Dynamic Prediction of Landslide Displacement Using Time Series GRU and Incorporating Environmental Variables

Landslide is one of the most common geological disasters globally, which has caused serious impact on human society and natural environment.High-precision prediction of landslide displacement has important effects on prevention and early warning against landslide disasters.Most existing landslide displacement prediction models focus on static model methods and focus minimally on the effects on the external environmental variables of landslide.To solve these problems, this study proposed a time series gate recurrent unit (GRU) dynamic prediction model that considers the effects of environmental variables.First, landslide displacement was decomposed into the trend and periodic term displacements by exponential smoothing.Second, a GRU model was built by considering the influences of external environmental variables on landslide displacement to predict the periodic term displacement.Third, the individual component displacements were aggregated to attain a dynamic forecast of the landslide movement.Lastly, a case study was conducted based on the landslide of Baishui River, China.The effectiveness of the proposed prediction model was verified by comparing with traditional intelligence algorithms (e.g., back propagation (BP) and extreme learning machine (ELM)).Results demonstrate that the proposed model conforms well to the evolution process of landslide displacement with consideration to the influences of external environmental variables on the fluctuation characteristics of periodic term displacement.The memory structural function of the GRU model can automatically adapt to the dynamic variation characteristics of landslide data during landslide prediction.The minimum and maximum prediction errors of the GRU model are 0.01 mm and 12 mm, respectively.The GRU model, compared with the BP and ELM static models, effectively increases the prediction accuracy (RMSE is increased by 6.7 times and the MRE is increased by 3.5 and 7.6 times, respectively).This study provides an important evidence for the prediction, early warning, prevention, and reduction of landslide disasters.