Long-term irrigation promotes the infiltration of water in the thick, stratified loess layer, significantly raising the groundwater table and triggering a series of landslides in loess platform areas. The soil–water characteristic curve (SWCC) of loess buried at different depths affects the unsaturated infiltration process and is intricately connected to the soil’s microstructure. The SWCCs, scanning electron microscope (SEMs), and pore size distributions (PSDs) for five sets of undisturbed loess samples at depths ranging from 3.4 to 51.9 m are shown in this paper. The results indicate that the fitting parameter air entry value (AEV) of the SWCC rises from 13.67 kPa to 40.19 kPa as the depth increases from 3.4 to 51.9 m. And the saturated volumetric water content drops by 10.9%, with a notable SWCC shape difference between the transition and residual zones observed. Additionally, the total porosity of undisturbed loess falls by 12% when the depth increases from 3.4 to 51.9 m, while the macropores and mesopores reduce by 3.6% and 12.1%, respectively. These findings highlight the control of the pore structure on the SWCC and emphasize the correspondence between the SWCC and PSD. The conclusions also illustrate that the compaction effect changes the microstructure characteristics of loess, thereby affecting the soil’s water retention behavior.

Frequent agricultural irrigation events continuously raise the groundwater table on loess platforms, triggering numerous loess landslides and significantly contributing to soil erosion in the Chinese Loess Plateau. The movement of irrigation water within the surficial loess layer is crucial for comprehending the mechanisms of moisture penetration into thick layers. To investigate the infiltration and evaporation processes of irrigation water, a large undisturbed soil column with a 60 cm inner diameter and 100 cm height was extracted from the surficial loess layer. An irrigation simulation event was executed on the undisturbed soil column and the ponding infiltration and subsequent evaporation processes were systematically monitored. A ruler placed above the soil column recorded the ponding height during irrigation. Moisture probes and tensiometers were installed at five depths to monitor the temporal variations in volumetric water content (VWC) and matric suction. Additionally, an evaporation gauge and an automatic weighing balance measured the potential and actual evaporation. The results revealed that the initially high infiltration rate rapidly decreased to a stable value slightly below the saturated hydraulic conductivity (K_s). A fitted Mezencev model successfully replicated the ponding infiltration process with a high correlation coefficient of 0.995. The monitored VWC of the surficial 15 cm-thick loess approached a saturated state upon the advancing of the wetting front, while the matric suction sharply decreased from an initial high value of 65 kPa to nearly 0 kPa. The monitored evaporation process of the soil column was divided into an initial constant rate stage and a subsequent decreasing rate stage. During the constant rate stage, the actual evaporation closely matched or slightly exceeded the potential evaporation rate. In the decreasing rate stage, the actual evaporation rate fell below the potential evaporation rate. The critical VWC ranged from 26% to 28%, with the corresponding matric suction recovering to approximately 25 kPa as the evaporation process transitioned between stages. The complete evaporation process was effectively modeled using a fitted Rose model with a high correlation coefficient (R² = 0.971). These findings provide valuable insights into predicting water infiltration and evaporation capacities in loess layers, thereby enhancing the understanding of water movement within thick loess deposits and the processes driving soil erosion.

Loess landslides represent a severe geohazard in the Loess Plateau, posing significant challenges to sustainable development and regional safety. This study analyzed 111 landslides (1976–2018) and 529 marginal fissures in the South Jingyang Tableland (SJT) to reveal the distribution and failure modes of the landslide cluster. Preparation factors, including stratigraphic lithology, geomorphic features, and discontinuity systems (vertical/tectonic joints, unloading fissures, weak layers, and marginal fissures), were evaluated alongside triggering factors such as precipitation, irrigation, and river erosion. The main takeaways of this study are as follows. (1) A comprehensive database of 111 loess landslides (1976–2018) and 529 marginal fissures was constructed for the SJT, representing the longest and most complete chronological record available for this region. (2) Landslide activity within the Jiangliu landslide cluster (LC) exhibits an approximately decadal clustering pattern, with a temporal spacing of 9.6 ± 2.8 years between successive active periods, providing a basis for medium-term hazard forecasting in comparable settings. (3) The evolution of marginal fissures is controlled by the margin’s geometry and their interaction with slope failure processes, and the strong correlation between fissure development and landslide initiation enables the use of fissure monitoring as an early warning indicator. (4) The evolutionary mechanism follows a sequence of initial tectonic activity → rising groundwater → formation of the critical slide structure → water accumulation in fissures → single landslide → subsequent clustering. This process, termed a “water-induced loess catastrophe” (WILC), provides a conceptual framework for understanding landslide clustering on loess tablelands. (5) The database construction methodology, the decadal clustering analysis, and the WILC model collectively offer transferable tools for medium-term hazard forecasting, fissure-based early warning, and sustainable land-use planning on loess tablelands, not only in the SJT, but also in other loess regions with similar geological and hydrological settings.