Terrace formation linked to outburst floods at the Diexi palaeo-landslide dam, upper Minjiang River, eastern Tibetan Plateau

River terraces are frequently investigated with the aim of extracting information regarding tectonic or climate forcing on the evolution of landscapes. Terraces formed following the blockage of valleys by large-scale landsliding have received limited attention despite the high likelihood of their prevalence in landslide-dominated mountain belts. Here, we investigate the geomorphology, sedimentology, and chronology of two outstanding sets of terraces upstream of the giant river-blocking Diexi palaeo-landslide on the upper Minjiang River, eastern Tibetan Plateau. The first set occurs at Tuanjie village and has seven levels (T1–T7); the second set, at Taiping village, has three levels (T1–T3). All the terraces display a consistent sedimentary sequence comprising lacustrine muds topped by fluvial gravel sometimes capped by loess and a palaeosol. Based on field examination, lithofacies analysis, elevation data, and chronometric data (optically stimulated luminescence and radiocarbon dating), we correlate T1, T2, and T3 at Taiping with T5, T6, and T7 at Tuanjie. Our analysis suggests that two damming and three outburst events have occurred at the Diexi palaeo-landslide over the past 35 000 years. A giant landslide (> 300 m high) blocked the river before 35 ka, followed by the first outburst flood at ∼ 27 ka; the river was blocked again between 27 and 17 ka, followed by a second outburst at ∼ 17 ka. A third outburst at ∼ 12 ka was followed by gradual fluvial incision of the palaeo-dam crest to its current level. We attribute the terraces at Diexi to the recurrent blockage and outburst events, which reflect the shifting sediment transport capacity and incision at the palaeo-dam crest. Here, climatic fluctuations play a minor role in terrace formation, and tectonism plays no role at all.

1 Introduction

River terraces are temporary sediment storage sites along valleys that provide a natural archive of information on sediment transport and deposition through time (Chen et al., 2020; Liu et al., 2021), processes that are typically sensitive to the impacts of tectonism and climate (Pan et al., 2003; Singh et al., 2017; Avsin et al., 2019; Gao et al., 2020; do Prado et al., 2022). Terraces have been shown to reflect a wide range of geomorphic controls, such as rock uplift rate (Jansen et al., 2013; Pan et al., 2013; Giano and Giannandrea, 2014; Malatesta et al., 2021), fault activity (Caputo et al., 2008), crustal flexure (Yoshikawa et al., 1964; Westaway and Bridgland, 2007; Okuno et al., 2014), glacier melting (Bell, 2008; Oh et al., 2019; Vásquez et al., 2022), changes in sediment supply (Jansen et al., 2011), sea level (Yoshikawa et al., 1964; Malatesta et al., 2021), and even the internal dynamics of the fluvial system (Schumm and Parker, 1973). In tectonically active mountains, large-scale landslides, debris flows, and rockfalls (Molnar et al., 1993; Molnar and Houseman, 2013; Srivastava et al., 2017) can cause river blockages and associated sudden outburst floods that have a major impact on the sedimentary processes of the upstream and downstream reaches, including terrace formation (Korup et al., 2007; Hewitt et al., 2008; Korup et al., 2010; Hewitt et al., 2011). And yet few studies have explored the influence of extreme events on the formation and evolution of terraces (Montgomery et al., 2004; Yuan and Zeng, 2012; Zhu et al., 2013; Chen et al., 2016; Arzhannikov et al., 2018; Hu et al., 2018; Arzhannikov et al., 2020; Xu et al., 2020). We attempt to address that knowledge gap here.

Rapid uplift and climate change during the Quaternary led to frequent extreme geomorphic events in the area drained by the Minjiang River at the eastern margin of the Tibetan Plateau (Gorum et al., 2011; Fan et al., 2017, 2018; Wu et al., 2019; Dai et al., 2021; Yang et al., 2021). The upper Minjiang, for instance, displays many terrace sequences with origins that remain debated (Yang, 2005). But due to the lack of detailed sedimentological, chronological, and geomorphological information, the role of extreme geomorphic events, such as landslides and outburst floods, is still being explored (Yang et al., 2003; Yang, 2005; Gao and Li, 2006; Zhu, 2014; Luo et al., 2019).

A set of outstanding terraces occurs just upstream of the 300 m high Diexi palaeo-landslide dam, one of the largest, best-preserved, and longest-duration landslide-dammed lakes in a tectonically active setting (Fan et al., 2019). The Diexi terraces (Fig. 1) have been examined by previous researchers (Wang et al., 2005; Yang et al., 2008; Fan et al., 2019), but a systematic analysis has yet to be conducted. A set of terraces at the village of Tuanjie is thought to have resulted from repeated outburst floods from the Diexi palaeo-dammed lake between 30 and 15 ka – each terrace corresponding to a different outburst (Duan et al., 2002; Wang et al., 2005; Wang, 2009; Zhu, 2014; Ma et al., 2018). At least two blockage events have also been suggested (Yang, 2005; Yang et al., 2008), together with four periods of fluvial progradation (Xu et al., 2020). However, mechanistic details of the terrace formational processes based on the sedimentology and a comprehensive dating analysis are lacking. Here, we seek to address the unresolved questions of the origins of the Diexi terraces, including the following aims: (1) to conduct a detailed analysis of terrace sedimentology, (2) to obtain absolute depositional ages of the terraces (at Tuanjie and Taiping), and (3) to understand the evolution of the Diexi palaeo-dam since its formation at about 35 ka (Wang et al., 2020b). Our broader objective is to provide a better understanding of terrace formation linked to extreme geomorphic events in mountain regions.

