Studying On DEM Generated By DSM Vegetation Filtering

Digital Elevation Model(DEM)is the basic input data for many earth science researches.Among them,the DEM products obtained from satellite stereo images are the most abundant and most widely used.The penetration ability is extremely weak,so what they get is actually an elevation model containing vegetation,buildings and other non-ground data,that is,a digital surface model(Digital Surface Model,DSM).Especially in forest areas,the height of vegetation will make the elevation expressed higher than the actual ground elevation.This kind of error will seriously affect the accuracy of many terrain studies.The traditional process of generating DEM from DSM is to remove the uniform vegetation deviation irrelevant to the type of vegetation.This method may cause the processed DEM to be too high or too low in some areas,and the reliability cannot be guaranteed.With the help of vegetation height model(Canopy Height Model,CHM),filtering out a certain percentage of vegetation height from DSM is an intuitive and effective method.However,the continuous vegetation height products in large areas are not easy to obtain,so it is still difficult to use this method to perform vegetation filtering on regional or even global DSM to generate DEM.Based on the above research background,this study takes the coniferous forest and broad-leaved forest coverage area in the upper Heihe River as the research area,takes the AW3D30 DSM as the research object,and uses the DSM / DEM / CHM generated by airborne lidar as the benchmark data.The method of fusing the Spaceborne lidar(Geoscience Laser Altimetry System,GLAS)data with Landsat TM image can obtain the continuous vegetation height in a large area,and then carry out the research of DSM vegetation filter generation DEM method.The main research contents and results are as follows:(1)Vegetation height estimation method of spaceborne lidar.First,normalize the terrain of the airborne LiDAR point cloud data to generate small-scale DEM,DSM,and CHM,respectively.Then,extract the waveform parameters related to the vegetation height from the GLAS data,including the peak length,the leading edge length,the trailing edge length and so on.Finally,using the airborne LiDAR vegetation height as the reference data,a linear regression model of GLAS waveform parameters and vegetation height was established to obtain the discrete vegetation height of all GLAS laser points in the study area.(2)Regional vegetation height inversion method combining spaceborne lidar and optical remote sensing data.First,multi-scale segmentation of Landsat TM images in the study area,based on each segmented object,extract spectrum,texture,and vegetation features,divide vegetation types according to these features,and extract forest coverage areas.Then,a BP neural network is used to establish the relationship model between the image object features and the GLAS laser point vegetation height data,so as to realize spatial extrapolation and obtain a large area continuous vegetation height model.(3)DSM vegetation filtering method based on regional vegetation height model.First,taking the airborne LiDAR DEM / DSM as the reference data,the relationship between the error distribution of each pixel of the AW3D30 DSM and the airborne LiDAR DEM / DSM and the vegetation height and slope is analyzed.Then,in the airborne radar scanning area,some pixels are randomly selected to establish a multiple linear regression model between the error of AW3D30 DSM and LiDAR DEM and vegetation height and slope.Finally,using the GLAS large area continuous vegetation height model and the terrain slope calculated by the AW3D30 DSM to estimate the filtered elevation value of the AW3D30 DSM in the study area,subtract the elevation value from the AW3D30 DSM to obtain the filtered DEM.