Abstract: In the subsidence monitoring of coal mining areas,the conventional geodesy,InSAR and other remote sens- ing methods have some limitations. LiDAR technology can be used to scan ground subsidence area,and the fine fea- tures of surface subsidence basin can be quickly acquired by the superposition of multi-stage point clouds. However, the subsidence model based on the existing mainstream point cloud filtering and interpolation algorithms often contains significant noise,which limits the practical application of this technology in mining areas. The surface of a mining working face in Yushen mining area in China was taken as the experimental area. In view of its topographic fluctuation and low vegetation coverage geographical environment,the UAV LiDAR was used to obtain four sets of ground point cloud data in two phases. The improved method of mining subsidence modeling based on laser point cloud was studied by combining with the measured data of conventional surface movement. Through the comparison and analysis of DEM errors corresponding to commonly used point cloud interpolation and filtering algorithms respectively,including author- ized digital elevation model,inverse distance weighting,kriging,natural neighborhood,spline interpolation and progres- sive triangulated irregular network densification,multi-scale curvature classification,maximum local slope,and progres- sive morphological filtering,the results of using authorized digital elevation model interpolation and progressive triangu- lated irregular network densification fil-tering were better,but the generated initial subsidence model was still not accu- rate,which mainly included the model error caused by point cloud plane position error,non-ground point noise,point cloud interpolation error and water coverage range change. Based on the analysis of the above error distribution charac- teristics and the im-provement approach,an optimization scheme of the subsidence model denoising based on the wave- let threshold was proposed. Aimed at the different wavelet parameters of the subsidence basin and the non-subsidence region,the whole region were decomposed by using prior condition that the subsidence value of the non-subsidence re- gion was zero,the subsidence region were decomposed by the low-level wavelet,and then the two results were pro- cessed by embedding. The results showed that the accuracy of the subsidence model was significantly improved after the wavelet denoising,and the detailed characteristics of the subsidence basin were effectively retained. The overall standard deviation of the subsidence model was within 50 mm,which can meet the basic requirements of the large sur- face deformation monitoring in the western mining area. According to the random error characteristics of the subsidence model edge,a subsidence boundary extraction method based on subsidence slope was proposed,which could provide a feasible solution for the airborne LiDAR technology for the efficient monitoring and fine modeling of mining subsidence in the western mining area.