Study On Forest Structure Parameters Inversion Based On GEDI Data

Traditional forest monitoring methods are limited and cost a lot of manpower and financial resources.Satellite remote sensing has the characteristics of being not limited by national boundaries,long distance and large range of earth observation,and has unique advantages in acquiring geographic information.As a new type of remote sensing payload,lidar has the advantages of high accuracy,high resolution and active observation,and can obtain forest vertical structure information,so it has been widely concerned and applied in forest monitoring and protection.The Global Ecosystem Dynamics Investigation(GEDI)is the first multi-beam full waveform lidar for forestry,it is mainly used for accurate measurement of forest canopy height,understory topography and aboveground biomass in tropical and temperate regions between 51.6°N and 51.6°S.In order to explore the accuracy of GEDI inversion of forest structure parameters,the study took Penobscot experimental forest in Maine,USA as the research area,and GEDI L2A and L4A data products as the research data,and processed and analyzed the two types of data products respectively.L2A data products perform Gaussian smoothing on L1B waveform data,set background thresholds,and input the extracted results to L2A.For L2A,10 parameters were extracted for spot location,data screening,and canopy height inversion model was established.L4A aboveground biomass was obtained by quantifying relative height(RH)index in L2A.For L4A data product,11 parameters were selected to establish biomass inversion model.Using NASA Goddard’s Lidar,Hyperspectral and Thermal Airborne Imager(G-LIHT)as validation data,the accuracy of forest canopy height and biomass inversion by GEDI was evaluated.The main research contents and results are as follows:(1)Based on the GEDI L2A elevation and height data,extract the parameters of the forest canopy height and understory topography,first locate the position of each footprint according to the shot＿number,lon＿lowestmode and lat＿lowestmode indicators,and then use the quality＿flag,degrade＿flag and sensitivity to filter the high quality footprint data,and finally establish the inversion model of forest canopy height and understory topography according to elev＿lowestmode,elev＿highestreturn and mean＿sea＿surface.The accuracy of forest canopy height retrieval from GEDI data is discussed from three perspectives.According to different scenarios,in order to meet the situation that the canopy height is too large or the ground elevation is too low,different noise,signal,start and end threshold combinations are set to finally generate 6 sets of different L2A elevation data.The highest inversion accuracy is for the a2 algorithm group,R~2=0.87,RMSE=1.71m,MAE=0.84m,and it is consistent with the results of inversion of understory topography.The inversion accuracy of forest canopy height is different for different time periods and different beam types,and more effective footprint data are obtained at night and with full power beam types,but the results show that the inversion accuracy is not necessarily higher.(2)Inversion of above-ground forest biomass data based on GEDI L4A footprint-level biomass data.For the L4A data,the relative height index in L2A was used as the input of the parametric linear model to predict the above-ground biomass of the forest,and the above-ground biomass prediction value was generated for each relative height.For the study of L4A data,geolocation is also carried out according to shot＿number,lon＿lowestmode and lat＿lowestmode,and then the five indicators of l2＿quality＿flag,l4＿quality＿flag,algorithm＿run＿flag,degrade＿flag,sensitivity are used to screen L4A data to determine high-quality footprint points,and finally according to agbd,agbd＿pi＿lower,agbd＿pi＿upper to establish the inversion model and verify the accuracy.In view of the defects of small number of a5 algorithm groups and low precision,L4A additionally set up a10 group instead,which improves the accuracy.Compared with 47 in a5 group,the number of samples in a10 group has increased by 44 footprint points,RMSE and MAE increased by 5.50Mg/ha and 5.10Mg/ha,respectively.Among the 7 sets of data,the a2 algorithm group has the highest accuracy,with R~2=0.92,RMSE=10.80Mg/ha,and MAE=4.99 Mg/ha.