Figure 1 The Diexi study area. (a) Location of Diexi at the eastern margin of the Tibetan Plateau. (b) Geological setting (maps modified from Guo, 2018; Wang et al., 2020a; Zhong et al., 2021). (c) Oblique view of the seven Tuanjie terraces, including elevations (m a.s.l.). (d) Oblique view of the three Taiping terraces including elevations (m a.s.l.).

2 Study area

The Diexi palaeo-landslide dam is located on the eastern Tibetan Plateau in the upper reaches of the Minjiang River. The area exposes rocks of the eastern part of the Bayan Har Block (Fig. 1a), spanning the Devonian, Carboniferous, Permian, Triassic, and Quaternary periods (An et al., 2008; Zhang et al., 2011; Ma, 2017; Zhong, 2017). This region of the Tibetan Plateau has been affected by intense and frequent earthquakes (Yang et al., 1982; Chen and Lin, 1993; Li and Fang, 1998; Shi et al., 1999; Hou et al., 2001; Lu et al., 2004) linked to the ongoing collision of the Indian and Eurasian plates (Fig. 1b).

The Diexi study has an arid to semi-arid climate (Shi, 2020), with a strong effect of the prevailing winds. Cumulative evaporation averages 1000–1800 mm a⁻¹ (Yang, 2005), and mean temperature and precipitation are 13.4° and 500–600 mm a⁻¹, respectively. Vegetation patterns show major elevational zonation and comprise mainly mountain coniferous forests, alpine meadows, and low shrubs at the highest elevations.

The Minjiang valley widens downstream, overall varying from 60 to 300 m wide at the valley floor (Yang, 2005; Jiang et al., 2016; Ma, 2017; Zhang, 2019) and up to 3000 m deep flanked by steep hillslopes that are typically 30–35° (Zhang et al., 2011; Guo, 2018). The Diexi palaeo-dammed lake (31°26^′–33°16^′ N; 102°59^′–104°14^′ E) is situated on the bend of the V-shaped Minjiang valley, which in turn lies in the well-known “north–south earthquake tectonic zone” (Tang et al., 1983; Huang et al., 2003; Yang, 2005; Deng et al., 2013).

The Diexi palaeo-landslide is located on the left bank of the Minjiang River at Jiaochang village (Fig. 1b). The palaeo-landslide length and width are ∼ 3500 and 3000 m, respectively, and volume is ∼ 1.4 to 2.0 km³ (Zhong et al., 2021). The highest parts of the palaeo-landslide reach up to 3390 m a.s.l. (metres above sea level), whereas the elevation of the dam crest is ∼ 2500 m a.s.l. (Dai et al., 2023).

At Taiping village (32°12^′ N, 103°45^′ E), three terraces occur near the mouth of Luobogou Gully (Fig. 1d) (Wang et al., 2005; Fan et al., 2021), while a suite of seven terraces occurs 12 km downstream at Tuanjie village (32°2^′ N, 103°40^′ E) near the mouth of the Songpinggou tributary (Fig. 1c). Further downstream, high-energy gravel outburst deposits occur at scattered locations, including near the villages of Xiaoguanzi, Shuigouzi, and Manaoding (Fig. 1b).

3 Materials and methods

3.1 Geomorphic and sedimentary description

Field surveys were carried out from October to November 2018. We describe sedimentary structure, geometric shape, sorting, roundness, and palaeo-flow direction of the gravel by applying the lithofacies approach primarily based on Miall (2000), but also including previous work conducted by Yang (2005) and Yang et al. (2008) (Table 1). The terraces were numbered according to elevation from the lowermost terrace (T1) to higher terraces (Tn).

Terrace elevations were measured using light detection and ranging (lidar) data with ∼ 0.5 m vertical accuracy and the Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model (ASTER GDEM) with ∼ 30 m vertical accuracy (Fan et al., 2021).

Table 1 Lithofacies of terrace sediments at Diexi. Adapted from Miall (2000), Yang (2005), and Yang et al. (2008).

3.2 Chronology

Two independent dating methods, optically stimulated luminescence (OSL) and radiocarbon dating, were employed to establish a reliable chronostratigraphic framework for the Tuanjie and Taiping terraces. We collected samples from the top and bottom of the lacustrine units, the top of the gravel units, and the base of the loess and palaeosol units with the aim of clarifying the timing of the damming and outburst processes and terrace stability: 19 OSL samples and 9 radiocarbon samples in total (Figs. 2 and 3).

