Photon Counting LiDAR Data Processing And Forest Parameters Estimation

The spaceborne Li DAR(Light Detection And Ranging)system combines the capacity to detect the accurate vertical structure of vegetation and the ability to obtain data in a large area,which has great advantages in the quantitative inversion of forest parameters at large scales.NASA(National Aeronautics and Space Administration)has launched the ICESat(the Ice,Cloud,and land Elevation Satellite)-2 mission in September 2018,which adopts a new photon counting approach for the first time in a spaceborne platform.Therefore,relevant research to investigate data from the pre-launching time will contribute important guidance for the future forestry application.This paper used data from the airborne prototype and ICESat-2's ATLAS(Advanced Topographic Laser Altimeter System)simulation data in the United States.The main topics involved data simulation,data pre-processing and forest parameter inversion.We thoroughly discussed the influence of different settings of footprint size and sampling interval,the method to extract vegetation photons from strong background noise,and the application of forest parameters estimation from both airborne and simulated spaceborne data.The main research works are summarized as followings:(1)Data simulation from photon counting Li DAR: The FLIGHT(Forest LIGHT interaction model)radiative transfer model was used to generated the simulation datasets of ICESat-2's ATLAS sensor by using the Monte Carlo method to sample the realistic combination of various forest parameters.The related results demonstrated that: With a larger footprint size,the accuracy between the metrics of photon point cloud and the corresponding forest parameters showed a decreasing trend.For the trend in all different footprint sizes,the optimal footprint size should be set in the range of 10-20 meters.With a larger sampling distance,the relationship between the metrics of photon point cloud and the corresponding forest parameters showed a decreasing trend.When the sampling interval exceeds 0.8 m,theaccuracy indicators shows an accelerated descending trend,indicating that the optimal sampling interval should be as small as possible to achieve better detection accuracy.(2)Data pre-processing of photon counting Li DAR: We developed two kinds of methods to extract the forest photons from both unsupervised and supervised classification approach.For the unsupervised approach,we proposed an automatic algorithm by using a local outlier factor(LOF)modified with ellipse searching area,and the relevant results proved that: The horizontal ellipse searching area always gives the best result compared with the circle or vertical ellipse searching area.Furthermore,our method has a good performance not only in lower noise rate with relatively flat terrain surface but also works even for a quite high noise rate environment in relatively rough terrain such as MATLAS data.In addition,for the supervised approach,we proposed 12 statistical features of the photon counting Li DAR data and further used to classify the photons using a machine learning method.Our study highlight that only a very small number of samples are required to train the classifier and this robust model is capable to apply among a relatively large area.Further analysis proved the potential of model transferability across different sites which have similar terrain and SNR(signal to noise ratio)circumstances.(3)Forest parameter inversion of photon counting Li DAR: We first used a combination of SIMPL,G-Li HT(Goddard's Li DAR,Hyperspectral & Thermal Imager),and field measurements in Howland Research Forest,to investigate the potential of forest parameter estimation of photon counting Li DAR data.The relevant results confirm that: We found good consistency between the metrics derived from the photon counting Li DAR from SIMPL and airborne small footprint Li DAR from G-Li HT.We also found that the optimal scale size is around 50 m for the estimation of forest parameters expect the max tree height,of which the best scale size is 40 m.Further analysis of forest parameter sensitivity study and inversion models by using various link-scale sized ATLAS simulation data showed that: The metrics we proposed,which is the percentage of photons above 2 m,served as a good representative to indicate canopy closure and light interception in our simulated scenes.Furthermore,the statistical scale size which could be used as potential forest parameter inversion is 50 m.The main innovations of this paper are summarized as follows:(1)Based on the radiative transfer model,a comprehensive analysis of the sensitivity of forest parameters towards different footprint sizes and sampling intervals are thoroughly analyzed.The hardware parameters recommendation for spaceborne photon counting Li DAR is given: the optimal footprint size should be 10-20 m,the sampling interval between footprints should not exceed 0.8 m.This result could provide theoretical support for the hardware design of sensors.(2)An automatic forest signal detection algorithm by using a local outlier factor(LOF)modified with ellipse searching area was proposed,and the recommended ellipse parameter configuration is given: the ratio of the major and minor axes is 6:1.The average detection accuracy of the algorithm is not less than 85%,and can be implemented to a high noise rate environment in relatively rough terrain.