Landslide Susceptibility and Runout Modelling Along the Karakoram Highway Using PS-InSAR and Hybrid CNN-LSTM–XGBoost
Rock Mechanics Letters · 23.01.2027 · 4 (1) · 451-461
DOI: 10.70425/rml.202701.55 ↗Kurz gefasst
Fachartikel in der Zeitschrift Rock Mechanics Letters, online erschienen am 23.01.2027. Verfasst von Mir Mujtaba Laghari und 2 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The Karakoram Highway (KKH) traverses a tectonically active and climatically sensitive mountain corridor that is repeatedly affected by landslides. This study develops an integrated hybrid framework combining convolutional neural network–long short-term memory (CNN-LSTM) learning, extreme gradient boosting (XGBoost), persistent-scatterer interferometric synthetic aperture radar (PS-InSAR), runout modelling, climate projections, and multi-hazard risk assessment along a 320-km section of the KKH in northwestern Pakistan. Eighteen topographic, hydrological, geological, climatic, anthropogenic, and geodetic conditioning factors were used. The CNN-LSTM was pre-trained on the NASA Cooperative Open Online Landslide Repository and fine-tuned using an inventory of 612 landslides, while class imbalance was addressed using SMOTE and focal loss. The stacked hybrid model achieved an AUC of 0.974, accuracy of 0.928, and Cohen’s κ of 0.856, outperforming standalone CNN-LSTM (AUC = 0.951) and XGBoost (AUC = 0.938) models. SHAP and Grad-CAM analyses identified slope, PS-InSAR line-of-sight velocity, annual rainfall, and elevation as the dominant controls. Flow-R and RAMMS simulations reproduced representative runout patterns with κ = 0.74. Under CMIP6 SSP5-8.5, high-to-very-high susceptibility is projected to increase from 24.0% at baseline to 35.2% by 2050 and 41.6% by 2100. The multi-hazard assessment further indicates that 23.7% of the highway lies within high-to-very-high composite risk zones. The framework provides an explainable, climate-sensitive, and operationally transferable basis for landslide-risk reduction along high-relief transport corridors.
Zitieren
APA
Laghari, M. M., Saeed, K., & Cheng, N. (2027). Landslide Susceptibility and Runout Modelling Along the Karakoram Highway Using PS-InSAR and Hybrid CNN-LSTM–XGBoost. Rock Mechanics Letters, 4(1), 451-461. https://doi.org/10.70425/rml.202701.55
Vancouver
Laghari MM, Saeed K, Cheng N. Landslide Susceptibility and Runout Modelling Along the Karakoram Highway Using PS-InSAR and Hybrid CNN-LSTM–XGBoost. Rock Mechanics Letters. 2027;4(1):451-461. doi:10.70425/rml.202701.55
BibTeX
@article{Laghari2027,
author = {Laghari, Mir Mujtaba and Saeed, Khuzaima and Cheng, Nirandoal},
title = {Landslide Susceptibility and Runout Modelling Along the Karakoram Highway Using PS-InSAR and Hybrid CNN-LSTM–XGBoost},
journal = {Rock Mechanics Letters},
year = {2027},
volume = {4},
number = {1},
pages = {451--461},
doi = {10.70425/rml.202701.55}
}
EUVIMED