Introduction

The Jishishan earthquake occurred at approximately 23:59:30 CST (15:59:30 UTC) on 18 December 2023 in northwestern China. The mainshock epicenter (35.745°N, 102.827°E) was located in Jishishan County, Gansu Province (Fig. 1a, b) (Zuo and Zhao 2024). The focal mechanism of the approximately 10-km-deep mainshock was interpreted as a shallow thrust-fault event (Jia et al. 2024; Wang et al. 2024a; Zhang et al. 2025). The maximum seismic intensity was VIII according to the China Earthquake Administration (CEA) (Shi et al. 2024) (Fig. 1c). By 24 December 2023, the CEA had reported 12 aftershocks with magnitudes greater than 3.0 (Guo et al. 2024; Yang et al. 2024). A total of 151 people were killed and 983 were injured, making this the deadliest earthquake in China since the 2014 Ludian earthquake in southwestern China. The earthquake also destroyed 70,000 houses, damaged another 99,000, and caused an estimated direct economic loss of RMB 14.612 billion (Ministry of Emergency Management of the People’s Republic of China).

Fig. 1
Fig. 1
Full size image

Location of the Ms 6.2 Jishishan earthquake on 18 December 2023 in northwestern China. The seismic-intensity map and active-fault data were obtained from the China Earthquake Administration; aftershock data are from Guo et al. (2024)

The Jishishan earthquake triggered 3,767 coseismic landslides, most of which had small scars (A < 0.5 km²) and limited runout (Chen et al. 2025). Coseismic landslides caused few casualties except at the Caotan landslide (hereinafter the CT landslide), a rare case involving the combined effects of seismic shaking and water. The CT landslide, with a total length of 2,917 m, a width of 650 m, and an elevation drop of 81 m, occurred on the Yellow River terraces. Its angle of reach, calculated as arctan(H/L) from the elevation drop H and horizontal travel distance L, is 1.6°. This angle is used only as a bulk mobility index and must not be interpreted as the intrinsic or equivalent friction angle of the soil. The landslide killed 19 people, buried 74 houses, and damaged approximately 3.2 km of roads and 13 ha of farmland.

Several comparable cases have been documented. In 1920, the Ms 8.5 Haiyuan earthquake triggered the giant Shibeiyuan loess landslide in Ningxia, northwestern China. The landslide covered approximately 9 km² and had an exceptionally gentle slope (2–5°) and long runout within loess-paleosol sequences (Zhang and Wang 2007; Wang et al. 2014; Xu et al. 2021a; Chai et al. 2023). On 22 July 2013, persistent rainfall followed by the Ms 6.6 Minxian earthquake induced the 420,000-m³ Yongguang landslide–mudflow in Gansu Province. The landslide initiated on a 15–22° loess slope underlain by Neogene mudstone and sandstone and rapidly evolved into a mudflow along the valley (Zhuang et al. 2020; Wang et al. 2021a). Similar behavior has been reported in pyroclastic-fall deposits and fill. Thousands of highly mobile landslides with low travel angles were triggered in pyroclastic-fall deposits by the 2018 Mw 6.6 Eastern Iburi earthquake in Japan, partly because of hydrated halloysite (Zhang et al. 2019; Li et al. 2020a; Wang et al. 2021b). The flow-like Kataragai landslide in a fill slope was triggered on a gentle slope outside the area of strongest shaking during the 2018 Mw 5.7 Western Shimane earthquake (Li et al. 2020b). These cases illustrate the high mobility that can develop when water-rich deposits are subjected to strong ground motion (Sassa et al. 1996, 2005). Their abrupt onset and long travel distances can cause severe damage; therefore, explaining the formation and long runout of the CT landslide is important for mitigating similar hazards.

Recent rapid-response studies have mapped the regional coseismic-landslide distribution and documented the field characteristics, material composition, and preliminary liquefaction mechanisms of the CT landslide (Chen et al. 2025; Liu et al. 2024; Wang et al. 2024c; Xu et al. 2025). The relationship between source-area failure, water recharge, and successive flow transitions remains uncertain. To examine that relationship, we integrate post-event field mapping, UAV photogrammetry, borehole and trench logs, standard penetration testing, and interviews to reconstruct the source geometry, slide-to-flow transition, and downstream movement. Because all observations were collected after the event and no direct pore-pressure or continuous velocity records are available, the inferred liquefaction depth, basal shear surface, and movement sequence remain interpretations that require validation by laboratory testing, additional subsurface investigation, monitoring, and dynamic simulation.

