Estimation Of Forest Stock Based On Optimal Design Of The Sampling System In Forest Inventory And Analysis

Although China's current forest resources monitoring system meets the basic requirements for forest resource surveys,there is some deficiency,such as poor survey results sharing,high survey costs,lack of annual monitoring,inadequate forest health monitoring and so on.As a forestry developed country,the United States' forest resources inventory system provides a solution and reference for our solving this situation.This study referred to the United States' FIA(Forest Inventory and Analysis)three-phase sampling system,with Yanqing District in Beijing as the study area,and forest area and stock volume as the research object,designing different sampling plans for sampling and comparison of forest resources.Finally we got the following conclusions:(1)At the first-phase sampling level,regular hexagonal sampling grids with different side lengths were systematically placed in the study area as the sampling frame.The optimal hexagonal sampling frame was determined by analyzing the coefficient of variation,and the optimal side length of hexagon was determined.The optimal length of the sampling grid frame was 3000m,and the area of each first-order plot was 2338.27 hectares;(2)At the second-phase sampling level,the intrinsic factors of the four-point cluster plots were studied.The internal factors of the cluster plots include the size of the plots,the plot distances within the clusters,and the plot numbers within the clusters,etc.The optimal number of sub-sample plots within the group was 45 and the three non-center subsample plots are located at 0°,120°,and 2400 of the central subsample plot.The scheme of the optimal plot distance within the cluster was 36 m.The best design plan for the size of the sample plot was a circular plot with a radius of 7m.(3)Based on the determined sampling frame size and cluster plots,using the ground sample data and GF-1 remote sensing image data,the forest volume index estimation model was construct based on partial least-squares regression and k-NN methods.The remote sensing feature variables used were NDVI,RVI,DVI,SAVI,NLI,and other four variables.And through independent variable selection,the final selection of EVI5 NDVI,NLI,RVI,SAVI was the five variables as a preferred variable for the accumulation model construction.Based on partial least-squares regression method and k-NN method,the forest volume was modeled respectively,and two methods were used to evaluate root mean square error(RMSE)and relative root mean square error(RRMSE).The final result was that the use of partial least-squares regression model for the reservoir volume inversion result was 2.7088 million m3,the relative error was 0.4579 million m3,and the estimation accuracy was 78.82%;and the k-NN method for the accumulation volume inversion result was 1.8440 million m3,the relative error was 0.3117 million m3,and the estimation accuracy was 85.46%.Under the premise that the continuous investigation of forest resources in Beijing stipulates the survey accuracy of 85%,the inversion results of the forest volume in the Yanqing District based on the k-NN method are better than the inversion results of the partial least-squares regression method.The root mean square error of forest volume estimated based on partial least squares regression was 9.323 m3/hm2,and the relative root mean square error was 37.1%.The root mean square error of forest volume estimation based on k-NN method was 7.739 m3/hm2.The relative root mean square error was 33.6%.In summary,the k.NN method is more effective than the partial least-squares regression method.It can be used as a reference for future research on remote sensing inversion of forest volume in Yanqing or the same latitudes.In summary,the forest volume estimation method based on the third-phase sampling could be implemented in Yanqing District of Beijing,China.The specific optimization design scheme was feasible and efficient.