Linking SSD structures with earthquakes is commonly based in tectonically active areas, deformation of fine grained material, vertical repeatability, lateral traceability, separation by non-deformed layers, occurrence of SSD types simulated in laboratory, and exclusion of gravity collapse (Sims, 1975; Owen and Moretti, 2011; Jiang et al., 2016). According to these six criteria, seismic events can be recognized from SSD structures in the lacustrine sediments. Intriguingly, the SSD layers are always covered by coarse-silt layers in the Lixian sequence (Jiang et al., 2017). Based on the sedimentological analysis of high-resolution grain-size proxies, such as end-member analysis, C-M (C: one percentile, M: median diameter) diagram, Sahu value (Sahu, 1964), and so on, these layers are caused by earthquakes that triggered dust storms and massive landslides and then changed the source of dust particles (Figure 3) (Jiang et al., 2017). As a result, the Lixian section displays a much higher deposition rate of 1.75 mm/yr than that of the southern Chinese Loess Plateau (∼0.08 mm/yr, Ding et al., 1994). This analysis is corroborated by continuously fine-grained (mean grain size: 15.9 μm) deposition at the Lixian lakeshore instead of coarse-grained sediments usually caused by fluvial flow (Supplementary Figure S2). Therefore, SSD structures and coarse-silt layers are used as sedimentological indicators of seismic events (Jiang et al., 2016, 2017).

Seismic shaking with different intensity can trigger the different types of SSD structures of the water-saturated sediments, and the stronger earthquakes usually result in stronger deformation (Rodrı́guez-Pascua et al., 2000; Wetzler et al., 2010; Lu et al., 2020). Accordingly, SSD of similar types in different regions probably correspond to the same triggering mechanism and intensity conditions. On this basis, fluid dynamics modeling is used to obtain the peak ground acceleration (PGA) required to trigger the typical types of deformation (Wetzler et al., 2010; Lu et al., 2020).

Previous studies indicate that earthquakes with different intensity can lead to different types of SSD structures, and the fluid dynamics modeling are used to obtain the PGA required to trigger different types of SSD structure (Wetzler et al., 2010; Lu et al., 2020). The lower-bounding magnitude of SSD is Mw ≥ 5.3 with PGA ≥0.13 g Modified Mercalli Intensity (MMI) ≥ VI½ in the Dead Sea (Lu et al., 2020).

In this study, we collectively analyzed the SSDs in the Lixian section. The clastic dyke and micro-faults are cross-layer and often occur locally in the strata (Jiang et al., 2016), so continuity is not guaranteed, and the age is uncertain. The clastic gravel in Lixian section is composed of bedrock debris, which is possibly triggered by bedrock landslides induced by earthquakes of different intensity (Jiang et al., 2016). We eliminate these SSDs in the whole Lixian sequence. In addition, ball-and-pillow structure in Jiang et al. (2016) are named according to deformation patterns, which cannot clearly reflect the strength of the deformation. Corresponding to the fluid dynamics modeling, the ball-and-pillow structures exhibit minimal deformation, so we redefined them as “micro-folds”. The micro-folds can reflect the initial stage of the flame structures and slump folds (Jiang et al., 2016). As the deformation continues to intensify, clastic breccia layers are formed when the original rock strata break up.

As mentioned above, four typical types of SSDs, i.e. micro-folds, flame structures, slump folds and clastic breccia layers are identified corresponding to linear waves, asymmetric billows, coherent vortices and intraclast breccia layer, respectively (Figure 2A,B) (Jiang et al., 2016; Lu et al., 2020). We project the different types and thicknesses of SSD into the acceleration-thickness diagram to get the PGAs required to trigger different types of SSD in the Lixian section.