Geological setting and climate conditions

The study area is located at the northeastern margin of the Tibetan Plateau and belongs to the transition zone between the Loess Plateau and Qinghai-Tibet Plateau (Fig. 1a, b). Geotectonically, the study area lies in the southern part of the South Qilian Block, at the junction of the nearly NW-trending Qilian and West Qinling orogenic belts, and contains strike-slip and thrust faults and fault-bounded rhombic basins. These faults comprise the Lajishan Fault, the Daotanghe-Linxia Fault, and the West Qinling Fault (Fig. 1c). The Lajishan Fault, characterized by a typical imbricate thrust structure, is the principal fault traversing the region (Yuan 2004; Zhao et al. 2022), and the Lajishan North Fault is considered the seismogenic fault for the 2023 Jishishan earthquake (Wang et al. 2024b). The Lajishan Fault was highly active during the Neogene and early Quaternary, but its activity has decreased since the late Pleistocene (Zhang 2012). Seismic activity is generally low, and 20 historical earthquakes with magnitudes below 6.0 have been documented. The 2023 Jishishan earthquake is the largest recorded to date.

The vicinity of the CT landslide is widely mantled by Holocene alluvial deposits (Qh al), Holocene alluvial-proluvial deposits (Qh pal), Upper Pleistocene alluvial-proluvial deposits (Q3 pal) and loess (Q3 m1), and Lower Pleistocene alluvial-proluvial deposits (Q1 pal). The main strata outcropping in the study area comprise Miocene sandstone and silty mudstone interbedded with conglomerate (N1 m), Oligocene conglomerate and sandstone interbedded with mudstone (Et), Lower Cretaceous conglomerate and sandstone (K1 H 1), Middle Jurassic basal conglomerate, shale, and sandstone (J2 y), and Caledonian effusive and intrusive rocks (Fig. 2a). Paleogene and Neogene mudstone and sandstone, together with Upper Pleistocene loess, are the principal strata associated with landslide formation in this area.

Fig. 2
Fig. 2
Full size image

Geological setting (a) and precipitation record (b) near the study area. The pluviometer location is shown in Fig. 1c

In terms of climate, the study area is characterized by an arid continental climate with a mean annual temperature of 9 °C. The mean annual precipitation of the study area is approximately 300 mm, with the maximum daily precipitation and hourly precipitation being 142.5 mm and 42.0 mm, respectively (Ma et al. 2002). The 2023 Jishishan earthquake and the subsequent CT landslide occurred during the winter season, with daily precipitation of 10.6 mm recorded in a pluviometer approximately 7.3 km west of the CT landslide (Fig. 2b). This suggests that antecedent precipitation made little contribution to the initiation of the CT landslide.

Geometry and movement characteristics

The 15-day field campaign began on 23 December 2023, five days after the earthquake, and continued through 6 January 2024. It included UAV surveys, trench observations, borehole drilling, and in situ tests; the two trenches, two boreholes, and standard penetration test described below were completed during this campaign. The landslide boundary was delineated by comparing pre-failure satellite imagery and a digital elevation model (DEM) with post-failure UAV orthophotography and the resulting DEM (Fig. 3a, b). The total affected area was approximately 476.88 × 10³ m².

Fig. 3
Fig. 3
Full size image

Plan view of the CT landslide. a Pre-failure image obtained from Google Earth. b Post-failure image acquired by UAV. The dashed red line delineates the landslide boundary

The CT landslide originated on the third terrace of the Yellow River (T3) at an elevation of approximately 1,753 m above sea level, traveled rapidly down the Qijia Valley across T2, and reached the first terrace (T1). It descended only 81 m over a total horizontal travel distance of approximately 2,917 m and reached a maximum width of about 650 m. The resulting angle of reach, arctan(H/L), is 1.6°. Residents of Meiyi Village stated that the flow reached the vicinity of their homes shortly after the earthquake, while officials from Zhongchuan Town and residents of Jintian Village reported that it reached Caotan and Jintian villages within 10 min. An interview-based estimate of the mean velocity from the source area to the Meiyi Village dam exceeded 10 m/s, which exceeds the 5 m/s threshold for the IUGS ‘extremely rapid’ velocity class (IUGS Working Party on World Landslide Inventory 1995). We therefore classify the event as an extremely rapid, long-runout landslide–mudflow.

