Forest Canopy Height Inversion Using Spaceborne Full Waveform LiDER

In the context of climate change,accurately estimating forest’s carbon storage has become an important step in the United Nations’plan to reduce emissions caused by deforestation and forest degradation.Forest canopy height,as an important component of forest vertical structure parameters,is a key parameter for estimating carbon storage,and its quantitative esti-mation has important theoretical significance and practical application value.Lidar technology has been proven to be a powerful tool for estimating forest canopy height and other vegetation parameters,which are important information for ecosystem surveys.The GEDI system is the latest full waveform spaceborne Lidar equipment specifically designed for observing global forests and has been reliably applied to the inversion of forest pa-rameters.This study focuses on the forest area of Genhe City,Inner Mon-golia Autonomous Region in China,using GEDI L1B waveform data and canopy height reference data within the study area for regression analysis.The study builds an inversion model for forest canopy height and combines it with optical remote sensing image data to extrapolate the results to obtain continuous distribution of forest canopy height data.The research results provide new ideas and methods for the inversion of forest canopy height using spaceborne full waveform GEDI data.The main research content and conclusions are as follows:(1)The canopy height model(CHM)obtained from airborne LiDAR point cloud data was used to validate the accuracy of the GEDI L2A forest canopy height variables in the study area.The results showed that the rh95height variable had the highest correlation with the forest canopy height extracted from the CHM,with R~2=0.466,RMSE=6.692m.However,it was found that the accuracy of this study was far lower than the estimated accuracy of the GEDI L2A forest canopy height in the US.(2)Based on convolutional neural networks(CNN),a forest canopy height inversion model was built for the study area at the patch scale using GEDI L1B waveform data as input variables and the CHM obtained from lidar as the canopy height reference data.The results show that the built CNN inversion model can reliably estimate the forest canopy height,with R~2=0.606,RMSE=3.185m.Compared with the GEDI L2A canopy height product data,R~2 increased by 23.1%and RMSE decreased by 52.4%.(3)The required bands were extracted from optical remote sensing Sentinel-2 imagery and the relevant vegetation indices were calculated.Combined with the forest canopy height data predicted by the CNN model,a spatially continuous forest canopy height inversion model was con-structed using the random forest(RF)regression algorithm,which achieved forest canopy height extrapolation in space.The predicted results were validated using airborne lidar data,with an accuracy of R~2=0.635,RMSE=3.013m.