The degree of ground shaking during earthquakes can be determined by the documented PGA and MMI (Bilal and Askan, 2014). In various earthquake studies, MMI is inferred from regional datasets, peak ground-motion data, isoseismic maps, and earthquake damage reports (Worden et al., 2012), especially in paleoseismic studies. Du et al. (2019) established the relationship between MMI and PGA in the Longmen Shan fault zone based on 34 moderate to large earthquakes. In this study, we use this relationship to infer the MMI information of the paleoseismic records in the Lixian section.

The equation is expressed as MMI = 3.311     log   PGA - 0.354 (1)

Establishing a seismic attenuation relationship is an important action in regional evaluation of seismic risk (Lei et al., 2007). Due to the complex tectonic background in China, there are significant differences in seismic attenuation relationships established in different regions.

Lei et al. (2007) established a seismic attenuation relationship based on 96 recent destructive earthquakes in Southwest China as follows. Long axis:   I a = 7.3568   +   1.2780 M - 5.0655 l g ( R a   +   24 ) (2) Short axis : I b = 3.9502 + 1.2780 M- 3.7567 g ( R b + 9 ) (3) where I is the intensity of the MMI and R is the epicentral distance.

Considering that the Wenchuan earthquake intensity isoseismal map is distributed along the strike of the Longmen Shan fault zone (NE-SW), and the Lixian section is located on its short axis (Figure 1B). In this study, we use the short-axis relationship to study the seismic record of the Lixian section.

The b value is the most important parameter in the magnitude-frequency relationship (Gutenberg-Richter distributions) (Wesnousky, 1995), and plays an important role in seismicity studies and seismic hazard analysis. In order to obtain a reliable complete earthquake catalog and b value, we use the empirical relations (Lei et al., 2007; Du et al., 2019) to obtain the fault distribution range and epicenter distance that triggered the lowest-degree deformation in the Lixian section based on the upper limit of potential source earthquake. We extract the seismic data (6,435 data from 1970 to 2020) within the range of 95 km away from the Lixian section.

The minimum completeness magnitude (Mc) is identified by the methods of cumulative number of earthquakes plotted against time and maximum curvature (Woessner and Wiemer, 2005). On this basis, the maximum likelihood method is used to fit the corresponding seismicity parameters (b value).

In the Lixian section, we identified 17 SSD layers (Supplementary Table S1). Due to the absence of paleoseismic records from the middle Holocene (∼6.0 ka) to present in the Lixian section, the 2008 M _w 7.9 Wenchuan earthquake provided a favorable condition for the magnitude limitation of large earthquakes in this study. Analysis of empirical equation and inversion suggests that the instrumental data (PGA = 0.21–0.38, Liu and Li, 2009) around the Lixian section correspond to the magnitudes of 7.9–8.5 and MMI of Ⅶ–Ⅷ of the Wenchuan earthquake, by taking R = 50 km (epicentral distance). The above results are consistent with the actual data (M _w of 7.9, MMI of Ⅷ–Ⅸ). These results increase the reliability of the empirical equation applied in the eastern margin of the Tibetan Plateau.

In this study, six layers of micro-folds, five layers of flame structures, four layers of slump folds and two layers of clastic breccia have been identified in the Lixian lacustrine sequence (Supplementary Table S1, Figure 2A). According to the distribution of different types of the Lixian SSD strcutures in the acceleration-thickness diagram (Figure 2C), the PGA for triggering SSD is ≥0.13g, ≥ 0.19g, ≥ 0.43 g and ≥0.63 g (Supplementary Table S2). Because the projection points are on the boundary of different deformation types, we obtained the lower boundary of acceleration that can reflect the triggering of typical deformation on the eastern margin of the Tibetan Plateau. According to the regional empirical equation (Du et al., 2019), we constrain the intensity of the four typies of SSD paleoearthquake records with MMI ≥ Ⅵ½, ≥ Ⅶ, ≥ Ⅷ½ and ≥ Ⅸ (Supplementary Table S2).

