Forest Height Retrieval Of China With A Resolution Of 30m Using ICESat-2 And GEDI Data

Global climate change is one of the most serious challenges,which is related to the survival and development of human beings.Therefore,it has become a hot issue that governments are increasingly concerned about.As the main component of the terrestrial ecosystem,the forest ecosystem plays an irreplaceable role in mitigating global climate change and regulating global carbon balance.Accurate estimation of forest height over large areas is helpful to reduce the uncertainty of carbon sink,and is of great significance to the study of the terrestrial carbon cycle and global change.As an active remote sensing technology,LiDAR can quickly and accurately obtain the vertical structure information of forest canopy,thus it has been regarded as the most powerful tool for forest height retrieval.Space-borne LiDAR has been widely used for forest height retrieval and mapping due to the global coverage.In fact,the first generation of space-borne LiDAR ICESat-1/GLAS has been successfully used to map forest height at a global/regional scale.However,due to the low density of GLAS footprints,the spatial resolution of forest height maps is generally low(500 m or 1 km).Compared to ICESat-1/GLAS,the new generation of space-borne LiDAR ICESat-2/ATLAS and GEDI have smaller footprint sizes and higher sampling densities,which makes them possible for forest height mapping with high spatial resolution over large areas.This study aims to map forest height with 30 m resolution in China utilizing the new generation of space-borne LiDAR data and other remote sensing data.There are four specific objectives: 1)to study the key technologies of photon-counting LiDAR data processing,including noise photons removal and signal photons classification;2)to build the footprint-scale forest height inversion models for GEDI and ICESat-2/ATLAS data respectively;3)to study the forest height consistency of different types of space-borne LiDAR data,and 4)to construct the forest height extrapolation model based on both active and passive remote sensing data for mapping the forest height with a resolution of 30 m in China.The main contents and findings are as follows:(1)This study proposed a noise removal algorithm based on improved OPTICS for photon-counting LiDAR data.Firstly,the elevation statistical histogram was built to remove obvious noise photons.Secondly,the improved OPTICS algorithm was used to separate signal photons from noise photons.Finally,the remaining noise photons are removed based on the elevation distribution characteristics of photons.The results indicated that the new proposed algorithm is insensitive to the input parameters and has high self-adaptability.It performs much better than the noise removal algorithm of ATL08 in extracting the signal photons over complex terrains.Additionally,the F value is increased by 5% on average compared with the improved DBSCAN algorithm.(2)This study proposed a multi-level progressive signal photons classification algorithm.This algorithm can reduce the influence of canopy photons and noise photons by conducting the steps of initial ground photons extraction and pseudo-ground photons removal,and thus improve the extraction accuracy of ground photons in dense forests with complex terrains.Compared with the classification algorithm of ATL08,the proposed signal photon classification algorithm is more suitable for areas with dense vegetation cover or sparse ground photons.The overall accuracy is improved by 8%.(3)This study built the footprint-scale forest height inversion models for GEDI and ICESat-2/ATLAS data and verified the accuracy of these models in different scenarios.The correlation between space-borne LiDAR-derived forest height percentiles and airborne LiDAR-derived forest height was explored.To verify the accuracy of forest heights extracted from space-borne LiDAR data,this study adopted a large amount of public airborne LiDAR data in the United States.For GEDI and ICESat-2/ATLAS data,the forest height percentiles with the highest correlation with airborne LiDAR data are RH95 and RH100,respectively.The accuracy of GEDIderived forest heights is higher than that of ICESat-2/ATLAS.For different scenarios of ICESat-2/ATLAS data,the accuracy of forest heights collected in the night time is generally higher than that in the daytime,and the strong beams perform much better than weak beams in extracting the forest heights.(4)This study established the forest height consistency models for two different types of space-borne LiDAR data.The forest heights extracted by GEDI data were regarded as the reference values,and the relationship between the GEDI-derived forest heights and feature parameters of ICESat-2/ATLAS data was built by machine learning and stepwise linear regression algorithm.The portability of the consistence model was explored between different forest types and study areas.The results indicated that the difference of forest heights extracted by different types of space-borne LiDAR can be greatly reduced and the accuracy of forest heights is generally consistent by establishing the forest height consistency model.For different study areas and forest types,a universal forest height consistency model can be established.(5)This study mapped the forest height with a resolution of 30 m in China.The space-borne LiDAR data was first taken as the main data source and combined the 30 feature parameters extracted from Sentinel 1,Sentinel 2,Landsat 8 and other data.Then,the forest height extrapolation models corresponding to different ecological regions and forest types were constructed by machine learning algorithm,the forest height map was finally produced based on these forest height extrapolation models.The mean error is0.08 m and the RMSE value is 2.67 m,and the spatial resolution is improved from 500 m to 30 m.