Landslide displacement prediction using kinematics-based random forests method: A case study in Jinping Reservoir Area, China

Abstract Empirical and numerical methods are widely used in the landslide and slope forecasting fields. The combination of the two methods for prediction is rare. This study proposes a displacement prediction method based on the Verhulst inverse function (VIF) and the random forest (RF) algorithm. The VIF model is applied to describe the kinematic behavior of landslides and slopes based on the three-stage creep deformation. The fitting displacement curve has a smoothing trend and this trend can encompass the fluctuating features when the nonlinear simulation involves the influences of external factors. The effects of external factors on displacement are quantified by the RF algorithm. The training and predicting results are obtained by the combination of VIF and RF, called VIF-RF model. The VIF-RF model has a skill in predicting the long-term period and judging the deformation stage. The approaches are illustrated using a reservoir landslide dataset along a high head dam reservoir bank in Southwestern China, in which landslides deformation were correlated to the reservoir levels and rainfall. The VIF-RF model is established by the daily recorded displacement and indexes of rainfall and reservoir level in the 1274-day period. The fitted rate (V′) and acceleration (a’) indicate that the landslide deformation is under the primary or secondary creep stage. The accuracy and reliability of the developed VIF-RF model are manifested by the error analysis, which is performed with the root mean square error (RMSE) and mean absolute percentage error (MAPE). The future displacement was predicted by the hybrid model. The features of accelerated deformation in the future 564 days agree well with the records in the monitoring periods. The VIF-RF model can provide a novel insight that can be used to forecast the future displacement for landslides and slopes.