Figure 2 OSL and calibrated radiocarbon (denoted as cal ka BP) dating results from Tuanjie. (a) The highest lacustrine deposits. (b) Lacustrine deposits and palaeosol at T7. (c) Gravel unit at T5. (d) Lacustrine deposits and palaeosol at T5. (e) Loess at T4. (f) Lacustrine deposits and palaeosol at T4. (g) Palaeosol at T3. (h) Lacustrine deposits at T3. (i) Gravel unit and palaeosol at T2. (j) Lacustrine deposits at T2. (k) Lacustrine deposits at T2. (l) Palaeosol at T1. (m) Lacustrine deposits at T1. Dashed white lines mark unit boundaries.

Figure 3 OSL and calibrated radiocarbon (denoted as cal ka BP) dating results from Taiping. (a) Paired OSL and radiocarbon samples collected from the highest lacustrine deposits. (b) Palaeosol at T3. (c) Lacustrine deposits in T3. (d) Lacustrine deposits in T3. (e) Lacustrine deposits at T2. (f) Lacustrine deposits at T2. (g) Lacustrine deposits at T1. (h) Lacustrine deposits at T1. Dashed white lines mark unit boundaries.

3.2.1 OSL dating

At the Tuanjie terraces, 12 samples were taken from the lacustrine deposits (excluding T6 and the highest lacustrine deposits), 2 samples were collected from the gravel units of T2 and T5, and samples were taken from palaeosols at T1 to T5 and T7 (Figs. 2 and 3). At the Taiping terraces, four samples were taken from the lacustrine deposits at T1 to T3 and the highest deposits, and another was taken from the palaeosol unit at T3 (Figs. 2 and 3). Samples were collected from freshly dug exposures by inserting stainless-steel tubes followed by careful sealing from light.

Samples were processed and measured at the Institute of Earth Environment, Chinese Academy of Sciences. The quartz grains were extracted following standard laboratory pre-treatment procedures (Kang et al., 2013, 2020).The sediment at the tube ends, which may have been exposed to daylight during sampling, was discarded and the unexposed samples were prepared for equivalent dose (D_e) and environment dose rate determination. Approximately 50 g samples were treated with 30 % HCl and 30 % H₂O₂ to remove carbonates and organic matter, respectively. The samples were then washed with distilled water until the pH value of the solution reached 7. For samples IEE5542 and IEE5550, the coarse fractions (90–150 µm) were sieved out and etched with 40 % HF for 45 min, followed by washing using 10 % HCl and distilled water. For the other 17 samples, the fine polymineral grains (4–11 µm) were separated according to Stokes' law. These fine polymineral grains were immersed in 30 % H₂SiF₆ for 3–5 d in an ultrasonic bath to extract quartz. Finally, the purified fine (coarse) quartz was deposited (mounted) on stainless-steel discs with a diameter of 9.7 mm. The purity of quartz was verified by the IRSL intensity and OSL IR depletion ratio (Supplement Figs. S1 and S2a; Duller, 2003).

All OSL measurements were performed on a Lesxyg Research measurement system, with blue light at (458±10) nm and infrared light at (850±3) nm for stimulation and a ⁹⁰S/⁹⁰Y beta source (∼ 0.05 Gy s⁻¹) for irradiation. Luminescence signals were detected by an ET 9235QB photomultiplier tube (PMT) through a combination of U340 and HC340/26 glass filters.

The single-aliquot regenerative-dose (SAR) protocol (Table S1 in the Supplement; Murray and Wintle, 2000; Wintle and Murray, 2006) was utilised to determine the equivalent dose (D_e) following Kang et al. (2020). Quartz grains were preheated at 260 °C for 10 s for a natural and regenerative dose, and a cut heat at 220 °C for 10 s was applied for the test dose. The quartz was stimulated for 60 s at 125 °C with blue LEDs; the OSL signal was calculated as the integrated value of the first 0.5 s of the decay curve minus the integrated value of the last 0.5 s as the background. For D_e determination, approximately 10 aliquots were measured for each sample. The mean D_e value of all aliquots was used as the final D_e value. Conventional tests in SAR protocol, including recuperation ratio, recycling ratio, quartz OSL brightness and fast-component-dominated nature, growth curve shape, and D_e distribution (Figs. S2 and S3), indicated that the protocol is adequate for the samples in this study.

The environmental dose rate was estimated from the radioisotope concentrations (U, Th, and K) and cosmic dose rates. U and Th concentrations were determined by inductively coupled plasma mass spectrometry, while K concentration was measured by inductively coupled plasma optical emission spectrometry. The cosmic dose rates were calculated using the equation proposed by Prescott and Hutton (1994). The α value of fine-grained (4–11 µm) quartz was assumed to be 0.04±0.002 (Rees-Jones, 1995). Considering the sedimentary texture, as well as current and past climate conditions since deposition, the water content of the gravel and palaeosol was assumed to be 10±5 %, while the water content of lacustrine deposits was estimated to be 20±5 %. Dose rate was calculated using the Dose Rate and Age Calculator (DRAC) (Durcan et al., 2015). Finally, the quartz OSL ages were obtained by dividing the measured D_e (Gy) by the environmental dose rate (Gy ka⁻¹).

