In karst mountainous areas where high-dimensional features coexist with extremely limited sample sizes, accurate landslide susceptibility mapping remains challenging. To address this issue, we propose an ensemble framework termed the Triple-Source Probabilistic Fusion Bidirectional Long Short-Term Memory network (TSPF-BiLSTM). The approach was tested in Sangzhi County, Hunan Province, by integrating three base learners—Random Forest (RF), LightGBM, and AdaBoost. Their raw outputs were first calibrated using five-fold Platt scaling to generate posterior probabilities on a unified scale. A bidirectional LSTM was then employed to perform deep nonlinear fusion of these cross-model probability features. Using a total of 618 landslide and 618 non-landslide samples (split into training and testing sets), the TSPF-BiLSTM model achieved a mean AUC of 0.9525 (±0.0115) under ten-fold cross-validation, outperforming not only the individual base learners but also standalone deep learning models (CNN and Transformer). The frequency ratio in the very high susceptibility zone reached 3.97, significantly exceeding all benchmark models and confirming its superior capability in high-risk area identification. Multi-model importance analysis identified NDVI, elevation, and annual rainfall as the dominant regional landslide predisposing factors. Within the specific ranges of NDVI 0–0.686, elevation 155–462 m, and annual rainfall 1273.6–1301 mm, landslide frequency ratios consistently exceeded 1.96. The proposed framework, with its probability-level fusion and embedded regularization mechanisms, effectively mitigated overfitting despite the small sample size, providing a robust technical solution for geological hazard risk identification and prevention in the data-scarce karst terrain of the Wuling Mountains.

The Tibetan Plateau, formed by the Indian-Eurasian collision, is dissected by the north-south trending South Tibet Rift system, but the formation mechanism of these rifts within a continuing compressional setting remains debated. Using P-wave receiver functions and joint inversion with surface wave dispersion along a ~230 km broadband seismic profile crossing the Nyima-Tingri Rift (NTR) and Xianza-Dingjie Rift (XDR), we investigated the detailed crustal structure beneath the Himalayan and Lhasa terranes. Our results reveal three key findings: (1) The crustal thickness ranges from 60 to 80 km, with the XDR exhibiting a pronounced Moho uplift (~10 km) beneath the rift axis, whereas the Moho beneath the NTR remains flat and continuous, indicating contrasting evolutionary stages. (2) A mid-crustal low-velocity layer at ~30 km depth is consistently observed west of 87.2° E, suggesting the presence of partial melt or fluids that decouple upper crustal deformation from mantle flow. (3) A prominent intracrustal discontinuity at 50–70 km depth produces a “Moho doublet” pattern; we interpreted this as the subducted Indian lower crust in the Himalayan terrane but as the relict Tibetan Moho in the Lhasa terrane, reflecting progressive northward modification. Collectively, these observations demonstrate that the north-south structures in southern Tibet lack the deep structural characteristics of mature rifts and instead represent an “infant stage” of extension. Their formation is best explained by east-west crustal stretching driven by an ongoing north-south convergence and eastward flow of lower crustal and upper mantle materials rather than by classical lithosphere-scale rifting.

Landslides pose significant threats to sustainable development by causing infrastructure damage and ecosystem degradation, particularly in densely vegetated mountainous regions. To support sustainable land-use planning and disaster-resilient development, this study integrates three advanced vegetation metrics—Vegetation Formation Group (VFG), aboveground biomass (AGB), and forest canopy height (FCH)—into landslide susceptibility modeling. Using Yuanling County, a subtropical vegetated region in China, as a case study, we developed a novel ensemble model, AdaBoost-CB (AdaBoost-CatBoost), and compared its performance with mainstream machine learning models including RF, XGBoost, and LGB. The results show that AdaBoost-CB achieved the highest Area Under the Curve (AUC) value of 0.915. Furthermore, it yielded the highest landslide frequency ratio of 6.51 in the very-high-susceptibility zones. The dominant landslide-controlling factors—NDVI, elevation, slope gradient, slope aspect, and rainfall—were consistently identified across six models. These findings provide a scientific basis for sustainable land-use planning and disaster risk reduction strategies, contributing directly to the goals of sustainable development in vulnerable mountainous regions.