| Forests,as one of the largest ecosystems on Earth,play an important role in global climate regulation,biodiversity,and water resource management.However,obtaining actual forest data across large regions requires considerable human and material resources and can only provide limited data at a sample level.Therefore,the use of remote sensing technology to observe forest parameters has become increasingly important.Among them,spaceborne LiDAR,as a high-precision and high-resolution remote sensing technology,can obtain three-dimensional information about objects and observe large forest areas,achieving precise inversion of forest canopy height.In this study,using ICESat-2/TLAS data mainly from Genhe City,Hulunbuir,Inner Mongolia,noise reduction and classification were performed,and forest canopy height inversion was performed.At the same time,this study analyzed the influence of terrain slope and vegetation cover on noise reduction and classification accuracy,and discussed the impact of sampling window size and canopy height percentile on forest canopy height inversion.This study provides new ideas and methods for noise reduction and classification of ICESat-2/ATLAS photon point clouds,and improves the accuracy of forest canopy height inversion.The specific research content and results of this article are as follows:(1)An adaptive photonic point cloud denoising algorithm is proposed.ICESat-2/ATLAS data may produce bias in the positioning of signal photons in the presence of large terrain undulations.Therefore,in this study,the search shape was improved to an ellipse,and the size of the search ellipse was adaptively calculated based on photon characteristics.The terrain slope was also used as an input parameter to correct photon coordinates and improve the accuracy of the denoising algorithm.The results showed that the algorithm performed well in all nine datasets in the study area,with an average F-value of 0.98 for denoising results.(2)Analyzed the effect of terrain slope on the denoising algorithm.As the slope increases,the accuracy of the denoising algorithm gradually decreases.In this study,the F-value range of the denoising results is0.970-0.984,while the F-value range of ATL08 denoising results is0.943-0.973.Therefore,compared with ATL08 denoising algorithm,the algorithm proposed in this study can effectively reduce the impact of terrain slope.(3)A multi-level photon classification algorithm was proposed.A photon density-based approach to ground photon and canopy top photon extraction can effectively reduce the influence of noise photons near the canopy surface.Based on the extracted ground photons and canopy top photons,the terrain surface and canopy surface were fitted and compared with the reference terrain surface and canopy surface data,and the R~2 and RMSE of the terrain surface fit were 0.999 and 1.086 m,respectively,and the R~2 and RMSE of the canopy surface fit were 0.990 and 3.603 m,respectively.(4)The study discussed the influence of terrain slope and vegetation coverage on the classification algorithm.The results showed that terrain slope had a greater impact on the extraction of ground photons than vegetation coverage,while vegetation coverage had a greater impact on the extraction of canopy top photons than terrain slope.Higher accuracy classification results can be obtained under smaller terrain slope and vegetation coverage.(5)The influence of different window sizes and canopy height percentages on the accuracy of forest canopy height retrieval was studied.The results showed that the optimal window size for ATL08 algorithm was 50 m and the optimal canopy height percentile was RH95,which corresponded to a retrieval accuracy of RMSE=3.440 m and MAE=2.595 m.The optimal window size for the proposed algorithm was 35 m and the optimal canopy height percentile was RH100,which corresponded to a retrieval accuracy of RMSE=3.039 m and MAE=2.313 m. |
PDF Full Text Request