3.2.2 Radiocarbon dating

Nine samples (all bulk sediment) were collected for radiocarbon analysis: two from the highest lacustrine deposits in the Tuanjie and Taiping terraces, one from the loess cap at Tuanjie T4, three from the bottom lacustrine deposits at Tuanjie T2, and three from the bottom lacustrine deposits at Taiping T1, T2, and T3 (Figs. 2 and 3). The radiocarbon sample collected from the highest lacustrine deposits at Taiping was used to compare with the OSL sample (TP19-1) taken from the same position. The radiocarbon samples collected from the highest lacustrine deposits at Tuanjie and the equivalent at Taiping were compared. Utilising the same dating method for age comparison enhances the robustness of our analysis. We sampled the loess unit at Tuanjie T4, as it was the most complete and easiest to access. Six samples taken from the bottom lacustrine deposits were used to determine the depositional ages of the terraces.

All samples were tested for organic matter and analysed using the NEC accelerator mass spectrometer and thermo-infrared mass spectrometer at the Beta Analytic Radiocarbon Dating Laboratory. All radiocarbon ages reported here are calibrated using IntCal 20 (Reimer et al., 2020).

4 Results

4.1 Terrace geometry and distribution

The seven terraces at Tuanjie and three terraces at Taiping terraces are all developed on thick lacustrine deposits (Fig. 4), which are naturally highly erodible. At Tuanjie, the lacustrine deposits are > 200 m thick, and the longitudinal (streamwise) lengths of the seven terraces range from 150 to 1000 m (Fig. 4, Table 2). The Taiping terraces are developed on a hillside with a slope of 40–60° and are therefore influenced by landslides and some human activity. The lateral extent of T1, T2, and T3 varies from 190 to 520 m (Table 2). Correlations between the terrace levels at the two sites are given in Table 2 and Fig. 4.

Figure 4 Sketch showing correlation between the Tuanjie and Taiping terraces (see Table 1).

Table 2 Elevation and correlation of terraces at Tuanjie and Taiping. Diexi Lake currently stands at ∼ 2150 m a.s.l.

4.2 Terrace lithostratigraphy

4.2.1 Tuanjie terraces

Tuanjie terraces T1, T2, T3, T4, and T6 are characterised by a sequence of silts, sands, gravel, loess, and palaeosol units. T5 and T7 lack the loess unit (Fig. 5a), probably due to erosion via human activities, and for the same reason T4, T5, T6, and T7 show strong signs of deformation and collapse. The lithostratigraphy (Table 1 and Fig. 5a) is summarised as follows (starting from the base).

1. Silt clay (Fl) has intense weathering, horizontal bedding, and wave bedding, characteristic of lacustrine deposits.

2. Gravelly (Gh, Gci, Gmm) fluvial deposits are separated by an unconformity with the underlying lacustrine deposits. The flow orientation of the gravel is predominantly parallel to the Minjiang River, suggesting it is the source of this gravel. The gravel units at T1, T4, T5, and T7 (Gh) are generally poorly sorted and well-rounded, with grain sizes ranging 2–30 cm. Longitudinal bedforms, lag deposits, and sieve deposits (Fig. 5a) are present. At T2 (Gci) the gravel shows inverse grading, with grain sizes ranging 2–25 cm (clasts > 35 cm are rare), and appears as poorly sorted and sub-circular to round clasts without orientation. At T3 (Gci), the gravel units exhibit inverse grading and are poorly sorted, with sub-circular to round clasts of grain size ranging 3–25 cm. Gravel at T6 (Gmm) shows graded bedding with good sorting and rounding.

3. Loess (Ls) units of T1 and T2 are brick red in colour; the loess at T3 contains angular fragments of phyllite.

4. Palaeosols (Ps), if present, are developed capping the fluvial strata and contain abundant roots (Fig. 5a). Lacustrine deposits extend above T7 with a thickness of 30 m; these deposits show undulating bedding and severe denudation (Fig. 4).

Figure 5 Terrace sedimentary sequences, lithofacies, and dating results (radiocarbon dates are denoted cal ka): (a) Tuanjie T1, T2, T3, T4, T5, T6, and T7, as well as the highest lacustrine deposits. (b) Taiping T1, T2, and T3, as well as the highest lacustrine deposits. All lithofacies labels are linked to Table 1; see Table 2 and Fig. 4 for terrace correlations.

4.2.2 Taiping terraces

Taiping terraces are characterised by a sequence of lacustrine silts, muds, gravel, loess, and palaeosol units (Fig. 5b). The lithostratigraphy (Table 1 and Fig. 5b) is summarised as follows (starting from the base).

1. Silt–clay (Fl) underlies all three terraces. Note that the highest extent of the lacustrine units reaches > 70 m thick.