The landslide was further divided into sliding, flow, and deposition areas using field observations and pre- and post-failure digital orthophoto maps (DOMs). These areas were 127.13 × 10³ m², 83.81 × 10³ m², and 265.94 × 10³ m², respectively (Fig. 4a).

Fig. 4
Fig. 4
Full size image

Zonation of the CT landslide (a) and damage resulting from the long-runout movement (b)

The sliding area lies on the nearly flat third terrace (T3), where the mean slope is approximately 3.63° (Fig. 5). The main scarp is nearly straight, and most of the subtriangular source mass converged southward toward the head of the Qijia Valley; a smaller part moved eastward (Fig. 6a). Residual material forming scattered stripes records these movement directions (Fig. 6a, b). In the frontal part of the source area, away from the main scarp, the preserved morphology is consistent with coherent displacement above a possible deep basal shear zone. Here, the terms upper, middle, and lower are morphological descriptors rather than formal stratigraphic units: the upper interval lies above the relatively intact remnant, the middle interval is the residual band visible in Fig. 6b–d and may correspond to the interval of higher SPT resistance at 5–9 m below the test location, and the lower interval extends from beneath that remnant to the interpreted basal shear zone. These bounds describe the frontal source morphology; they are approximate and do not establish laterally continuous depth intervals. The upper and lower intervals collapsed, whereas part of the middle interval remained relatively intact on the sliding bed (Fig. 6b–d). The basal shear surface shown in Fig. 5 was interpreted by integrating the main-scarp geometry, exposed contacts and movement traces in Fig. 6b–d, the boundary between displaced and in situ material, pre- and post-failure DEM differencing, borehole and trench stratigraphy, and the low-resistance interval in the SPT profile. Its buried portions could not be traced directly, and the surface in Fig. 5 should be read as an approximate interpretation rather than a mapped continuous contact. The borehole and SPT data provide local supporting constraints rather than a continuous trace of the shear surface. Near the main scarp, the rear source mass underwent lateral spreading under traction from the front mass and came to rest in a wave-like pattern after a short displacement (Fig. 6e). Headward erosion was less extensive near the left flank than near the right flank (Figs. 3 and 6a). The farmland had been continuous irrigated three days until the earthquake, and the higher moisture content near the right flank may have contributed to this difference, consistent with a role for winter irrigation in shallow failure (Watkinson and Hall 2019; Xu et al. 2021b).

Fig. 5
Fig. 5
Full size image

Longitudinal profile of the CT landslide. Elevation and distance use different scales. The post-failure surface was derived from UAV data, and the pre-failure surface was derived from the freely available ALOS DEM. The blue inverted triangle marks the post-event groundwater-table position measured in the source area

Fig. 6
Fig. 6
Full size image

Details of the sliding area. a Front view of the source area; green polygons delineate damaged roads, irrigation ditches, and the transmission-tower area and indicate the movement direction. b Scattered residual stripes and the middle remnant. c–d Remaining sliding mass (middle layer). e Wave-like displaced mass at the rear of the source area

After initiation, most of the displaced mass moved southeastward and a smaller part moved eastward. The eastward-moving mass turned south after crossing a lower platform and stopped after approximately 227 m because of energy loss (Fig. 7a, b). The southeastward mass entered the Qijia Valley and disintegrated as it descended the steep valley head (Fig. 7c). Local interviews indicated a winter water depth of approximately 0.5 m in the upstream valley; water released by damaged irrigation ditches also entered the channel. Mixing with this water likely reduced the shear resistance of the fragmented mass and promoted its transition from sliding to highly mobile flow (Fig. 7d). The upstream valley slope is approximately 17°, and comparison of the pre- and post-failure DOMs shows channel widening, indicating substantial lateral erosion. Entrainment from both valley sides increased the solid volume. The flow then reached the dam near Meiyi Village. A mud mark on a house wall and the former dam geometry indicate a maximum surge height greater than 10 m (Fig. 7e). Interviews place arrival at the dam approximately 2 min after the earthquake and underpin the mean-velocity estimate discussed above. Near the village, the valley turns from southeast to south. The dam deflected the flow both northward and southward; the northward branch stopped after approximately 134 m as the ground rose (Fig. 7f). A pre-existing pond, confirmed by pre-failure imagery and residents, supplied additional water when it was damaged. Turning of the valley, blockage by the dam, a downstream slope below 5°, and relatively dry channel conditions progressively reduced the flow velocity (Fig. 7g). The surge height at the memorial archway was approximately 4 m (Fig. 7h).

