Abstract:

Based on the annual monitoring and evaluation data of forest resources in Guangxi Zhuang Autonomous Region,the characteristic variables of airborne LiDAR were selected with the stepwise regression,and multiple linear regression,logistic regression and random forest models were established to predict the volume of sample plots of Eucalyptus robusta,Cunninghamia lanceolate and natural broad-leaved forest in southern collective forest region.The results showed that:1) For the plots of Eucalyptus robusta and Cunninghamia lanceolate,the stepwise regression characteristic variables were mainly height and intensity variables,while for the natural broad-leaved forest plots,they were comprehensive variables,such as clearance ratio,coverage,and leaf area index;2) For Eucalyptus robusta and natural broad-leaved forest plots,the estimation accuracy( Eucalyptus robusta:R ²=0.97,RMSE=12.60 m³/hm²;natural broad-leaved:R²=0.90,RMSE=18.45 m³/hm²) of the volume of the random forest model was higher than that of the multiple linear regression and logistic regression models,for the Cunninghamia lanceolate plot,the multiple linear regression model obtained the best estimation result(R ²=0.91,RMSE=24.30 m³/hm²);3) Among the estimation accuracy of the three models,the artificial Eucalyptus robusta and Cunninghamia lanceolate plots were higher than of the natural broad-leaved forest plots.It can be seen that the high-density LiDAR point cloud can obtain better characteristic variables.In view of the complex sample plot conditions,we need to flexibly select the estimation model to realize the volume survey,which can provide a scientific basis for the forestry and grassland authorities to carry out the forest resources survey,monitoring and management.

Key words: southern collective forest region, UAV LiDAR, regression model, random forest, stock volume