2. Gravelly (Gh, Gci, Gmm) fluvial deposits observed on the Taiping terraces all show a flow direction aligned with Luobogou Gully, indicating that this gravel is derived from the gully. Gravel at T1 (Gcm) is characterised as poorly sorted and subrounded with grain sizes of 5–10 cm. Similarly, the gravel units in T2 and T3 (Gcm) contain numerous broken phyllite fragments. T3 displays two beds of horizontal, angular phyllite fragments (Gh, Gci) with grain sizes of 2–5 cm.

3. Mud (Fm) units contain snail shells, suggesting that these may be overbank deposits, abandoned channels, or drape deposits.

4. Loess (Ls) units at T2 and T3 are mixed with some angular phyllite fragments

5. Palaeosols (Ps) cap all three terraces.

4.3 OSL ages

We measured 19 quartz OSL dates in total: 14 from Tuanjie and 5 from Taiping terraces (summarised in Fig. 5 and Table 3).

At Tuanjie, the depositional ages of the lacustrine deposits range from ∼ 32 to 10 ka and do not follow a simple elevational sequence. T1, T2, T3, and T4 display a younging trend with increasing elevation, while T5 and T7 yield similar ages but are older than T3 and T4. Gravel units from T2 and T5 yield ages of 28±2 and 22±2 ka, respectively. The palaeosols are all Holocene in age, mostly ranging from ∼ 12 to 9 ka, with T1 yielding a notably younger age of ∼ 4 ka.

At Taiping, the depositional ages of all three lacustrine samples (plus the highest lacustrine sample) are consistently ∼ 10 ka.

Table 3 Summary of OSL data.

^* Given that the terraces are not completely flat, elevation data vary slightly from those in Table 2.

4.4 Radiocarbon ages

Nine radiocarbon ages were measured in total, all from bulk sediment samples (Table 4). The highest lacustrine deposits at Tuanjie and Taiping yielded ages of ∼ 27 and ∼ 17 cal ka BP, respectively. The loess sample collected from Tuanjie T4 yielded an age of ∼ 13 cal ka BP. The bottom lacustrine deposits of T2 yielded ages of ∼ 34, ∼ 39, and ∼ 37 cal ka BP. The depositional ages of all three bottom lacustrine samples of Taiping T1, T2, and T3 are ∼ 30, ∼ 34, and ∼ 30 cal ka BP, respectively.

Table 4 Summary of the radiocarbon results for Tuanjie and Taiping.

5 Discussion

5.1 Reliability of dating results

First, we consider the reliability of our chronology. Given the relatively stable depositional environment of the silt-rich (lacustrine and palaeosol) samples and the normal distribution of D_e, we assume they were well-bleached before deposition and therefore yield reliable ages.

Our ages are consistent with those reported by previous studies at Diexi, which fall mainly between about 36 and 11 ka (Table S2). We note that the older ages of the Tuanjie and Taiping lacustrine deposits (Fig. 5a) are ∼ 35 and ∼ 30 ka, respectively; however, two other published sources support our result: (1) a basal radiocarbon age (calibrated to 35.1±0.3 cal ka BP) reported from the Diexi Lake ZK2 drill core (Wang et al., 2012) and (2) two radiocarbon ages from another lacustrine section at Tuanjie (calibrated to 35.8±0.4 and 30.7±0.03 cal ka BP) reported by Zhang et al. (2009).

Both gravel units at Tuanjie T2 and T5 (∼ 28 and 22 ka, respectively) yield OSL ages that are much older than the underlying lacustrine deposits (∼ 11 and 10 ka, respectively) (Fig. 5). In this case, we favour the lacustrine ages and exclude the samples collected from thin sand lenses within the gravel.

At Taiping, our radiocarbon–OSL dating pair collected from the highest lacustrine deposits yields ages of 17.2±0.2 cal ka BP and 10.8±1.2 ka, respectively (Fig. 5). In this case, we suspect the radiocarbon age is overestimated due to the “old carbon reservoir” effect. This reservoir effect in the sample can result from several factors, including (1) lower-¹⁴C-activity carbon and atmosphere–water exchange (Deevey et al., 1954; Keaveney and Reimer, 2012; Ascough et al., 2016); (2) landslides, debris flows, or other disturbances causing surface sediments to drop into the lake, mixing older sediments with new (Counts et al., 2015; Shi, 2020); and (3) the redeposition of older organic components, such as stored charcoal (Kaplan et al., 2002; Krivonogov et al., 2016).

5.2 Terraces along the upper Minjiang River

A minimum of 15 sets of river terraces occur along the upper Minjiang River valley, with 9 sets located upstream of Diexi (from Gonggaling to Zhangla), 2 sets near Diexi (Taiping and Tuanjie), and 4 sets downstream (the Maoxian–Wenchuan area). From previously published work, we compiled a total of 124 dates (OSL, infrared stimulated luminescence, thermoluminescence, radiocarbon, and electron spin resonance) measured on the terraces of the upper Minjiang River (Table S2). Terraces upstream of Diexi go as far back as ∼ 830 ka (Zhao et al., 1994) but fall primarily between ∼ 47 and 2 ka (Fig. 6). Terraces in the Diexi area span ∼ 505 to 2 ka (Kirby et al., 2000; Duan et al., 2002; Yang et al., 2003; Gao and Li, 2006; Wang et al., 2007; Wang, 2009; Mao, 2011; Jiang et al., 2014; Zhong, 2017; Guo, 2018; Luo et al., 2019; Zhang, 2019; Wang et al., 2020b), with the majority being 32–2 ka (Fig. 6). Downstream reaches host terraces ranging ∼ 400 to 50 ka (Zhao et al., 1994; Yang et al., 2003; Yang, 2005; Zhu, 2014), with a significant fraction falling between ∼ 40 and 20 ka (Fig. 6).