According to the regional empirical attenuation relations and regional geological survey, the middle segment of Longmen Shan fault zone is the main fault that triggered deformations in the Lixian lacustrine sequence, while the Miyaluo fault can only trigger the low-degree deformations (micro-folds and flame structures) (Supplementary Text S1). In this study, the minimum epicentral distance triggering deformation in the Lixian section is limited to R_min ≥ 10, and the R_max ≤ 95 km (Supplementary Table S3). According to the regional empirical equation (Lei et al., 2007; Du et al., 2019), the six layers of micro-folds (MMI ≥ Ⅵ½, PGA ≥0.13) and five layers of flame structures (MMI ≥ Ⅶ, PGA ≥0.19) recorded in the Lixian lacustrine sequence correspond to magnitudes of M ≥ 5.9, 6.3 (R_min ≥ 10 km) and M ≥ 6.8, 7.2 (R ≥ 30 km), respectively (Table 1). Previous studies show that SSD can be triggered by M _w > 5.7 in or near isoseismic lines with MMI ≥ VII (Monecke et al., 2006). Multiple faults in the region considered are likely to trigger low-degree deformations in the Lixian lacustrine sequence. Therefore, MMI ≥ Ⅵ½ (M ≥ 5.9) are taken as the lower-boundary conditions for triggering the deformation in the Lixian lacustrine sequence, by taking R_min ≥ 10 km.

Four layers of slump folds (MMI ≥ Ⅷ½) and two layers of clastic breccia layers (MMI ≥ Ⅸ) recorded in the Lixian section correspond to PGA of 0.43 and 0.63, respectively, which are probably a response to the larger seismic shaking in the middle segment of the Longmen Shan fault zone. According to the regional empirical equation (Lei et al., 2007; Du et al., 2019), the MMI ≥ Ⅷ½ and Ⅸ earthquake events correspond to M ≥ 8.1 and 8.6, by taking R_min ≥ 30, respectively (Table 1). The trigger conditions for slump folds (M ≥ 8.1, MMI ≥ Ⅷ½) are consistent with the actual data of the Wenchuan earthquake (M _w of 7.9, MMI of Ⅷ - Ⅸ), which well constrain the largest earthquake magnitude of the Lixian section. Because only two layers of clastic breccia are recorded in the Lixian lacustrine sequence which are less representative (Figure 2C), and the M ≥ 8.6 is much larger than the potential source magnitude (M = 8.0) in the eastern margin of the Tibetan Plateau (Supplementary Figure S3), the clastic breccia layers in this study are divided into slump folds. Thus, M ≥ 8.1 is the highest magnitude recorded by SSD structures in the Lixian lacustrine sequence.

Besides the SSD layers in the study area, there are also many coarse-silt layers, among which 17 directly overlie the SSD layers, indicating that deformation developed at the sediment surface (e.g. Alsop et al., 2022) (Figure 4A) and 46 layers exist independently (Figure 4B, Supplementary Table S4). These coarse-silt layers are caused by earthquakes, through which many more sources of clastic particles increased instantly and supplied plenty of dust particles for the study area (Figure 3) (Jiang et al., 2017). Generally, the thickness of the seismic event layers recorded by lacustrine sediments can be correlated with the seismic intensity, and further used to infer the regional palaeoseismic intensity (Moernaut et al., 2014).

The SSD thickness in the Lixian section has a good correlation (coefficient of 0.67) with seismic intensity, while the correlation weakens between thickness of coarse-silt layers that overlie SSD and seismic intensity with a low coefficient of 0.30 (Figure 4C,D). Intriguingly, when MMI is ≤ Ⅵ½, the thickness of coarse-silt layers are less than 10 cm (Figure 4D); when MMI is > Ⅵ½, the coarse-silt layers become thicker with the increase of intensity. The thickest is up to 19 cm (MMI ≥ Ⅷ½). Therefore, the thickness ≥19 cm of coarse-silt layer is linked with MMI ≥ Ⅷ½ event, and the 10–19 cm layer is linked with MMI ≥ Ⅶ event in this study.

