismic evidence is the 300x106 m 3 Karakul rockslide that blocked the Karasu-Left River mouth (41°38'N, 72°39'E). The 2.5x108 m 3 and about 250 m high rockslide is composed of Devonian limestone that collapsed from the 900 m high left bank of the valley and thrusted over Permian sandstone and conglomerates. Besides the Karasu-Left River mouth, it blocked also the Naryn River [Strom, 2010]. The resulting 150 m deep Karasu Lake invaded the Karasu-Left River valley. This dam is located at 41°34.5'N, 73°13.5'E exactly over the trace of the Talas-Fergana active fault [Mamyrov et al., 2009;Strom, 2010]. Comparable paleoseismological events occurred along the Northern-Kavak Fault, as evidenced by a series of caldera-like craters and rockslides (see Fig. 2, a). The eastern crater (Kyzylkul, 41°48.1'N, 73°45.3'E) is a roughly elliptic body, about 3x2 km in size and 3 km 3 in volume with steep, 250 to 700 m high edges and a relatively flat bottom [Strom, 2000;Strom, Groshev, 2009]. The second crater (Djuzumdy, 41°48.75'N, 073°22.85'E), about 1x0.5 km and 200-300 m deep, has a roughly rhomboid form. Its volume is about 0.12 km 3 . It is located 28 km to the west of the Kyzylkul dam in the upper reaches of the Djuzumdybulak stream. The sou thern boundary of the Djuzumdy crater is an active fault that ruptured about 2000 years ago [Strom, 2000;Strom, Groshev, 2009]. These craters were compared to Note. Sites (fault segments) are arranged as shown in Fig. 8 from SE to NW. Compiled by [Korjenkov et al., 2012] after [Burtman et al., 1996;Korjenkov et al., 2010;Mamyrov et al., 2009;Trifonov et al., 1990].
6. CONCLUSION This work shows that the geographically unexpected Ms=7.3 Suusamyr earthquake of August 19, 1992 is not the strongest event that may occur in the area. Stronger seismic events took place not only in the western, but also in the southern parts of the Suusamyr earthquake source zone, namely within Minkush-Kokomeren transpressional zone, along which the Naryn River.
The Kambarata HPP-2 projects is already completed. Con struction of the Kambarata Hydro Power Plant-1 has already started. Following the existing seismic zoning maps Turdukulov, 1996], these two power plants are designed with account of risks of possible earthquakes with <7.5. Our study shows that magnitudes up to 8 should be expected. We hope that experts in seismic zoning and earthquake-resistant construction will revise and reassess their project to face the risk.
With a similar warning conclusion, attention should be drawn to the western part of the Minkush-Kokomeren transpressional zone that includes the Toktogul water reservoir, the largest in Central Asia with a volume of 19.5 km 3 and an area of 284 km² [Simpson, Negmatullaev, 1978]. Filling of the reservoir started in 1973. At that time, the geological study of the territory subject to flooding consisted only of geological mapping (scale 1:200000) and aeromagnetic studies (scale 1:100000). Deep structures hidden beneath the Cenozoic cover were ignored.
Our data showing that the Minkush-Kokomeren transpressional zone extends as far west as the Talas-Fergana Fault give grounds to conclude that the reservoir is a high-risk site that should be surveyed with geophysical sounding techniques (electric, magnetic, seismic), and activities aimed at mitigation of any catastrophic event should be properly planned.

ACKNOWLEDGMENT
We are grateful to Alexander Strom for a comprehensive discussion of the Inner Tienshan seismodislocations. We thank SNSF for financial support of this research (Project No IB7320-110694).