Fig. 7
Fig. 7
Full size image

Details of the flow area. a Plan view and locations of representative photographs. b Eastward-moving displaced mass. c Disintegration in the upstream valley. d Transition from sliding to flow. e Mud mark preserved on a house in Meiyi Village. f Scratches on the right flank of deposition area 2, where the flow moved northward. g Downstream flow area. h Memorial archway in Caotan Village

No distinct channel was present in the Qijia Valley around Caotan Village. Beyond the memorial archway and road X881, the flow spread at low velocity and buried houses in Caotan and Jintian villages (Fig. 8a, b). Most houses and roads remained in place and were buried or partly damaged rather than completely destroyed, and the upper stories of buried houses showed little impact damage (Fig. 8c). Nevertheless, the reported flow depth of 2–3 m and sudden midnight onset caused severe casualties and losses. The deposit morphology is consistent with a transition from slurry to plastic flow in this reach (Iverson 1997) (Fig. 8d, e). Most of the mass stopped upstream of an irrigation ditch south of Jintian Village (Fig. 8f), while a smaller portion received additional water from damaged ponds and ditches, reverted to a shallow slurry flow, and traveled approximately 772 m farther across farmland (Fig. 8g, h).

Fig. 8
Fig. 8
Full size image

Details of the deposition area. a Plan view and locations of representative photographs. b Inundated areas of Caotan and Jintian villages. c Half-buried houses in Jintian Village. d Plastic-fluid flow in Jintian Village. e Plastic-flow deposits. f Irrigation ditch dammed by displaced material south of Jintian Village. g Shallow slurry deposit at the distal margin. h Distal deposition area

The proposed movement sequence comprises seven stages: coherent sliding above an inferred basal shear zone, lateral spreading, disintegration, slurry flow, plastic flow, renewed slurry flow, and final deposition.

Damage and impacts

From the source area through the Qijia Valley, the landslide damaged a transmission tower, roads, irrigation ditches, houses, ponds, and farmland. Figure 9a shows the locations of representative damage. Movement in the sliding area overturned one transmission tower (Fig. 9b), damaged four roads with a combined length of 1,180 m (Fig. 9c), and destroyed one 420-m-long major irrigation ditch and more than 10 smaller ditches (Fig. 9d). Houses on the left side of the Qijia Valley were severely damaged along the flow path (Fig. 9e, f). The greatest losses occurred in the deposition area, where buildings, roads, cropland, and other infrastructure were buried (Fig. 9g–m). Official reports and our preliminary inventory indicate that 19 people were killed and that 74 houses, 2,038 m of roads, approximately 13 ha of farmland, and two ponds were damaged or buried.

Fig. 9
Fig. 9
Full size image

Damage caused by the CT landslide. a Locations of representative photographs. b Overturned transmission tower. c Damaged road. d Damaged irrigation ditch. e–f Damaged houses in Meiyi Village. g Damaged memorial archway. h Buried vehicle. i–m Damage to houses, farmland, ponds, roads, and trees in the deposition area

Failure and initiation mechanisms

As described above, the CT landslide initiated on a nearly flat river terrace during the dry winter season, and the slope was stable under normal conditions. Cumulative precipitation during the month before the earthquake was less than 15 mm (Fig. 2b). Nevertheless, the earthquake triggered an extremely rapid landslide with an affected area of 476.88 × 10³ m² and substantial casualties. We therefore conducted detailed field investigations to evaluate the failure mechanism.

During the field campaign, two trenches were excavated and two boreholes were drilled near the main scarp and the right flank of the sliding area (Fig. 4a). The groundwater levels, moisture conditions, and seepage observations reported below are therefore post-event measurements rather than direct records of co-seismic conditions. Figures 10a and 11a show the stratigraphic sections of borehole 1 and trench 1, respectively. The boreholes and trenches indicate at least eight layers: dark-gray humus (0–0.6 m); brick-red alluvial-proluvial clay (Q4 pal, 0.6–1.0 m); dark loessial soil (1.0–1.3 m); a pebble layer (Q4 pal, 1.3–1.55 m); brick-red alluvial-proluvial silty clay (Q3 pal, 1.55–2.2 m); loess (Q3, 2.2–21.0 m); brownish-red alluvial-proluvial silty clay (Q3, 21.0–23.8 m); and gravelly sand (Q2 al, 23.8–40.0 m). The third and second terraces of the Yellow River in the study area formed at approximately 150 ka and 80 ka, respectively (Li et al. 1996; Zhao and Liu 2003), indicating that the loess in the main source mass is Late Pleistocene in age.

