Abstract

Rainfall-induced landslides are a significant geological hazard, causing severe economic losses and casualties. Accurate forecasting of these events is particularly challenging due to the complex interactions of spatial and temporal factors governing slope stability. The Iverson model, which uses the Richards equation to describe water infiltration in unsaturated soils, is a widely adopted framework for analyzing rainfall-induced landslides. However, its reliance on traditional numerical methods limits its scalability and efficiency, particularly for complex boundary conditions and transient behaviors near slope failure. To address these limitations, we propose a physics-informed neural network (PINN) enhanced with transfer learning (TL-PINN) to solve the Iverson model for landslide forecasting. The TL-PINN employs a transfer learning strategy, freezing specific hidden layers to improve performance in capturing the dynamics of slope stability under rainfall. This TL-PINN method predicts critical parameters such as the pressure head distribution, factor of safety (FS), and failure timing by efficiently solving the Richards equation. Validation through a case study demonstrates the ability of TL-PINN to handle discontinuous top-boundary conditions, sharp gradients, and nonlinear phenomena, outperforming traditional PINNs. The results highlight the potential of TL-PINN to advance the application of the Iverson model, providing valuable insights for analyzing and forecasting rainfall-induced landslides.

Department(s)

Geosciences and Geological and Petroleum Engineering

International Standard Serial Number (ISSN)

0895-0563

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2025 American Society of Civil Engineers, All rights reserved.

Publication Date

01 Jan 2025

Download

Full Text Link

Share

 
COinS