The increasing global demand for food necessitates the continuation of flood-irrigation-based agricultural techniques, especially in areas with river terraces and neotectonic activity. This practice, however, increases the risk of catastrophic landslides triggered by earthquakes and subsequent liquefaction. The Jintian Village landslide-mudflow, caused by the 2023 Jishishan earthquake (Ms 6.2), exemplifies this risk. This event, occurring on a seemingly stable, low-angle (<3°) slope and traveling ~2.8 km, resulted in significant casualties and highlights a critical gap in our understanding of earthquake-induced landslides in low-angle terrain. Previous landslide hazard assessments have significantly underestimated the risk in such areas. This paper aims to systematically investigate the failure modes, causes, and kinematic processes of the Jintian Village landslide-mudflow, focusing on the interplay of geomorphology, tectonics, and human activities. This study provides critical insights into the mitigation of similar catastrophic events worldwide.

Existing research on earthquake-induced loess liquefaction in China is limited, partly due to the typically deep groundwater levels before widespread agricultural irrigation. Notable past examples include the Shibeiyuan landslide triggered by the 1920 Haiyuan earthquake and low-angle landslides in Indonesia's 2018 Palu earthquake zone, where irrigation raised groundwater levels. However, the geological conditions, especially tectonics, geomorphology, stratigraphy, and hydrology, of the Jintian Village landslide differ significantly from these previously studied cases. The existing literature does not fully address the dual liquefaction layer failure mode observed in Jintian Village, nor does it sufficiently consider the combined impact of long-term tectonic activity and modern agricultural practices. This lack of comprehensive understanding necessitates the present research.

This study integrates various methodologies to investigate the Jintian Village landslide-mudflow. Remote sensing techniques, including high-resolution Google Earth satellite imagery (0.48 m) and DJI M300 UAV orthophotos (0.03 m) and digital surface models (DSM, 0.14 m resolution), were employed to map the river terraces and characterize the landslide's spatial extent and features. Pre- and post-landslide DEM data (resampled to 1 m resolution) were used to calculate the thickness of the sliding source area. Topographic profile lines were analyzed to identify surface runoff channels, while Spatial Analyst Tools (ArcGIS) helped map the water system. A detailed field investigation was conducted to map preferential channels, using data from sinkholes, excavation profiles, fracture planes, exposed sidewalls and backwalls, and sidewalls of liquefaction channels. The thickness of stratigraphic units (overburden and loess) was measured using exploratory pits and exposed profiles, creating variation curves for overburden thickness and burial depths of red clay and sand layers. Electrical resistivity tomography (ERT) was employed to analyze soil moisture content in both strongly and weakly irrigated areas. X-ray diffraction (XRD) and scanning electron microscopy (SEM) were used to analyze the mineralogical composition and microstructure of loess samples. Finally, undrained strength and permeability coefficient tests were performed on collected soil samples to assess the soil properties.

The study revealed a complex failure mode for the Jintian Village landslide-mudflow, dominated by a dual liquefaction layer of loess and sand. The liquefaction was not uniformly distributed, with variations observed across the sliding source area. The contact relationships between stratigraphic units showed unconformities resulting from liquefaction of different layers. The ERT results indicated that the loess in the strongly irrigated areas was significantly more saturated than in weakly irrigated areas, confirming that agricultural irrigation played a major role in increasing the saturation of the loess. The presence of a widely distributed network of pre-existing cracks, likely from past seismic activity, facilitated rapid water infiltration through preferential channels, leading to high saturation of the loess. These channels enhanced water-soil interaction, altering the loess fabric and reducing its liquefaction resistance strength. The liquefaction of the sand layer within the overburden also contributed to the overall failure. The kinematic analysis showed the landslide-mudflow as a low-angle, ultra-long-distance event, influenced by factors such as an initial high-energy mudflow that destroyed an earth dam, the presence of an ice layer (initially) reducing friction, and the narrowing of flow outlets. The later, ultra-long-distance mudflow resulted from a lower bulk density caused by a mixture of loess, overburden, and groundwater. The boundaries of the sliding source area were found to be linked to the geometry and distribution of loess liquefaction channels, further influenced by the spatial variations in overburden thickness. The overburden thickness partly controlled the extent of the liquefaction zone, with thicker overburden leading to greater lateral spreading.

The findings highlight the critical role of human activities, specifically prolonged flood irrigation, in exacerbating the susceptibility of loess river terraces to earthquake-induced liquefaction. The pre-existing tectonic cracks provided preferential pathways for irrigation water infiltration, dramatically increasing the saturation of the loess and weakening its strength. The combined effects of tectonics, prolonged irrigation, and the resulting changes in material fabric created conditions ideal for the dual liquefaction of loess and sand, leading to the catastrophic landslide-mudflow. The low-angle, ultra-long-distance movement of the mudflow was facilitated by the interaction between the initial high-energy mudflow and the topography. These findings emphasize that traditional landslide hazard assessments that do not account for the interaction between human activities and geological conditions significantly underestimate the risk of large-scale landslides in low-angle terrain.

The Jintian Village landslide-mudflow event represents a new understanding of earthquake-induced landslides in loess regions with extensive flood irrigation. The dual liquefaction failure mode and the influence of preferential channels, resulting from both tectonic activity and long-term irrigation, were highlighted. The study underscores the need for revised landslide risk assessments considering the interplay of geological conditions and human land use practices. Future research should focus on developing improved predictive models for such events, integrating geological data with hydrological and land use information. Mitigation strategies should prioritize transitioning from flood irrigation to more efficient techniques like drip or micro-irrigation and managing land use to reduce water infiltration.

While this study provides a comprehensive analysis of the Jintian Village landslide-mudflow, several limitations should be noted. The study primarily focuses on the Jintian Village event; thus, its generalizability to other regions might be limited. The long-term impact of the event on the region's hydrological cycle and soil properties requires further investigation. Further research is needed to quantify the exact contribution of each factor (tectonic activity, pre-existing cracks, irrigation practices) to the overall failure. More detailed microstructural analysis of the soil samples might provide a more detailed understanding of the changes in fabric during the water infiltration process.