Fig. 10
Fig. 10
Full size image

Stratigraphic section of borehole 1 on the right flank of the CT landslide (a) and standard penetration test results (b). The borehole location is shown in Fig. 4a

Fig. 11
Fig. 11
Full size image

Trench 1 stratigraphy and geotechnical properties on the right flank of the CT landslide. a Trench profile and interpreted layers. b In situ density, dry density, and void ratio. c Water content and degree of saturation. d Insertion resistance and volumetric water content. The trench location is shown in Fig. 4a

A standard penetration test was conducted near borehole 1 to identify potential weak zones and explore liquefaction susceptibility (Trifunac 1995). The empirical comparison discussed below focuses on the upper 20 m. A 63.5-kg hammer was dropped freely from a height of 0.76 m to drive a standard sampler 0.51 m long, with outer and inner diameters of 51 and 34.9 mm, respectively. The measured blow count Nd, distinct from the reference parameter N0in Eq. (1), was recorded for each 0.3-m penetration increment. An empirical reference blow count, Ncr, was calculated using Eq. (1), which was developed for liquefaction assessment of saturated sands and silts excluding loess (CMC 2010; Liu 2016).

$$\:{N}_{cr}={N}_{0}\beta\:\left[\text{ln}\left(0.6{d}_{s}+1.5\right)-0.1{d}_{w}\right]\sqrt{3/{\rho\:}_{c}}$$
(1)

where N0 is the reference blow-count value for the design basic acceleration and takes values of 7, 10, 12, 16, or 19 for design basic accelerations of 0.10, 0.15, 0.20, 0.30, or 0.40 g, respectively; ds is the soil depth (m); dw is the groundwater-table depth (m); ρc is the clay-particle content (%), taken as 3 when the measured value is less than 3% or when the soil is sand; and β is the adjustment coefficient, taken as 0.80, 0.95, or 1.05 for the first, second, or third design earthquake group, respectively.

For the exploratory comparison, N0 = 19 and β = 1.05 were adopted. A nearby strong-motion station recorded a maximum horizontal peak acceleration of 477.2 Gal (approximately 0.487 g; Wang et al. 2024c). The adopted N0 corresponds to the highest design-acceleration category tabulated in the cited code, 0.40 g; its use for this recorded motion is an approximation. The design characteristic period used was 0.45 s. The clay contents of the brick-red alluvial-proluvial silty clay and loess are 26.28% and 12.41%, respectively (Liu et al. 2024). Borehole 1, adjacent to the penetration test, recorded a groundwater-table depth of 11 m. The resulting reference Ncr profile is shown by the green line in Fig. 10b. The measured Nd profile first decreases, then increases, decreases again, and finally increases with depth. Comparison of Nd and Ncr identifies a low-resistance interval between 9 and 17 m in this exploratory comparison, with a local minimum near the measured groundwater table. The blow count also decreases between 2 and 4 m but increases between 5 and 9 m, which is consistent with the relatively intact middle part of the source mass.

Two approximately 4.0-m-deep trenches were excavated during the field campaign to measure in situ density, dry density, volumetric water content, and insertion resistance; saturation, gravimetric water content, and void ratio were then calculated. In Fig. 11b, dry density is approximately 1.5–1.6 g/cm³ at 1.0–2.2 m depth and decreases to 1.3–1.4 g/cm³ in the underlying Q3 loess, while the void ratio rises above 0.97. Water content first increases, then decreases, and finally increases with depth (Fig. 11c); the in situ volumetric water content follows a similar pattern (Fig. 11d). Moisture increases through the surface humus and brick-red alluvial-proluvial clay, and soil at 0.8 m depth is nearly saturated. This shallow saturation (0–1.0 m) is consistent with winter irrigation before the earthquake. The low permeability and water retention of the brick-red clay helped maintain saturation in the surface layers. Moisture decreases in the silty clay and loess beneath the pebble layer. Below 2.2 m, moisture increases with depth and the loess is saturated at 3.6 m. Seepage observed below 3.0 m one day after trench excavation (Fig. 12a–c) confirms water supply to the shallow loess. Near-vertical structural fissures filled with brick-red clay (Figs. 11a and 12h) may influence water migration, although their hydraulic role cannot be established from their appearance alone. Borehole 2 on the right flank recorded a stable groundwater-table depth of 12 m, and loess at 5.0 m depth was saturated and in a plastic state. Ponding appeared in trench 2 one day after excavation (Fig. 12d). Stagnant water was observed two hours after excavation in the displaced mass and disappeared within two days (Fig. 12e). Surface water on the displaced mass one week after the event and farmland irrigated three days before the event are shown in Fig. 12f and g.

