Uncertainty Quantification of Landslide Susceptibility Mapping Considering Landslide Boundary Geometry

Landslide Susceptibility Mapping (LSM) plays an important role in identifying, characterizing, and managing landslide disasters. The uncertainty of LSM is known associated with several geo-variables, including geology, land use, topography, vegetation cover, among others. However, the effect of the varying geometry of mapped landslide boundaries on the accurate characterization of landslide events has not been fully investigated in the literature. In this study, we performed a sensitivity test of a landslide susceptibility model, based on the varying complexity of mapping boundaries, including triangular, circle, square, and irregular shapes of mapping polygons. We trained a Bayesian network model to evaluate the posterior landslide probability for the Western Oregon area using model variables including slope, elevation, curvature, relief, land use/cover, soil types, and precipitation, and the model data consists of landslides inventories, established according to different complexity levels of mapping polygons. Results show that the circular and irregular polygons generated model predictions with higher precision and accuracy compared to those associated with triangular and square polygon shapes. The results of this analysis can serve as quantitative guidance for the faithful characterization of future landslide events and set the basis for uncertainty quantification of other participating sources in landslide susceptibility analysis.

Keywords: Landslide assessment, landslide susceptibility mapping, Bayesian network, uncertainty quantification.