The Yizhong River Basin, located in an alpine gorge on the southeastern Tibetan Plateau, experiences recurrent landslide-derived debris flows originating from moraine-covered source areas. This study integrates unmanned aerial vehicle (UAV)-based geomorphic interpretation, field geological profiling, laboratory direct shear tests, rainfall analysis, and sequential SEEP/W–FLAC3D modeling to examine source-area preconditioning and rainfall-related slope response. Present-day UAV-derived profiles and field observations support a two-layer configuration in which unconsolidated moraine deposits overlie fractured bedrock and are consistent with an inferred sequence of headward incision, gully deepening, and lateral expansion; this sequence is interpreted conceptually rather than as a chronologically resolved reconstruction. The moraine cover represents a substantial high-elevation sediment source, with an approximate potential volume on the order of 3.9 × 10⁵ m³. Laboratory wetting from the natural state (6.3% water content) to a controlled wet state (16.8%) reduced cohesion from 24.5 to 15.3 kPa and the internal friction angle from 30.2° to 26.4°. Under an idealized 8 h equal-total rainfall scenario, the sequential seepage–stability model yielded a factor-of-safety decrease from 1.33 to 0.92 and a maximum downward displacement of 22.3 cm, with deformation concentrated near the mapped collapse–slide zone. The recurrent activity is therefore interpreted as a feedback pathway linking moraine-slope instability, gully erosion, sediment recharge, and renewed failure. The results provide a case-based assessment of how moraine–bedrock preconditioning and rainfall-related weakening may jointly contribute to recurrent source-area instability, highlighting the combined roles of geological structure, sediment supply, and hydrological weakening in debris-flow initiation and recurrence.

Deep excavation reduces lateral confinement and can increase tangential stress, promoting brittle failure in highly stressed rock. However, the effects of bolt number and spacing on mechanical response and damage evolution under this stress path remain insufficiently understood. In this study, parallel RFPA^3D simulations were conducted for true-triaxial single-face unloading followed by continued axial loading, assuming perfect bonding at the bolt–rock interface. Two parameter series examined bolt numbers from zero to four (M0–M4) and bolt spacings from 10 to 80 mm. Stress–strain response, acoustic emission activity, failure morphology, damage area ratio (A_d), and box-counting fractal dimension (D_f) were analysed. Bolt reinforcement increased post-unloading peak stress and restricted the inward extent of damage, while damage persisted near the unloaded face and within the inter-bolt regions. Peak stress increased from approximately 345 MPa in M0 to 363 MPa in M3, then decreased slightly to 358 MPa in M4, showing that reinforcement gains were not proportional to bolt number. Bolt spacing had a modest effect on peak stress but a clearer influence on damage continuity and inward extent. Among the tested spacings, the 40 mm case (D40) achieved the highest peak stress of 370.0 MPa and the lowest A_d of 23.85%, with a low D_f of 1.431. In contrast, the 80 mm case (D80) exhibited more extensive and spatially complex damage (A_d = 30.78%, D_f = 1.677). A_d and D_f provided complementary descriptions of damage extent and spatial complexity. These numerical results provide a quantitative basis for comparing reinforcement configurations under the investigated specimen-scale conditions.

Climate-driven extreme rainfall is increasing landslide risk, whereas conventional regional rainfall thresholds inadequately represent spatial variations in slope preconditioning across complex mountainous terrain. This study develops a dynamic early-warning framework that couples slope unit landslide susceptibility with rainfall triggering for Jinyang County, Southwest China. The county was divided into 7249 slope units, and eight conditioning factors were used to train support vector machine (SVM) and extreme gradient boosting (XGBoost) models with Bayesian hyperparameter optimization. Bayesian-optimized XGBoost (BO-XGBoost) achieved the best discrimination, with an area under the curve (AUC) of 0.893. For temporal triggering, 113 historical rainfall-landslide events were analyzed in logarithmic intensity–duration (I–D) space. A power-law fit described the central trend, and the fifth percentile of residuals defined a 5% non-exceedance threshold exceeded by approximately 95% of triggering events. Residual-based thresholds were used to establish four rainfall-trigger levels. A gated risk matrix coupled these levels with susceptibility classes. Validation using 162 online monitoring sites during the 21 August 2023 extreme rainfall event yielded 85.2% accuracy, 83.6% precision, 81.2% recall, 88.2% specificity, and an F1 score of 82.4%, demonstrating practical potential for spatially refined regional landslide early warning.