Fig. 12
Fig. 12
Full size image

Hydrological conditions of the CT landslide. a Plan view and locations of representative photographs. b Locations in the sliding area. c Seepage in loess one day after trench excavation. d Ponding in trench 2 one day after excavation at the main scarp. e Stagnant water two hours after excavation in the displaced mass. f Surface water one week after the landslide. g Farmland saturated by winter irrigation before the landslide. h Near-vertical structural fissures filled with brick-red clay in loess. i Damaged irrigation ditch. j Accumulated ice blocks transported from the Qijia Valley. k Dead fish after pond damage. l Surface water released by pond damage

Figure 11d shows the relation between insertion resistance and depth. Insertion resistance, an index of in situ soil strength, first increases and then decreases with depth. The dense, clay-rich brick-red clay above the pebble layer has relatively high resistance, whereas the brick-red silty clay and Q3 loess have lower resistance. Together with the SPT results, these measurements indicate a shallow, saturated, low-strength interval at approximately 2–5 m depth. This interval may be susceptible to cyclic weakening. The screening based on an 11-m groundwater-table depth does not directly evaluate this shallower saturated interval, and its liquefaction susceptibility requires validation by cyclic testing.

The in situ data indicate localized weak intervals in the shallow loess and near the post-event groundwater table, where cyclic liquefaction may have initiated preferentially. Low penetration resistance in parts of the 9–17 m interval, including the 9–12 m zone, provides a local mechanical constraint, but the available data do not show that the entire interval liquefied. The pre-failure source area slopes gently southward toward the Qijia Valley, where perennial water from precipitation and subsurface seepage is present. Seepage can remove fine particles, degrade soil structure, and reduce liquefaction resistance (Derbyshire 2001; Li et al. 2020b). We therefore infer that fluctuations and seepage near the groundwater table may have contributed to the low penetration resistance at 9–12 m. The saturated shallow loess at approximately 2–5 m may have received water from irrigation or subsurface flow, but the source and timing of this recharge remain unresolved. This interpretation is constrained by only two boreholes, two trenches, one SPT profile, and post-event water-level observations. It does not imply that the entire saturated deposit liquefied, and the volume of liquefied material cannot be equated with the full volume of the 9–17 m interval. Laboratory cyclic tests, additional subsurface profiles, and groundwater monitoring are needed to test the proposed mechanism and better constrain the possible extent of liquefaction.

Factors controlling initiation and movement

The CT landslide occurred under a combination of geological, topographic, hydrological, geotechnical, and seismic conditions. The principal controls are summarized below.

Soil composition: Low-density, low-strength loess provided material potentially susceptible to cyclic weakening under strong shaking. Alternating low-permeability and permeable layers could have favored locally undrained conditions (Huang and Yu 2013).

Strong earthquake: Thick, loose deposits mantle the mountainous terrain around the CT landslide. Such conditions can amplify ground motion, with reported peak-ground-acceleration amplification factors of 1.1–2.0 (Wang et al. 2019). A peak ground acceleration of approximately 0.48 g was recorded nearby (Xu et al. 2025), and the seismic intensity at the landslide was VIII (Fig. 1c). Strong shaking was the immediate trigger.

Hydrological conditions: A relatively high groundwater table and locally saturated shallow soil could have favored cyclic weakening. Winter irrigation and subsurface seepage may have contributed to pre-failure saturation, whereas water released from damaged ditches also supplied the moving mass. These conditions promoted shallow disintegration and headward erosion and supplied water during the transition from sliding to flow.

Topography and seepage: The pre-failure source area was a gentle platform connected to the Qijia Valley. The upstream channel bed, which carries perennial water, lies below the groundwater table in the source area and provided a free face for failure. The gentle southward gradient and the connection to the valley head also provided an outlet for seepage, which may have progressively reduced soil resistance. Together, these conditions help explain why the most destructive flow-like landslide in the Jishishan event occurred at this location.