Figure 6 Frequency distribution histogram of terrace ages since 50 ka in the upper reaches of the Minjiang River (at Diexi, upstream, and downstream). By far the most frequent terrace age falls between 20 and 10 ka.

Terraces upstream (Zhangla basin to the source of the Minjiang) are attributed to tectonic uplift (Yang et al., 2003; Yang, 2005; Yang et al., 2008, 2011; Chen and Li, 2014; Zhu, 2014), whereas, by contrast, the Tuanjie and Taiping terraces are thought to relate to the evolution of the Diexi palaeo-dam (Duan et al., 2002; Wang et al., 2005; Wang, 2009; Zhu, 2014). Terraces downstream in the Maoxian–Wenchuan region share similar characteristics to those at Diexi, as they are also believed to have formed via outburst flooding from a palaeo-dammed lake (Zhu, 2014). However, those terraces are also strongly influenced by activity along the Maoxian–Wenchuan fault zone. We hypothesise that the formation and evolution of the Diexi terraces (at Tuanjie and Taiping) are distinct and independent of the upstream and downstream terraces. We test and discuss this idea further in the following sections.

5.3 Correlation of the Tuanjie and Taiping terraces

The highest lacustrine deposits at Tuanjie and Taiping occur at the same elevation (∼ 2390 m a.s.l.), suggesting that the two sets of terraces are also related somehow. The Tuanjie and Taiping terraces certainly share similar lithostratigraphy (Fig. 5). For instance, Tuanjie T5 and Taiping T1 share the same sedimentary sequence (from the base to top): silty clays (Fl), gravel (Gh at Tuanjie, Gcm at Taiping), and palaeosol (Ps), and very similar sequences are shared by Tuanjie T6 and Taiping T2 as well as Tuanjie T7 and Taiping T3. In addition, the chronologies (Table 3) we have from the lacustrine deposits at Taiping T1 (9.5±1 ka) and Tuanjie T5 (10.4±0.6 ka) compare closely, as do Taiping T3 (10±0.8 ka) and Tuanjie T7 (10±0.8 ka). Based on these considerations, together with their elevation, we suggest that Taiping T1, T2, and T3 correspond to Tuanjie T5, T6, and T7 (Fig. 4).

5.4 Controls on terrace formation at Diexi: tectonism, climate, or outburst floods?

The formation of terraces in mountain rivers is typically attributed to either tectonic activities (Burgette et al., 2017), climate change (Maddy et al., 2005; Gao et al., 2020), or some combination of those (Luo et al., 2019; Chen et al., 2020; Narzary et al., 2022; Ma et al., 2023). The impact of extreme events on terraces has come to the attention of researchers more recently (Hewitt, 2016; Wang et al., 2021; Yu et al., 2021). At Diexi, the great thickness (> 200 m) of lacustrine deposits carved by floodwaters and topped discontinuously by terrace gravel and loess–palaeosol sequences suggests a role for tectonism, climate, and outburst floods, but the relative influence of each has yet to be clarified. We pursue this question below.

5.4.1 Effects of tectonism on the Diexi terraces

The Tuanjie and Taping terrace sites are sufficiently close (12 km) to be considered subject to equivalent tectonic forcing. In Sect. 5.2, we divided the upper Minjiang River into three segments: Gonggaling to Zhangla (upstream of Diexi), the Diexi area, and the Maoxian–Wenchuan area (downstream of Diexi). Since the initial damming at the Diexi palaeo-landslide, the fluvial incision rates in these three segments of the upper Minjiang have been measured at 8.3–85.3, 13.6–198, and 58 mm a⁻¹, respectively (see Table S2). In comparison, the Minshan Block (which includes the reach from Gonggaling to Maoxian) is thought to have experienced an average uplift rate of 1.5 mm a⁻¹ during the Quaternary (Zhou et al., 2000). Clearly, recent incision rates in the Diexi area have been several times faster than the average uplift rate of the Minshan Block. This highlights the unique character of Diexi and suggests that tectonic activity is not a primary factor in the formation of the terraces.

5.4.2 Effects of climate changes on the Diexi terraces

The regional climate underwent three transitions from cold–dry to warm–humid between ∼ 40 and 30 ka (Zhang et al., 2009), followed by more than 10 alternations of cold to warm between 30 and 10 ka (Wang, 2009; Wang et al., 2014). The terraces at Tuanjie and Taiping span the past 39 kyr, so to investigate the influence of climate we examine the climate variations over the same period (Fig. 7). The four climate proxies reveal significant fluctuations from the end of the Last Glacial Maximum (LGM) to the early Holocene, followed by relative stability throughout the Holocene.

