Predicting rainfall induced slope stability using Random Forest regression and synthetic data

Water fluxes in slopes are affected by climatic conditions and vegetation cover, which influence the effective stress and stability. The vegetation cover is the intermediate layer between the atmosphere and the slope surface that alter water balance in the slope through evapotranspiration and leaf interception. This paper studies the datadriven approach for predicting the macro stability of an example grass-covered dike based on actual data and also synthetic data provided by numerical modelling. Two numerical models are integrated in this study. The water balance in the root zone is simulated through a crop model, whereas the hydro-mechanical and safety analysis of the example dike is done using a two-dimensional Finite Element model. The considered period for these analyses is 10 years (3650 daily instances) which will be used to generate a time-series dataset for a secondary dike in The Netherlands. The features included in the dataset are parameters that (i) have a meaningful relationship with the dike Factor of safety (FoS), and (ii) can be observed using satellite remote sensing. The output dataset is used to train a Random Forest regressor as a supervised Machine Learning (ML) algorithm. The results of this proof-of-concept study indicate a strong correlation between the numerically estimated FoS and the MLpredicted one. Therefore, it can be suggested that the utilized parameters can be used in a data-driven predictive tool to identify vulnerable zones along a dike without a need for running expensive numerical simulations.