The CT landslide traveled approximately 2,917 m while descending only 81 m, giving an angle of reach of 1.6°. This value is a bulk mobility index, not an intrinsic friction angle. The proposed sequence in Sect.  3 links source-area failure to successive downstream flow transitions. The principal controls on the slide-to-flow transition and long runout are summarized below.

Terrain control: The sharp descent from the source area into the Qijia Valley promoted disintegration and increased the kinetic energy of the moving mass. The narrow valley head then confined the flow along the channel. The contrast between the main runout path and deposition area 1 (Fig. 4a) illustrates the combined effects of the elevation change and channel geometry. Farther downstream, the gentle terrain around Caotan and Jintian villages allowed the flow to spread widely, contributing to the high casualty toll.

Water recharge: Winter irrigation increased the moisture content of the shallow soil, while earthquake- and landslide-induced damage to irrigation ditches supplied surface water (Fig. 12b, i). Perennial water in the upstream valley and leakage from damaged ditches promoted the transition from sliding to flow as fragmented soil mixed with water. Accumulated ice blocks near deposition area 2 were transported from the upstream valley (Fig. 12j). Two ponds in deposition area 3 were also damaged (Fig. 4b); dead fish and standing water document that they contained perennial water (Fig. 12k, l). Release from these ponds probably contributed to the renewed transition from plastic flow to slurry flow in part of the deposit and enlarged the affected area.

Solid entrainment: During downstream movement, erosion of both sides of the Qijia Valley increased the solid volume of the flow and contributed to its long runout and destructive consequences.

Conclusion

The CT landslide occurred immediately after the midnight Jishishan earthquake. Field surveys, UAV mapping, borehole drilling, trench observations, and in situ tests were used to characterize its topography, geological and hydrological conditions, source structure, and material composition. The main findings are as follows:

  1. 1.

    The CT landslide was initially described as a ‘sand boil’, a term widely used in media reports. Field evidence instead indicates that the moving material should be classified as a landslide–mudflow.

  2. 2.

    The CT landslide developed across Yellow River terraces T3, T2, and T1. The displaced mass mainly comprises brick-red alluvial-proluvial clay (Q4 pal), a pebble layer (Q4 pal), brick-red alluvial-proluvial silty clay (Q3 pal), and loess (Q3). It descended only 81 m, reached a maximum width of approximately 650 m, and traveled approximately 2,917 m. The sliding, flow, and deposition areas are 127.13 × 10³ m², 83.81 × 10³ m², and 265.94 × 10³ m², respectively. Its 1.6° angle of reach indicates high bulk mobility but is not a material friction angle.

  3. 3.

    The observations are consistent with localized cyclic weakening, potentially including liquefaction, in low-density, low-strength, highly saturated loess during initiation. A high groundwater table, winter irrigation, and subsurface seepage may have provided favorable hydraulic conditions, while low-permeability clay layers could have favored locally undrained behavior. The gentle terrace and the steep free face at the valley head favored release of the displaced mass. The occurrence, depth, and volume of liquefaction require further verification because subsurface coverage was limited, water levels were measured after the event, and the empirical screening was not validated for the site loess.

  4. 4.

    The proposed reconstruction links coherent sliding and lateral spreading in the source area to disintegration, slurry flow, plastic flow, renewed slurry flow, and deposition. The sharp descent into the Qijia Valley and water supplied by damaged irrigation ditches, valley flow, and ponds promoted the slide-to-flow transitions. Bends in the confined valley and gentle downstream terrain reduced velocity, while damage to the ponds probably contributed to renewed slurry flow and farther runout.

Cyclic liquefaction of saturated loess has been reported worldwide. The resulting flow-like landslides can have abrupt onset and long runout, causing substantial casualties and losses. Liquefaction potential in loess-covered regions should therefore be systematically assessed using site-specific topographic, stratigraphic, geotechnical, and hydrological evidence. This study provides a field-based reconstruction of the CT landslide, but it is limited by post-event observations, two boreholes, two trenches, one SPT profile, and the absence of direct pore-pressure and velocity records. Future work should combine laboratory cyclic testing, groundwater monitoring, additional subsurface investigation, and dynamic simulation to test the proposed liquefaction and slide-to-flow mechanisms.