Objective  Taking the overall forest resources of forest farms or counties as the object of investigation, timely and accurately investigating and monitoring of the mean stock volume density（MSVD）will play an important supporting role in the macro management of forest resources, the evaluation of ecological protection value, the measurement of forest carbon reserves, and the performance evaluation of the tenure of leading cadres by the superior departments (such as cities and provinces). It has become an important research direction at home and abroad using remote sensing data, such as satellites and unmanned aerial vehicles, combining with less sampling plot to effectively estimate the overall parameters. At present, there is no comparative validation of several estimation methods for mean stock volume based on multi-source data in domestic. In order to promote the application of remote sensing technology in the forest resource survey, there is an urgent need to compare and evaluate the methods for estimating overall parameters of forest farms or counties.

  Method  The main plantation tree species of Wangyedian Forest Farm in Inner Mongolia of northern China were taken as the research object. Based on the sampling UAV LiDAR (herringbone system distribution), Sentinel-2A data (full coverage) and a small amount of sample plot data obtained in 2019, four patterns of sample plot (p), sample plot-satellite (ps), sample plot-sampling LiDAR (pl), and sample plot-sampling LiDAR-full coverage satellite (pls) were estimated and compared with five methods, including DB_p, MD_ps, MD_pls, HY_pl and MA_ps, which were suitable for these four patterns and belong to design-based method (DB), model-assisted method (MA), model-dependent method (MD) and mixed or hybrid method (HY).

  Result  (1)The mean stock volume densities of DB_p, MD_ps, MA_ps, HY_pl, and MD_pls were 212.54, 202.09, 202.38, 210.07 and 208.96 m³/ha, respectively. The accuracy (P) was 90.44%, 91.54%, 91.69%, 96.35%, and 96.45%, respectively, with variances decreasing in turn. (2) The relative efficiency (RE) of other methods compared with MD_pls was greater than 1 (RE_DBp,MDpls = 5.39, RE_MDps,MDpls = 3.82, RE_MAps,MDpls = 3.69, RE_HYpl,MDpls = 1.07), and the RE of HY_pl compared with MD_pls method was greater than 1, but close to 1. Compared with the other three methods, HY_pl was more efficient (RE_DBp,HYpl = 5.02, RE_MAps,HYpl = 3.43, RE_MDps,HYpl = 3.56). (3) Both HY_pl and MD_pls methods containing LiDAR data had positive efficiency compared with MD_ps and MA_ps methods containing Sentinel-2A data (RE_MAps,HYpl = 3.43, RE_MDps,HYpl = 3.56, RE_MDps,MDpls = 3.82, RE_MAps,MDpls = 3.69). The RE of MD_ps and MA_ps was close to 1, but the efficiency of MA_ps was slightly higher than that of MD_ps (RE_MDps,MAps = 1.04).

  Conclusion  Compared with the estimation method that only used the sample plot data, when the remote sensing data was used as an auxiliary variable to establish the estimation model, it can effectively improve the estimation accuracy of the MSVD of the forest farm, whether the Sentinel-2A multispectral remote sensing data, which is fully covered the forest farm but not sensitive enough to the stock volume, or the sampling LiDAR data which are sensitive to stock volume. Among four methods involving the application of remote sensing data, MD_pls has the highest estimation accuracy, followed by HY_pl. Both of these two methods including the application of LiDAR remote sensing sampling observation data are suitable for the MSVD estimation of forest farms. The MD_pls method has the highest estimation accuracy and relative efficiency, which can realize the synergistic application of the Space-Air-Earth multi-source observation data, and can be used as the preferred method for annual monitoring of forest stock in forest farms.