Inversion Of Forest Aboveground Biomass Using Airborne LiDAR And Multispectral Remote Sensing Data

Forest biomass is an important factor in the study of global environment and climate change,and plays an important role in the study of carbon cycle.Sustained deforestation and forest degradation have led to a sharp decrease in forest biomass and carbon storage,which has seriously affected the productivity of forest ecosystems.It is of great significance to estimate regional forest biomass quickly,quantitatively and accurately.It can not only evaluate forest ecosystem reasonably,but also provide reference factors for carbon storage research.In this paper,we use airborne LiDAR data and multispectral Landsat-5 TM data to invert large-scale forest biomass.Firstly,based on LiDAR data,the vertical structure of forest was restored,and a high-precision biomass estimation model was established.Then,using LiDAR estimation results as validation data,the spatial expansion of forest biomass was carried out using Landsat-5 TM image,and the feasibility of inversion of forest biomass by a small amount of LiDAR data and large-scale Landsat-5 TM image was explored.The main conclusions are as follows:(1)The superior penetration of LiDAR is one of the best remote sensing techniques to retrieve forest biomass.In the case of good point cloud density in the study area,a high-precision biomass estimation model can be directly established by extracting the vertical structure characteristics of forest.In the absence of point cloud penetration,the accuracy of biomass estimation can be improved by the strategy of post-event feature restoration.Experiments show that the results of biomass estimation can be improved by combining the vertical structure characteristics of the restored forest with different empirical models.(2)When estimating forest biomass,the selection of regression model is very important.The experimental results show that partial least squares regression is superior to multiple stepwise regression and support vector machine model in the case of only small samples.In the case of LiDAR estimation results as validation data,the CNN feature extraction ability of convolutional neural network is stronger,and the results are better than the empirical model.(3)When estimating ascending biomass based on Landsat-5 TM data,a multispectral inversion model was established with a small amount of LiDAR data as validation data,and the accuracy of the model estimation could reach 0.605.This shows that combining a small amount of LiDAR data with a large range of multi-spectral data can be used for regional forest biomass survey,and at the same time can reduce the cost relatively.Through multi-source data processing,vertical structure feature restoration and regression model selection,this study explored a method for estimating forest biomass based on a small amount of LiDAR and multi-spectral data.Finally,the regional biomass distribution map was obtained,and the spatial distribution characteristics of biomass were analyzed.