Estimating Forest Volume Using Airborne Lidar In Lage-scale Subtropical Forest

The airborne Li DAR data can accurately estimate forest parameters and detect forest structure in forest survey,which not only has low-cost,time-efficient,and portrays the three-dimensional forest structure information effectively but also can monitor forest resources in a large area.In the study,a subtropical region with an area of 237,600 km~2 in southern China was used as the study area,which has species-rich and complex forest structure.It can accurately obtain information on forest structure and forest stock volume in the subtropical region and provide a basis for sustainable forest management,maintenance of biodiversity and carbon balance and others.In this study,airborne Li DAR data were collected from October 2016 to January 2020 for the whole region,and sample plot survey data were deployed accordingly.The lidar data were pre-processed to obtain the digital elevation model,and the point cloud data were normalized by the digital elevation model,and the Li DAR variables were extracted based on the normalized point cloud data.Then,forest stock volume estimation models based on airborne Li DAR data were developed for different forest types.The models were combined regularly according to the three groups of Li DAR height variables,density variables and vertical structure variables.The multiple regression power model was constructed,and there were84 models with 3-5 variables of forest type or structure.The adaptability of the models was evaluated by the hold-out method(70%training sample,30%testing sample).According to the error analysis,the optimal model of each forest type was obtained.A stratification algorithm was used to classify the forest structure based on height and density variables so that the three-dimensional structure of forest stock volume in each class was nearly the same.Estimation models of forest stock volume were constructed for fir and Masson pine forests with different vertical canopy structures of the same forest type.Then,the accuracy of the estimation model between different forest types and different vertical canopy structures of the same forest type was discussed.The following conclusions were drawn:(1)In this study,the estimation model of forest stock was constructed for different forest types based on airborne Li DAR data.It was found that the estimation model of eucalyptus forest had the highest accuracy among different forest types(r RMSE=19.14%,R~2=0.80);the model of the broadleaf forest had the lowest accuracy(r RMSE=36.08%,R~2=0.60);the model of Masson pine had higher accuracy(r RMSE=19.69%,R~2=0.82);the model of fir forest had lower accuracy(r RMSE=22.23%,R~2=0.74).The Li DAR-derived variables accurately estimate the forest stock volume of different forest types.(2)Forest structure classification based on stratification algorithm using height variable and density variable,the total number of samples for this classification is 1195,of which 127 are young forest sample plots,and the stratification sample size is 1068.The forest structure classification of the single-layer forest was 720 sample plots,and 688 sample plots were measured,with a user accuracy of 98.98%;348 sample plots of the multiple-layer forest and 380 sample plots were measured,with a user accuracy of 89.73%;the overall accuracy of the classification was 95.69%,and the kappa was 0.90,indicating that the classification results of this study were of high accuracy.(3)The stratification algorithm was used to obtain the sample data of single-layer and multi-layer forests of fir and Masson’s pine,and the forest stock volume estimation model was constructed by the above method.The results show that:the estimation model of the Chinese fir single-layer forest has higher accuracy(r RMSE=18.50%,R~2=0.71),and the multi-layer forest has a lower accuracy(r RMSE=20.53%,R~2=0.68).Compared with the optimal model without stratified samples,the error of single-layer forest(r RMSE)was reduced by 3.73%and the multi-layer forest was reduced by 1.7%.However,the estimation model accuracy of Masson’s Pine multi-layer forest was relatively high(r RMSE=17.31%,R~2=0.88),followed by the single-layer forest model accuracy(r RMSE=18.42%,R~2=0.78).Compared with the unstratified Masson’s pine forest,the error of the single-layer forest was reduced by 1.27%,and the multi-layer forest was reduced by 2.38%.The accuracy of forest stock estimation was improved after vertical canopy stratification.Forest vertical canopy structure classification estimation has a broad application prospect in forest inventory and ecological environment protection.