Abstract: The accurate prediction of landslide displacement is mainly affected by the prediction model and parameters. Traditional landslide prediction models, such as regression model and grey prediction model, have the shortcomings of single factor and low accuracy of long-term prediction. However, gradient enhanced multiple regression model can realize long-term prediction of landslide surface displacement on the basis of considering multiple factors, which can effectively make up for the above shortcomings. At last, it will be used in nanshan landslide in xining city as an example, based on the various influence factors of landslide, merit-based landslide induced key factors (precipitation, geological structure, topography). And a variety of methods are used for comparative analysis, including gradient boosting regression, Bayesian ridge, elastic net and support vector regression of four model. Experimental results show that the EV, MSE and MAE of GBR are 0.99 mm², 0.04 mm and 0.15 mm, respectively. In addition, GBR is used to predict the surface displacement in December 2020, and the relative error is found in the range from-0.8% to 0.8%. The prediction accuracy is the highest. Therefore, the gradient enhanced multiple regression prediction model is relatively more accurate and efficient. It can accurately reflect the change state of landslide surface displacement, accurately monitor and warn the landslide body all day long, and ensure the safety of the surrounding environment of the landslide body.

Key words: landslide, BeiDou monitoring, prediction, gradient boosting regression