In addition, the China seismic intensity scale shows that MMI of Ⅵ can produce slight damage, and MMI ≥ Ⅶ is a destructive earthquake (http://www.gb688.cn/bzgk/gb/index). The boundary condition for deformation of linear waves in the Lixian section is MMI ≥ Ⅵ½ (PGA ≥0.13 g), which is sufficient to produce geomorphic destruction. Therefore, 35 event layers of coarse particles (5–10 cm) occurring independently correspond probably to the moderate-strong earthquakes with MMI ≥ Ⅵ½.

Considering that the aftershocks of the 2008 Wenchuan earthquake are too many to represent the long-term seismic activity in the study area, the space-time window method (Gardner and Knopoff, 1974) is used to delete the fore- and aftershocks. Based on the analysis of seismic data from 1970 to 2020 (Supplementary Figure S4A), the quality of seismic records after 1986 (Supplementary Figure S4B) is high with M _c of 3.5 and b value of 1.06 (Figure 5F).

The frequency-magnitude relationship shows that the moderate-strong earthquakes (M ≥ 5.9) and the strong earthquakes (M ≥ 6.3) records in the Lixian section extends the fitting line of Gutenberg-Richter relationship from 5.3 to 6.3 with b value of 1.06 (Figure 5F), which indicates that the palaeoseismic records revealed by this study make up for the vacancy of the regional moderate-strong earthquakes and increase their completeness. The large earthquakes (M ≥ 8.1) are much more frequent than in the fitting line of Gutenberg-Richter relationship (Figure 5F), which possibly indicate regional characteristic earthquakes (Wesnousky, 1995). Nevertheless, the significantly low b values (0.5–1.2) in Mianzhu-Maoxian of the middle-north segment of the Longmen Shan fault zone reflect the sliding state of frequent moderate-small earthquakes under the relatively high stress (Yi et al., 2006; Yi et al., 2008; Yang and Zhang, 2010), and strong earthquakes are most likely to occur in the future. This is consistent with many strong earthquakes in the middle segment of Longmen Shan fault zone revealed by this study, which implies that the magnitude constraint for strong seismic shaking events in this study is appropriate.

Linear interpolation was carried out based on the existing optically stimulated luminescence (OSL) dating results (Supplementary Figure S5, Jiang et al., 2016) to define the timing of paleoseismic events in the Lixian section. The 63 moderate-strong earthquakes of MMI ≥ Ⅵ½ (M ≥ 5.9) in the Lixian lacustrine record have a mean recurring time of 210 years (Figure 5A and Table 2). The mean recurring time of strong earthquakes of MMI ≥ Ⅶ (M ≥ 6.3) is about 600 years. The large earthquakes of MMI ≥ Ⅷ½ (M ≥ 8.1) have the longest recurring time of 1,200 years.

The dimensionless coefficient of variation (COV) is a commonly used parameter in describing earthquake recurrence model, which includes “quasi-periodic” (COV ≤0.7), random (COV ≈1) and “clustered” (aperiodic, COV >1) (Kagan and Jackson, 1991; Berryman et al., 2012; Griffin et al., 2020; Moernaut, 2020). For intraplate settings, a Poisson or clustered recurrence model seems the most appropriate, while plate boundaries generally show quasi-periodic or weakly periodic of recurrence intervals (Williams et al., 2019; Moernaut, 2020). The MMI ≥ Ⅵ½, ≥ Ⅶ and ≥ Ⅷ½ events recorded in the Lixian section have the COV of 0.66, 1.06, 0.77, respectively (Table 2; Figure 5). In the following, we discuss the recurrence models and trigger mechanism of these three types earthquake.

