Study On Inversion Of Forest Parameters Based On Airborne LiDAR

With the rapid development of remote sensing technology,airborne laser radar has been widely used to estimate forest parameters as an active remote sensing technology.Taking the Inner Mongolia Eagan area as the study area,the study combined field measurement data and airborne LiDAR data to estimate individual tree and forest parameters,the main methods and results are as follows:(1)The four polynomial fitting method was proposed to estimate individual tree crown diameter.In this study,the variable window was applied to search the top of tree,and individual tree height was extracted by the local maximum method.We extracted elevation data on canopy height model at the north-south and the east-west directions centering on the top of the individual tree.Two fourth degree polynomial fitting curves were extracted on the two group data respectively.The horizontal distance between two pits on the curve was considered as crown diameter at this direction.The crown diameter was calculated as the average of two values measured along two directions.The results showed that the shape of variable window had influence on success rate of searching the top of tree.Circular window(recognition rate is 85.2%)is better than the square window(recognition rate is 82.1%).The mean absolute error of individual tree height and crown diameter are.2.33m and 0.99m,respectively.(2)The study proposed a method to compute the penetration index in order to estimate LAI.All of collected LiDAR echoes were grouped into many categories depending on the laser emissions they were generated from,and then the single laser penetration index(LPIs)was obtained by the echo intensity and LPImean was obtained by using LPIS accordingly.To compare the results of LAI estimation by using LPImean with that by using ungrouped LPI,the theoretical models and the empirical models were established respectively in four LiDAR sampling scales(circle with diameter 5m,10m,15m and 20m).The results showed that the LAI estimation from LPImean was obviously better than ungrouped LPI.Both of the empirical model(R2=0.80,MAD=0.11)and the theoretical model(R2=0.77,MAD=0.16)got the best LAI estimation accuracies when LiDAR sample scale was at 15m.The LAI distribution of Birch forests in the study area was produced finally by combination of the empirical model and the theoretical model with the best accuracies.(3)The above ground biomass model was established by BP neural network.The net input layers were maximum estimated tree height,minimum estimated tree height,average estimated tree height,the standard deviation of tree height,maximum estimated crown diameter,minimum estimated crown diameter,average estimated crown diameter,the standard deviation of estimated crown diameter and LPImean.The output layer was the field above ground biomass.There were 1 hidden layer and 19 hidden notes.When the number of iteration reached 6,the network performance is best.The results showed that RMSE is 21.12kg.The AGB map of birch in the study area was produced by network model.