Displacement prediction of landslides at slope-scale: Review of physics-based and data-driven approaches

In this paper, a critical review of the landslide displacement prediction is conducted, based on a database of 359 articles on landslide displacement prediction published from 1985 to 2023. The statistical analysis of this database shows that the methods taken for the landslide displacement prediction could be categorized into physics-based and data-driven approaches. In the context of the physics-based approaches, the displacement of a landslide is characterized and predicted by a physics-based model that approximates the deformation mechanism of the landslide; whereas, the displacement, in the data-driven approaches, is often characterized and predicted by a mathematical or machine learning model, established based on analyses of the historical data. Note that although physics-based approaches were generally adopted in the early studies, data-driven approaches are becoming more and more popular in recent years. The main components involved in the physics-based approaches, including principles for establishing the prediction model, determination of model parameters, solution strategies of the model built, evaluation of the model's predictive performance, are first reviewed based on the literature database; then, those of the data-driven approaches, including methods for pre-processing the landslide displacement and influencing factors, algorithms for establishing the prediction model, calibration of model parameters, probabilistic prediction methods of landslide displacement, and evaluation of the model's predictive performance, are analyzed. Based on analyses of the information collected from the literature and our experience, we further discuss the challenges faced in landslide displacement prediction and offer recommendations for future research. We suggest that a hybrid prediction framework that takes advantage of both physics-based and data-driven approaches, a multi-field and multi-parameter landslide monitoring scheme, and an efficient strategy for the calibration of model parameters warrant further investigations.