AUTHOR=Li Chaofei , Li Tianbin , Lan Fuan , Ren Yang , Wen Yan , Cai Wencheng 

TITLE=The evaluation of landslide comprehensive susceptibility based on stacking ensemble learning fusion model and SBAS-InSAR: a case study in lexi highway

JOURNAL=Frontiers in Earth Science

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2025.1675848

DOI=10.3389/feart.2025.1675848

ISSN=2296-6463

ABSTRACT=IntroductionFrequent landslides along the Lexi highway have significantly hindered the construction and operation of engineering projects, impeding regional development. This study aims to clarify the distribution patterns and regional risks of these landslides to support risk management.MethodsAn ensemble learning fusion model, combining Random Forest (RF) and Extreme Gradient Boosting (XGBoost) via a Stacking algorithm, was first constructed to evaluate landslide susceptibility. Subsequently, the SBAS-InSAR method was applied to analyze long-term Sentinel-1A ascending and descending orbit data to determine surface deformation rates. Finally, a comprehensive susceptibility evaluation matrix was developed by integrating the susceptibility results with the deformation rates to generate a landslide comprehensive susceptibility map.ResultsLandslide sites are densely distributed along the Lexi highway, with an areal density of 15 landslides per 100 km2 and a linear density of 0.89 landslides per kilometer; The influence of distance to the fault zones, human activity intensity and rainfall on the distribution of landslides along the Lexi highway is the most significant, with the importance indexes of 0.27, 0.24, 0.21, respectively; Compared to other models, the Stacking ensemble learning fusion model shows superior predictive performance and generalization ability, achieving an AUC of 0.977 in evaluating landslide susceptibility along the Lexi highway; The landslide comprehensive susceptibility map effectively identifies regions with significant deformation, reducing very low and low susceptibility zones while increasing very high susceptibility zones by about 1.1%.DiscussionThe ensemble learning fusion model with InSAR-derived deformation data significantly improved the accuracy of the landslide susceptibility assessment. This comprehensive approach effectively reduces false alarms in areas with intensive engineering and high deformation rates, providing a more scientifically-grounded basis for landslide risk prevention and control along the Lexi highway.