Above-ground Biomass Estimation Of Plantations Using Airborne LiDAR And Hyperspectral Data

Forest above-ground biomass is an important indicator for assessing the forest carbon sequestration capacity,and it is essential for maintaining the stability of forest ecosystems.Meanwhile,plantation is an important part of forest resources in the southern region of China.Therefore,rapid and accurate acquisition of above-ground biomass information is of great significance for carbon sequestration measurement and ecological benefit assessment.In recent years,the wide application of remote sensing technology makes it possible to accurately and efficiently estimate forest biomass on a large scale.Airborne hyperspectral remote sensing data can obtain rich spectral information and spatial structure information of the forest canopy because of its high spatial and high spectral resolution characteristics,which in turn can estimate forest biomass.Airborne LiDAR data can acquire forest three-dimensional structural features and can provide vertical structural information related to biomass.Based on the characteristics of the two data sources,this study selected Gaofeng forest farm in Nanning City,Guangxi as the study area.Chinese fir,pine tree,eucalyptus and other broad-leaved tree were taken as the research objects.Then,the fusion method of two remote sensing data was studied using airborne LiDAR and hyperspectral data.In addition,this study established forest above-ground biomass models based on single data source and fused data source respectively to achieve high-precision estimation of forest above-ground biomass by tree species,which provides a data basis for carbon storage assessment and forest ecosystem assessment.The main research conclusions are as follows.(1)Extraction of airborne LiDAR data features to construct and optimize biomass estimation models by different tree species.Firstly,a total of 63 feature variables were extracted from the four aspects of forest canopy structure,point cloud structure,point cloud density and terrain features.Then,the important feature variables of the four tree species(chinese fir,pine tree,eucalyptus,and other broad-leaved tree)were screened separately using random forest.Finally,multiple stepwise regression,ridge regression,principal component regression and nonlinear regression were selected for the modeling methods.The screening results of feature variables showed that most of the features were point cloud structure features related to the height.The nonlinear model had the highest modeling accuracy(R~2was 0.9,0.98,0.56,and 0.79,respectively),and the verification accuracy was the lowest(R~2was 0.09,0.13,0.11,and 0.32,respectively).Among the linear models,the multiple stepwise regression model had the best modeling accuracy(R~2was 0.23,0.72,0.72,and 0.48,respectively),and the verification accuracy was good(R~2was 0.19,0.76,0.71,and 0.40,respectively).From the comprehensive consideration of model form and model accuracy,it was more appropriate to use the multiple stepwise regression method to construct the above-ground biomass model of each tree species.(2)Extraction of airborne hyperspectral data features to construct biomass estimation models by different tree species.Firstly,the wavelet transform and edge detection algorithm were used to extract the transformed spectral features and texture features,And the original spectral reflectance features,spectral derivative features,vegetation index and texture features of gray-level co-occurrence matrix were also extracted.There were a total of 11 types of feature sets above.Then,the random forest method was used for two screenings.In the initial screening,the importance features of 11 types of feature sets were selected respectively,and then the multiple stepwise regression method was used to construct the above-ground biomass model of each tree species.Finally,the feature variables with the initial modeling accuracy higher than 0.5 were fused and screened again to establish the above-ground biomass model of each tree species.The results of the two feature screening showed that the original spectral reflectance features,spectral derivative features and wavelet transformed texture features played a key role in the estimation of forest above-ground biomass.The modeling accuracy of chinese fir,pine tree,eucalyptus and other broad-leaved tree was 0.89,0.84,0.78 and 0.89 after secondary screening.And the verification accuracy had a big difference,the accuracy was 0.38,0.79,0.03 and0.13,respectively.The RMSE range was between 9.67～350.14 t/hm~2,and the error was large.(3)Fusion method of feature level and model level to construct biomass estimation models by different tree species based on airborne LiDAR and hyperspectral data.Feature level fusion used the optimal feature obtained from airborne LiDAR and hyperspectral data and then screened again to establish the above-ground biomass model of each tree species.The fusion results showed that the optimal features of chinese fir and pine tree were from the airborne LiDAR and hyperspectral data.The optimal features of eucalyptus were from airborne LiDAR data,and the optimal features of other broad-leaved tree were from airborne hyperspectral data.The modeling accuracy of chinese fir and pine tree were 0.78 and 0.95,respectively.The verification accuracy were 0.44 and 0.91,respectively.RMSE was 11.02 t/hm~2and 12.94 t/hm~2,respectively.The above-ground biomass model of each tree species based on airborne LiDAR and the above-ground biomass model of each tree species based on hyperspectral data were fused at model level.Entropy weight method,CRITIC weight method and independent weight method were used to determine the model weight respectively.The two models constructed using a single data source were weighted and fused to obtain the above-ground biomass model of each tree species.The fusion results showed that entropy weight method had the best effect on model fusion.The modeling accuracy of chinese fir,pine tree,eucalyptus and other broad-leaved tree were 0.89,0.944,0.811 and 0.88,respectively.The verification accuracy were 0.45,0.904,0.001 and 0.22,indicating that the model of eucalyptus had a serious over-fitting problem.After comparing the different above-ground biomass models of each tree species,it could be concluded that the optimal above-ground biomass model of chinese fir was based on model level fusion,the optimal model of pine tree was based on feature level fusion,the optimal model of eucalyptus was based on airborne LiDAR data,and the optimal model of other broad-leaved tree was based on hyperspectral data.The optimal model accuracy of each tree species was above 0.7.This study showed that different tree species were suitable for different fusion methods,and data fusion methods should be selected according to the features of specific tree species.In general,using two data sources to estimate the forest above-ground biomass was better for coniferous forest.The accuracy of eucalyptus above-ground biomass was the highest using airborne LiDAR data,indicating that tree height played a key role in the estimation of eucalyptus above-ground biomass.The accuracy of other broad-leaved tree was mainly related to the vertical texture features of the wavelet transform using hyperspectral data.After feature screening and model comparison,the model form of each tree species was simpler,and the estimation accuracy was greatly improved.The above research shows that the proposed above-ground biomass estimation model of each tree species has high applicability and can provide references for forest monitoring and production application.