The Estimation Of Aboveground Biomass Of Larix Olgensis Plantation Based On Bayesian Approach And Multi-source Remote Sensing Data

Forest aboveground Biomass(AGB)is one of the most basic quantitative representations of forests.It reflects the energy flow and biomass cycle between trees and the environment and is an important indicator for the assessment of forest carbon sources and sinks.Therefore,accurate and efficient estimation of forest AGB is very important for addressing global climate change and regulating the global carbon balance.With the emergence and maturity of remote sensing technology,the application model of combining field survey data with multi-source remote sensing data has become the main method to estimate forest AGB at the scale of individual tree,sample plot(pixel)and forest.However,the uncertainty of parameter estimation of existing forest AGB parameter estimation models shows that the Bayesian approach and hierarchical Bayesian approach under the framework of Bayesian statistics provide an alternative to the traditional frequency statistical method which represents model parameters with fixed values,and have better applicability in the application of hierarchical spatial data structure in forestry.In addition,at the current stage,the forest AGB estimation based on remote sensing means still needs to take the verification data obtained by sampling technology as the support of statistical inference or model test of the research.Spatial sampling method combines classical sampling technology and spatial statistics theory,which can effectively optimize the sample redundancy and low sampling efficiency of classical sampling method.The advantages of integrated spatial sampling method and Bayesian statistical method in small sample estimation also provide a new idea for scientific selection of validation data to improve the efficiency and accuracy of forest-scale AGB estimation by remote sensing means.In conclusion,it will provide reliable theoretical and technical support for efficiently and non-destructively monitoring of forest biomass to verify and explore the effects and advantages of multi-source remote sensing data combined with Bayesian statistical method in forest AGB estimation at various scales.In this study,the larch(Larix olgensis Henry)plantation in Maoershan Experimental Forest Farm was taken as the research object.The near-surface LiDAR data,optical data and the measured data of the plot were combined with the Bayesian approach and hierarchical Bayesian approach under Bayesian statistical theory,the individual tree AGB estimation models were constructed based on individual tree segmentation under different sample sizes.The Individual Tree-based Approach(ITA)and Area based Approach(ABA)were used to compare and explore the effects of AGB estimation at sampled location(pixel)scales.Based on the acquisition of small range and high precision AGB product,the actual sampling accuracy and sampling cost of three spatial sampling methods and three classical sampling methods were further compared.Based on the sampling points obtained by different sampling methods and the multi-source remote sensing data covering the forest farm,a forest-scale AGB inversion model was constructed by hierarchical Bayesian approach.In this study,the applicability and advantages of Bayesian statistical method in forest AGB estimation based on remote sensing were systematically and comprehensively explored at various scales for larch plantation,the main afforestation tree in Northeast China,and the effects of scientific selection and spatial distribution of validation data on the efficiency and accuracy of forest scale AGB estimation were also explored.It provides a theoretical basis for developing a more scientific and reasonable integrated forest biomass estimation scheme of"sky(satellite remote sensing)+air(LiDAR)+ground(field investigation)".The main conclusions of the study are as follows:(1)In the individual tree AGB estimation,high-precision individual tree parameters(individual tree diameter at breast height(DBH),tree height(TH)and canopy projection area(CPA))can be obtained without damage based on multi-platform fusion LiDAR data(R~2>0.9);Compared with the traditional frequency statistics method,the Bayesian statistical method had smaller standard error of parameters,narrower 95%confidence interval,and more stable estimation of model parameters for the estimation of individual tree AGB.In addition,the results of Bayesian approach for estimating the individual tree AGB under large sample size were similar to the traditional frequency statistical methods,but with the decrease of sample size,the RMSE values of Bayesian approach were reduced compared with the traditional frequency statistics method,and the model fitting effects were better.(2)In the individual tree-based approach(ITA)to estimate the AGB at plot scale,the Bayesian approach and hierarchical Bayesian approach were compared for the estimation effect of individual tree DBH model and individual tree AGB model.It was found that hierarchical Bayesian approach has improved the model estimation effect compared with Bayesian approach.In the area-based approach(ABA)to estimate the AGB at plot scale,we compared the estimation effect of the traditional frequency statistics and Bayesian statistics method.It was found that the hierarchical Bayesian approach under the Bayesian statistical can obtain the optimal sample scale AGB results(RMSE=18.296 t/ha,r RMSE=12.5%),and the standard error of model parameters were smaller,and the parameter estimation was more stable.(3)By comparing the plot scale AGB results under ITA and ABA methods,the optimal plot scale AGB inversion model based on UAV-LiDAR data was determined,and the pixel scale high-precision forest AGB product(U＿AGB)retrieved within the UAV flight range was obtained.Moreover,the actual sampling accuracy and sampling cost of the classical sampling methods and the spatial sampling methods were compared,it was found that the spatial sampling methods can save nearly half of the sampling cost compared with the classical sampling methods under the condition of 85%-90%sampling accuracy.In addition,based on the multi-source remote sensing data and the advantage of hierarchical Bayesian approach in small sample estimation,it was found that the forest scale AGB model constructed based on the sampling points laid out by the spatial stratified sampling method had the best fitting and accuracy.Finally,the average AGB of larch plantation in Maoershan Experimental Forest Farm was 121.53 t/ha.