Figure 7 Palaeoclimate (δ¹⁸O) proxies compared with the OSL and radiocarbon chronologies obtained from the Diexi terraces. (a) Sanbao Cave (Wang et al., 2008); (b) Hulu Cave (Wang et al., 2001); (c) East Asian monsoon (Cheng et al., 2016); (d) GISP-2 (Grootes et al., 1993); (e) the Diexi terraces at Tuanjie (solid symbol) and Taiping (hollow symbol). The early Holocene, Younger Dryas (YD), Bølling–Allerød interstadial (BA), and Last Glacial Maximum (LGM) are labelled.

It is tempting to speculate that warmer periods triggered wetter conditions or glacier melt, leading to the overtopping of the palaeo-dam and formation of terraces; however, we cannot see any clear relationship between the age of the terraces and the climatic variations over the past 39 000 years (Fig. 7). Nevertheless, four important points are worth making.

1. A fluctuating climate may be seen in terrace geometry. In papers by Mao (2011), Jiang et al. (2014), and Shi (2020), it is argued that Tuanjie T2 displays an irregular sequence of ages with depth that suggest repeated fluctuations in the lake level by up to 11 m between 19 and 11 ka (the deglaciation period) (Table S2). Regarding Tuanjie T1, we note the extraordinary terrace width. There are three possible factors that created the very wide T1 terrace: (i) during this period, strong monsoon activity resulted in high discharges and low sediment input, leading to river incision (Malatesta and Avouac, 2018; Tian et al., 2021; Yu et al., 2021). (ii) We note that some additional erosion may have occurred owing to the positioning of the Tuanjie terraces on the concave margin of the valley (Fig. 1b) where lateral fluvial erosion tends to be accentuated. (iii) As the lowest terrace, Tuanjie T1 was subjected to frequent erosion during the progressive outburst of the palaeo-dam (Phase IV to Phase VII; Fig. 9).

2. Some degree of climate control can be recognised in terms of the aeolian and weathering processes. The loess unit at Tuanjie T4 (∼ 13.4±0.1 cal ka BP) dates to just before the Younger Dryas, reflecting a cool depositional environment; loess observed at Tuanjie T3 and T2, as well as Taiping T3 and T2, suggests ages slightly younger. Most of the palaeosol units relate to the warming conditions of the early Holocene.

3. The three outburst floods (∼ 27, ∼ 17, and ∼ 12 ka, as reported in Sect. 5.5) in the Diexi area happened in the climate fluctuation periods. We speculate that these floods may be the result of glacial melting; Wang et al. (2012) mentioned that during the last glacial period, the melting of glaciers triggered massive hillslope instability and formed palaeo-dammed lakes.

4. The absence of an outburst flood in the Holocene may be related to the warm and stable climate.

5.5 Terrace formation and the evolution of the Diexi palaeo-landslide dam

Damming and outburst floods can strongly impact upstream and downstream areas, causing aggradation and incision (Fig. 8) (Hewitt et al., 2008; Korup and Montgomery, 2008). A lake formed by the blockage of a river can raise water levels upstream, resulting in potential upstream flooding (Guo et al., 2016), and following an outburst flood, the lake level drops as a result of sudden erosion at the crest of the dam. During this lower lake level, the river cuts through the easily eroded lacustrine deposits, forming terraces.

Figure 8 Model of palaeo-landslide dam evolution through time, starting with a blocking event (e.g. a major landslide), which then becomes a natural dam on the river, causing a lake to form. Lacustrine deposits accumulate behind the dam, sometimes to great depth (at Diexi lacustrine sediments are > 200 m thick). A positive water balance in the lake triggers overtopping of the dam, causing potentially catastrophic outburst floods downstream. The outburst flood typically erodes the crest of the dam, subsequently lowering the lake level and allowing fluvial processes to resume along parts of the valley. This repeated process yields a terrace stratigraphy comprising (from base to top) lacustrine deposits topped by fluvial deposits perhaps capped by loess and palaeosol development.

Figure 9 Schematic model of the evolution of the Diexi palaeo-dam and Tuanjie terraces. See Sect. 5.5 for detailed descriptions of each phase. Brown text denotes the ages of loess and palaeosol units.

The triangle formed by Tuanjie, Jiaochang, and Xiaohaizi (Fig. 1b) marks the centre of the palaeo-dammed Diexi Lake. We suggest that this ancient lake experienced multiple damming and breach events leading to major outburst floods down the Minjiang River. For instance, high-magnitude outburst sediments are identified downstream around Xiaoguanzi–Manaoding (Fig. 1b). Based on our terrace lithofacies and chronological analyses, we attempt to reconstruct the history of river-blocking and outburst floods sourced from the Diexi Lake as follows.

