Answers are grounded in this paper and cite their sources.
As a highly destructive geological hazard, landslides require accurate monitoring, and prediction for effective disaster mitigation. However, many current deep learning‐based methods for predicting landslide displacement still struggle to achieve high accuracy and to explain landslide behaviors. To address this long‐standing problem, we proposed a generative deep learning model with high accuracy and explainability for continuously predicting landslide displacement. Herein, we used 10‐day resolution and both displacement and pore pressure data. To enhance explainability, our model integrates the pore (water) pressure, which can modulate landslide displacement, by estimating the landslide acceleration probability. Furthermore, the structure of the generative model facilitates the identification of long‐term patterns and enables a more accurate prediction of landslide displacement. Experiments on two typical slow‐moving landslides in Oregon monitored with extensometers and piezometers show that compared with the best‐performing baseline models, MG‐DS reduces Root Mean Square Error and Mean Absolute Error by 11.8% and 16.1% for the Johnson Creek landslide and by 21.2% and 31.5% for the Carmel Knoll landslide, respectively. The model advantages manifest in three aspects: first, the generative model uses parallel prediction and iterative refinement to effectively enhance the accuracy; second, the model is able to identify and leverage pore pressure–displacement variations; and third, the use of a temporal decomposition architecture balances explainability and accuracy. Our study not only demonstrates that explicitly integrating pore pressure improves prediction performance but also shows that generative models provide a promising approach for landslide displacement prediction.
No takes yet. Share an insight, caveat, or question.
Hu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: