The Inversion Study Of Forest Canopy Height In An Hui Province Base On Spaceborne LiDAR Data

Forests are the mainstay of carbon sinks in terrestrial ecosystems,and accurate estimation of forest canopy height is the basis for accounting for carbon stocks in terrestrial ecosystems.Therefore,mapping forest canopy height at large regional scales is essential for monitoring the dynamics of forest carbon stocks and assessing the carbon sequestration capacity of forests.Anhui province,located at the junction of subtropical and temperate zones,is a key province in China’s southern collective forest area,where forestry plays an important role in the province’s economic and social development.Nowadays,LiDAR technology has been widely used in forestry resource surveys and vegetation vertical structure parameter extraction;GEDI laser altimetry system,a new generation of space-borne full-waveform LiDAR,has been designed specifically for measuring vegetation structure,providing data support for monitoring large-scale forest canopy height.This paper aims to combine GEDI data and other multi-source remote sensing data to map the forest canopy height at 30 m spatial resolution in Anhui province,with the following three aspects:1)to verify the accuracy and analyze the error of the ground elevation and forest canopy percentile tree height values extracted from GEDI data;2)to build a topographic correction model of forest canopy height at footprint-scale based on stepwise linear regression method;3)to build an extrapolation model of forest canopy height based on random forest algorithm with cooperative active-passive remote sensing,and to generate a forest canopy height map with30 m spatial resolution in Anhui Province.The following findings were obtained:(1)This study verified the accuracy and error analysis of the ground elevation and forest canopy height by GEDI data extraction.The results show that the accuracy of GEDI-derived ground elevation is highest at bare area,with R~2 of 0.99 and RMSE of 2.83 m;Meanwhile,the CHM98 extracted from airborne LiDAR data matched the RH99 data extracted from GEDI data with the highest R~2 of 0.598 and lowest RMSE of 5.736 m;both the accuracies of GEDI-derived ground elevation and forest heights decrease with increasing terrain slope;additionally,the GEDI data collected at night performs better than the data in daytime,and the strong beams perform better than the weak beams in ground elevation and forest height retrievals.(2)This study gradually introduced multi-source characteristics variables based on the stepwise linear regression method,in order to improve the accuracy of footprint-scale forest canopy height inversion model.The results show that the RMSE value of model error is reduced by 0.347 m after introducing slope variable,and reduced by 0.288 m after introducing vegetation coverage variable.Therefore,the accuracy of forest canopy height inversion model at footprint-scale is improved by introducing multi-source feature variables.(3)This study constructs a forest canopy height extrapolation model based on the random forest algorithm,using the topography-corrected forest canopy height values as the main data source and collaborating with climate,topography and optical remote sensing data,and validates the accuracy of the model.The results showed that:the R~2 was 0.77,RMSE was3.34 m,and the model predictions were mainly in the range of 15 m～25 m;Furthermore,for the 30 m spatial resolution forest height distribution map of Anhui Province,the accuracy validation of both airborne LiDAR data(R~2 was 0.64,Bias was 1.16 m,RMSE was 5.14 m)and the measured data(R~2 was 0.69,Bias was-0.92 m,RMSE of 3.21 m)were both high.