Rapid prediction of rainfall-induced landslides over a wide area aided by a simulation-based surrogate model

This study proposes a framework for rapidly predicting rainfall-induced landslides over a wide area by using a precomputed dataset obtained from mechanics-based numerical simulation. Slope stability is evaluated for various rainfall scenarios by combining infiltration analysis and three-dimensional limit equilibrium analysis to obtain the spatial distribution of landslide hazard based on the factor of safety, and the results are stored in a database. The proper orthogonal decomposition is applied to the matrix created from the database to analyze the spatial distribution characteristics (spatial modes) of the dataset, and the spatial distribution of the factor of safety is expressed as a superposition of the obtained spatial modes. Additionally, the extracted spatial modes are utilized to create a surrogate model using Gaussian process regression. The created surrogate model can evaluate the spatial distribution of landslides under arbitrary rainfall conditions with a low calculation cost. The performance of the proposed framework was investigated by applying the surrogate model to landslide hazard analysis using actual terrain data, confirming that it can accurately reproduce the numerical analysis results.