Integrating UAV-LiDAR And Sentinel-2 To Estimate Forest Growing Stock Volume

Growing stock volume(GSV),as one of the important indicators of forest resources inventory,reflects the abundance of forest resources and the health of forest ecosystems.Therefore,accurate estimation of growing stock volume plays an important role in forest resource management and ecosystem dynamics monitoring.Traditional growing stock volume surveys are long,intensive and labor-intensive.In particular,Unmanned Aerial Vehicle LiDAR(UAV-LiDAR)can obtain forest vertical structure information quickly and accurately at low cost;at the same time,Sentinel-2 multispectral optical image data with high spatial,spectral and temporal resolution can be obtained openly also provide new opportunities for accurate estimation of growing stock volume.Therefore,the effective combination of UAV-LiDAR and Sentinel-2 data to estimate growing stock volume and improve the accuracy of growing stock volume estimation can provide effective reference information for future sustainable forest management and management.In this study,the typical temperate coniferous plantation forest in Wangye Dian Farm,Chifeng City,Inner Mongolia Autonomous Region was used as the study area,and the two-stage extrapolation approach was used to establish the Field～UAV-LiDAR～S-2 model,the Field～UAV-LiDAR model,and the Field～S-2 model using the random forest algorithm combined with field sample plots,UAV-LiDAR data and Sentinel-2 image data.Through comparative analysis,the framework of growing stock volume estimation by the actual sample plots-UAV sampling-satellite remote sensing ensemble,i.e.Field～UAV-LiDAR～S-2 model,was established,and finally the growing stock volume of the entire study area was estimated by Field～UAV-LiDAR～S-2 model.The estimation accuracies of different growing stock volume estimation models were compared,the effects of reducing the number of sample plots within UAV-LiDAR sampling on the estimation accuracy of growing stock volume were investigated,the effects of reducing the sampling intensity of UAV-LiDAR on the estimation accuracy of growing stock volume were analyzed,and finally the spatial distribution of growing stock volume in the study area was mapped with high accuracy.The main research results include:(1)The accuracy of growing stock volume estimation based on Field～UAV-LiDAR model is the highest(R~2=0.79,RMSE=42.34m~3/ha,r RMSE=19.24%),followed by the accuracy of growing stock volume estimation based on Field～UAV-LiDAR～S-2 model(R~2=0.68,RMSE=50.19m~3/ha,r RMSE=29.43%),and the lowest accuracy of growing stock volume estimation based on Field～S-2 model(R~2=0.49,RMSE=63.25m~3/ha,r RMSE=33.08%).(2)When the number of sample plots gradually was increased(52-82),the accuracy of growing stock volume estimation based on the Field～S-2 model also gradually increased(R~2:0.49-0.64,r RMSE:33.08%-29.85%).According to the generated linear regression equation,when 107 field sample plots are reached,the r RMSE of growing stock volume estimation using the Field～S-2 model will be equal to the r RMSE obtained by estimating growing stock volume through 52 field sample plots using the Field～UAV-LiDAR～S-2 model.With the UAV-LiDAR～S-2 model,the r RMSE obtained by estimating growing stock volume through 52 field sample plots was reduced by about 23%.(3)As the sampling intensity of UAV-LiDAR decreased(100%-10%),the accuracy of growing stock volume estimation based on the Field～UAV-LiDAR～S-2 model gradually decreased(R~2=0.68-0.58;r RMSE:29.43%-51.24%).When the sampling intensity was reduced from 20%to 10%,the accuracy of growing stock volume estimation decreased sharply,and 20%of the original UAV-LiDAR sample plots could produce acceptable accuracy.(4)The results of the estimated growing stock volume based on the Field～UAV-LiDAR～S-2 model were consistent with the distribution of actual growing stock volume,which was mostly distributed in the range of 100m~3/ha-300m~3/ha throughout the entire study area.This study verified the feasibility of the Field～UAV-LiDAR～S-2 model-based estimation of growing stock volume in plantation forests and also showed the effectiveness of the two-stage extrapolation method.Therefore,the effective combination of UAV-LiDAR and Sentinel-2 data can not only improve the accuracy of growing stock volume estimation,but also provide a new idea for future large-scale forest inventory.