The Longmen Shan fault zone is developed along the eastern margin of the Tibetan Plateau, which is the plate boundary between the Bayan Kala fault block and Sichuan Basin (Figure 1). It has a low slip rate (Supplementary Figure S1) (Zhang, 2013). The recurrence interval of the 13.3-ka-long large earthquake records of MMI ≥ Ⅷ½ (M ≥ 8.1) follows the Weibull distribution (Figure 5) with a small COV (0.77), which reflects the weak quasi-periodic recurrent model of the plate boundary (Williams et al., 2019; Moernaut, 2020). The 1,200 years of recurrence interval of large earthquakes in the Lixian section is much shorter than 2,000–6,000 years inferred previously (Zhang et al., 2008; Zhang, 2013; Ran et al., 2014), which indicates the continuous lacustrine sequence are of great scientific significance in revealing a continuous paleoseismic record and assessing the seismic risk. The 2008 M _w 7.9 Wenchuan earthquake occurred in the middle segment of Longmen Shan fault zone, which is the seismogenic fault that possibly triggered the high-degree deformations in the Lixian lacustrine sequence. In addition, the isolated and geometrically simple plate boundary faults exhibit relatively regular recurrence patterns (quasi-periodic) (Berryman et al., 2012). Therefore, the weak quasi-periodic recurrent model of large earthquakes recorded in the Lixian section may respond to the periodic accumulation and release of energy in the Longmen Shan fault zone caused by the mutual compression of the Tibetan Plateau and South China block (Zhang et al., 2003). In contrast, the MMI ≥ Ⅵ½ (M ≥ 5.9) earthquake records also show the stronger quasi-periodic (the Weibull distribution and COV of 0.66) (Figure 5), which may be due to the combined effect of multiple faults (Longmen Shan fault zone, Minjiang fault and Miyaluo fault, Supplementary Figure S4) leading to more frequent seismic records (Supplementary Text S1).

The MMI ≥ Ⅶ (M ≥ 6.3) earthquake records follow a power-law distribution with a COV of 1.06, which indicate a random (Poisson) recurrence model in the in-plate (Williams et al., 2019; Moernaut, 2020). The additive influence of different active seismic sources can result in an overall Poisson recurrence process (Gomez et al., 2015), and the strong stress interactions on the complex fault geometry and other active faults in the vicinity can lead to irregular seismic cycles (Visini and Pace, 2014). Therefore, the random recurrence model of 13.3-ka-long earthquake events of MMI ≥ Ⅶ (M ≥ 6.3) is a superposition response to the multiple active seismic sources of the middle segment of Longmen Shan fault zone and Miyaluo fault (Supplementary Text S1).

The number of MMI ≥ Ⅶ (M ≥ 6.3) and MMI ≥ Ⅷ½ (M ≥ 8.1) earthquakes recorded in the Lixian section during the last deglaciation (11.6–18.0 ka) is significantly less than that in the early Holocene (6.0–11.6 ka). The most critical aspect is that only one earthquake is recorded during the Younger Dryas (YD, 11.6–12.9 ka) and Heinrich 1 (H1, 14.6–17.8 ka) (Figure 5E). Entering the Holocene, the number of earthquakes in the Lixian section increases significantly (Figure 5) and shows a “quasi-periodic” earthquake recurrence model (COV of 0.59 and 0.62), while a random recurrence model fits for the last deglaciation (COV of 0.98 and 0.92). The frequency of earthquake recurrence in Bølling-Allerød (BA) and Holocene (warm period) is more frequent, but much less in YD and H1 (cold period), which reflects the obvious correlation between earthquake occurrence and climate change in the Longmen Shan fault zone. This is comparable with the behavior of strong earthquakes (M > 7) in Japan, more frequent in spring and summer (warm) than in autumn and winter (cold) (Heki, 2003). The increase in pore-fluid pressure caused by groundwater recharge can trigger seismic activity by reducing the effective normal stress on the fault (Heki, 2003; Christiansen et al., 2005; Montgomery-Brown et al., 2019), and is deserving of further investigation in the future.