Spaceborne ICESat-2 Sub-canopy Terrain Extraction Considering Spatial Density Distribution Of Photon Point Cloud

As an important national geographic information product,Digital Elevation Model(DEM)provides important data support for infrastructure construction,disaster detection and early warning,and geographic national conditions survey.The successful launch of the new generation of spaceborne lidar ICESat-2 has provided massive data support for large-scale and high-precision sub-canopy terrain mapping.However,using ICESat-2 photon point cloud data to extract sub-canopy terrain mainly has the following three problems: 1)The original ICESat-2photon point cloud is usually doped with a large number of background noise photons,which will seriously affect the filter and sub-canopy terrain performance;2)The residual near-ground photons and photons inside the canopy after filtering will interfere with the inversion accuracy of the sub-canopy terrain;3)The obtained sub-canopy terrain has a low spatial resolution,which limits its further application on a large scale.Therefore,to carry out research on the above problems,the main research contents and innovations of this paper are as follows:(1)A photon point cloud filtering algorithm based on backward local density is proposed.The backward local density parameter is used to describe the spatial density distribution difference between the signal photon and its adjacent noise photons,and the performance of photon point cloud filtering is improved.Due to the small difference in spatial density distribution between the signal photon and its adjacent noise photons,the existing filtering methods based on local statistical parameters cannot accurately describe the difference.In view of this,this paper defines a new feature parameter,the backward local density.The backward local density can better amplify the spatial density distribution difference between the signal photon and its adjacent noise photons,and then obtain pure signal photons.At the same time,the optimal search direction can well adapt to the influence of terrain slope changes on the photon distribution.The results show that the filtering method in this paper can achieve a good filtering effect under the condition of complex terrain and high background noise,and the comprehensive evaluation index F can reach 0.9915.In addition,the filtering method in this paper has good universality and stability,which can achieve better filtering accuracy no matter in tall dense forest or low sparse forest.(2)A ground photon extraction method based on circle search and local weighted elevation is constructed,which reduces the interference of near-surface noise photons and photons inside the canopy on the identification of ground photons,and improves the accuracy of ground photon identification.The residual near-surface noise and photons inside the canopy after filtering are two key factors that restrict the accuracy of sub-canopy terrain inversion.In this paper,a circle search method is used to remove the residual near-surface noise photons after filtering.Then,the photons within a specific quantile are regarded as potential ground photons and weighted to obtain the local weighted elevation of each photon point,which can avoid the interference of the photons inside the canopy to the ground photon identification effectively.Finally,by extracting the elevation information carried by the ground photons,the inversion of the sub-canopy topography can be realized.This paper selects research areas with different topography and forest types for testing.The results show that the RMSE of the sub-canopy terrain retrieved by this method can reach 1.63 m in tall and dense forest areas,and 1.02 m in low and dense forest areas.(3)A forest height deviation correction method for SRTM DEM based on ICESat-2 ground photons is developed,which solves the problem that SRTM DEM cannot characterize the sub-canopy topography due to forest interference,and improves the accuracy of SRTM DEM.Large-scale and high-precision control points are the key factors restricting the accurate topographic correction of SRTM DEM products.In this paper,the ICESat-2 ground photons obtained after filtering and classification are used as the elevation control points,and the elevation correction surface is obtained by spatial interpolation of inverse distance weights to correct the SRTM DEM forest height deviation.Finally,the elevation correction surface and the SRTM DEM image are superimposed to achieve the correction of the SRTM DEM.In this study,single scene SRTM DEM with large terrain fluctuations and a wide forest coverage area was selected for testing.The results show that the RMSE of the SRTM DEM before and after correction is reduced from 10.74 m to 4.70 m,and the accuracy is improved by 56.2%.