Abstract: At present, time series model for landslide deformation prediction has been univariate model which failed to take the inducing factors of landslide displacement into account. To establish multivariate time series model is necessary. An error correction model (ECM) for landslide deformation trend prediction was established by using multivariate non-stationary time series to comprehensively analyze the landslide’s inducing factors and dynamic displacement changes. The monitoring point ZG93 of Baishuihe landslide in Three Gorges Reservoir area was taken as an example to calculate the prediction errors. Results showed that except for several points, the prediction errors are all controlled in the range of ±2.3%.

Key words: landslide, multivariate non-stationary time series, ECM, deformation prediction