Abstract: In order to quantitatively analyze the time-lag effect between displacement and inducing factors of step-like landslide, as well as to improve the accuracy of displacement prediction, in this study, the authors proposed a new prediction model and conducted comparative analysis. First, the cumulative displacement was separated into trend term and periodic term based on time series analysis. Then, using maximum information coefficient (Cmi) and multivariate empirical mode decomposition (MEMD) for multi-factor analysis and multi-scale analysis, the multi-factor and multi-scale MEMD prediction model was constructed. Finally, taking  Bazimen landslide in  Three Gorges Reservoir area as an example, the optimal lag period inducing factors were selected as the model input through Cmi, and multi-scale prediction model was established based on the decomposition of multivariate sequence by MEMD. The proposed model was compared with other models (single-factor and single-scale model, multi-factor and single-scale model, single-factor and multi-scale EMD model). The results showed that the optimal lag periods of rainfall and reservoir water level in  Bazimen landslide were 2 d and 4 d. After decomposing the landslide multivariate sequence by MEMD, three groups of mode functions were obtained, each group had seven components, and the time-scale of each corresponding component was consistent. The response of the periodic term displacement to the inducing factors had a time multi-scale characteristic. Compared with the comparison model, the root mean square error of the multi-factor and multi-scale MEMD prediction model decreased by 49.4%, 36.9% and 27.4% on average, and the mean absolute percentage error decreased by 38.0%, 26.4% and 15.8% on average.

Key words: step-like landslide, displacement prediction, multi-factor analysis, multi-scale analysis, maximum information coefficient, multivariate empirical mode decomposition, Three Gorges Reservoir area