Research On The Methods Of Forest Stock Volume Inventory And Estimation

Forest stock volume（FSV）is an important indicator of the National Forest Inventory,which is a direct reflection of the total amount of forest timber in the country and region.It is directly related to forest biomass and carbon storage,and directly affects the national or regional timber strategic safety and the ability of fixed carbon dioxide and animal habitation.Estimation of FSV timely and accurately is one of the most important tasks for forestry workers.This study focused on FSV estimation from ground survey plot methods to remote sensing data at multiple levels,forming a more scientific road from point to surface.The main research results were summarized as follows:（1）Designed a new FSV ground survey sample plot marking device,which solved the guiding problem of sample plot tracking surveys,the problem of the recording of plot information,and the problem of movement and destruction of the plot marking device.Compared with the traditional sample plot marking device,the search time efficiency of the newly designed device was improved by 324.18%.The innovative FSV marking device could not only ensure the quick search and surveys,but also laid the foundation for the continuous investigation and monitoring of the FSV.（2）A new FSV sample plot positioning method was proposed.By using three base stations for static relative positioning test,it could be concluded that when the erection distance between the base stations within 30 km,and the common working time of the three base stations not less than 30 minutes,the positioning error would be in 0.50 m.Based on this high-precision sample plot positioning method,the FSV obtained from the ground sample plot survey could be more accurate to match with the remote sensing pixels,which laid the foundation for the further improvement of the FSV estimation accuracy.（3）Based on the method of quantitative estimation of the saturation point of remote sensing data using the spherical model mentioned in previous studies,this research proposed a new simple binomial model to calculate the FSV estimation saturation point,and calculated the maximum saturation point of Landsat-8 L8＿B3 and Sentinel-2 S2＿B5 variables were 323.08 m³ ha^-1 and 327.46 m³ ha^-1 respectively.（4）Based on high-precision positioning FSV sample plot data and multi-source remote sensing data,combined with variable selection method and machine learning algorithms,the FSV was modeled and tested.By comparing the results before and after the variable selecting,it was found that the variable selection method had almost no impact on the random forests（RF）algorithm,but it had an unstable effect on the support vector regression（SVR）algorithm.Comparing the performance of RF and SVR,it was found that the RF algorithm performed better than the SVR algorithm in the training and test phases.Based on variable selection and RF algorithm to compare the performance of Landsat-8 data,Sentinel-2data and the combined data of Landsat-8 and Sentinel-2 in the test phase,the combined data performed best in the estimation of FSV,with R²=0.51,MAE=52.45 m³ ha^-1,RMSE=67.56 m³ ha^-1.Combining Landsat-8,Sentinel-2 and mean forest height variables,under the variable selection and RF algorithm,the performance in the test phase was R²=0.71,MAE=35.10 m³ ha^-1,RMSE=51.57 m³ ha^-1.This showed that the mean forest height was an important variable for FSV estimation.Based on this,this study tried to analyze the large-footprint Li DAR data to extract the mean forest height,and the accuracy of the estimated results was R²=0.91,MAE=0.77 m,and RMSE=0.91 m.Finally,the FSV of Hunan province was calculated under the RF algorithm,with an overall accuracy of 59.67%.