Forest Biomass Estimation In Alpine Mountain Based On GEDI

Accurate estimation of forest above ground biomass is essential for regional and global carbon cycling and climate change mitigation.Mountain forests are difficult to conduct human surveys because of the complex terrain.There are large uncertainties in the estimation of both forest structural parameters and biomass.In order to obtain continuous biomass mapping in large scale mountain forest inversion,it is necessary to extrapolate the biomass of sample plots using suitable models with the help of active or passive remote sensing data.The satellite-based LiDAR GEDI,as a large spot active remote sensing technology that can extract 3D structural information of forests,can improve the accuracy of large scale forest biomass estimation.In this paper,we estimated forest biomass in the study area using ground survey data,UAV-LiDAR data,GEDI data,Sentinel-1 and Landsat 8 OLI image data,and evaluated the accuracy by extrapolating the geographically weighted regression and random forest models,respectively,using Qilian Mountains National Park as the study area.The main findings of this study are as follows:(1)At the sample scale,forest structure parameters and point cloud feature parameters extracted from UAV-LiDAR point clouds can be used as a proxy for ground measurement data to accurately predict forest biomass(R~2=0.84,RMSE=44.95Mg/ha,r RMSE=26.97%).GEDI data are effective for improving the accuracy of forest biomass prediction at large scales(R~2=0.75>0.69,RMSE=28.75<29.44Mg/ha).At the regional scale,the extrapolated forest biomass accuracy was(R2=0.66,RMSE=19.08 Mg/ha,r RMSE=11.04%),and the total forest biomass in the study area was 3.07×10~7Mg.(2)In estimating forest biomass using the random forest model,an upscaling approach is used to scale up the biomass from the sample plot scale to the regional scale.The parameters in the UAV-LiDAR point cloud were first extracted and used in the random forest model to estimate the forest biomass at the sample plot scale;the forest biomass estimation results of UAV-LiDAR were used as the input of the random forest model in the next step,and the metrics of GEDI data were used to estimate forest biomass at its footprint points was obtained;the forest biomass at discrete GEDI footprint points was used as input to the random forest model at extrapolation,and the continuous forest biomass data of the study area was estimated using multiple characteristic variables from Sentinel-1 and Landsat 8 OLI images.(3)In this study,based on GEDI L2A height products,topographic data,meteorological data and other auxiliary data,forest tree height in the study area was estimated using a geographically weighted regression model(R~2=0.43,RMSE=6.39m).Based on the forest tree height,the forest biomass values were obtained by establishing the allometric equations for different forest types.It was found that the forest biomass obtained using the geographically weighted regression model and the allometric equations were smaller than the random forest model(0.78×10~7<3.07×10~7Mg),which was due to the cumulative error caused by the fact that the anisotropic growth equation only focused on the maximum height of a single tree at the GEDI footprint point and ignored the cumulative error in estimating the biomass of a single tree.