The Minjiang River was blocked by the Diexi palaeo-landslide sometime before 35 ka (Phase I: > 35 ka), as indicated by five lines of evidence: (1) the bottom lacustrine deposits of Tuanjie T2 date to ∼ 35 ka; (2) the deposition age of Taiping T2 is ∼ 34 ka; (3) the basal radiocarbon age in a drill core from Diexi Lake is 35.1±0.3 cal ka BP (Wang et al., 2012) (note that the lacustrine pile extends ∼ 80 m deeper); (4) at Xiaoguanzi, lacustrine sediments dated to 34.9±0.8 and 35.6±0.8 cal ka BP (Wang et al., 2012) are observed capping part of the palaeo-landslide dam; and (5) the same occurs at Manaoding dated to 34.5±0.2 cal ka BP (Wang et al., 2012).

After being initially blocked by the palaeo-landslide, Diexi Lake reached its highest level around 27 ka (the highest lacustrine sediments at Tuanjie date to 27.1±0.2 ka; Fig. 5a). This matches the timing of evidence of the first known outburst flood (Phase II: ∼ 27 ka), a gravelly unit near Xiaoguanzi (Fig. 1b) OSL-dated by Ma et al. (2018) at 27.3±2.8 ka. Further evidence of an outburst flood (or floods) around 27 ka is indicated by two other nearby sites dated with OSL and radiocarbon, respectively: (1) a 35 m thick sequence of deformed lacustrine bedding at Shawan (Wang et al., 2011, 2012) and (2) convolution structures exposed near Jiaochang (Fig. 1b) (Wang et al., 2012). Around ∼ 27 ka appears to have been a time of major perturbation in the upper Minjiang River: a palaeo-landslide at Qiangyang (Fig. 1b) is radiocarbon-dated to 26.5±0.5 ka and 27.3±0.4 cal ka BP (Wang et al., 2012), and downstream, a palaeo-dammed lake at Maoxian is radiocarbon-dated to 26.8±1.0 cal ka BP (Wang et al., 2007).

The Diexi palaeo-dam was re-established and sedimentation in the lake resumed for about 10 000 years (Phase III: ∼ 27–17 ka), as indicated by the highest lacustrine sediments at Taiping dated to 17.2±0.2 cal ka BP (Fig. 5b).

The second outburst flood (or floods) occurred at ∼ 17 ka (Phase IV). This event incised the palaeo-dam, causing the Diexi Lake level to drop by ∼ 110 m (to 2279 m a.s.l.), as recorded at Taiping T1 and Tuanjie T5 (Fig. 5a, b). The lowering of the lake level exposed the highest lacustrine deposits at Taiping. The palaeo-landslide at Manaoding, dated to 16.8±0.6 cal ka BP (Wang et al., 2012), is possibly linked to this second outburst flood.

In the 5000 years that followed (Phase V: ∼ 17–12 ka), two more outburst floods may have lowered the palaeo-dam further (forming Tuanjie T4 and T3), although the timing is uncertain. Yet, we can say with confidence that an outburst flood at ∼ 12 ka (Phase VI) lowered the palaeo-dam by ∼ 70 m (to 2204 m a.s.l.), forming Tuanjie T2. From ∼ 12 ka to the present (Phase VII), it appears that the Diexi palaeo-dam crest has gradually incised to its current level at ∼ 2150 m a.s.l. (aside from the brief period at a higher level following the 1933 Diexi earthquake; Dai et al., 2021).

6 Conclusions

We set out to investigate the origin and chronology of two sets of outstanding terraces formed upstream of the giant river-blocking Diexi palaeo-landslide on the upper Minjiang River, eastern Tibetan Plateau.

The Tuanjie terraces have seven levels (T1–T7), while those at Taiping have three (T1–T3). All terraces display a consistent sedimentary sequence comprising thick lacustrine muds topped by fluvial gravel, which at a few sites are capped by loess and a palaeosol. We correlate T5, T6, and T7 at Tuanjie with T1, T2, and T3 at Taiping.

Our reconstruction of the history of terrace formation suggests that two damming and three outburst events have occurred at the Diexi palaeo-landslide over the past 35 000 years. The sequence of events is summarised as follows: a giant landslide (> 300 m high) blocked the river before 35 ka, followed by the first outburst flood at ∼ 27 ka; the river was blocked again between 27 and 17 ka, followed by a second outburst at ∼ 17 ka. A third outburst at ∼ 12 ka was followed by gradual fluvial incision of the palaeo-dam crest to its current level.

Our findings at Diexi provide a detailed case study of terrace formation linked to the evolution of the palaeo-landslide dam. The Diexi terraces (at Tuanjie and Taiping) are distinct and independent of the upstream and downstream terraces along the upper Minjiang River – they are not the direct product of either tectonic or climate forcing. Instead, terrace height and geometry are the result of the sequence of outburst floods that have progressively lowered the crest of the palaeo-landslide dam (the local base level to the terraces) since its emplacement more than 35 000 years ago.

This study proposes a new perspective on terrace formation in steep rivers draining landslide-dominated mountain belts. Given the frequent observation of valley-blocking dams in high mountain settings, we suspect that the terrace formation processes described here may be more widespread than has been previously recognised.