随着我国高分辨率遥感卫星的发射，采用高分辨率遥感图像结合少量地面样地调查资料进行森林蓄积量估测已成为研究热点。在确定了对森林蓄积量估测影响大的遥感和GIS因子后，可根据少量地面样地调查资料对应的遥感和GIS因子建立蓄积量估测方程，进行蓄积量预报，最大限度地减轻森林调查工作量。 传统的抽样，一般是抽取一部分样地建立方程，剩余样地预报精度。用少量样地建立估测方程，到底需要多少样地，样地数量太少，不能保证精度，样地数量太多，要耗费大量人力物力，体现不出估测的优越性。样地越多，包含的信息也就越多，建立的方程也越稳定。使用全部样地建立估测方程可以最大限度地利用样地包含的信息，但是，有两方面问题，第一，全部样地包含的信息可能有异常样地的信息，这对建立方程是有害的；第二，全部的样地参与建立方程，就无法对方程的预报精度进行检验，不知道估测方程是否满足实际精度需求。所以需要一种抽样方法，既能最大限度地利用样地的有利信息，又避免异常样地的有害信息，而且可以使估测方程的预报精度满足实际需求。 本文提出了多次随机抽样平均法。随机抽样，每个样地被抽中的概率是相等的，多次抽样可以保证所有样地都被抽中。对每次抽样的精度进行控制，第一，可以减少异常样地信息；第二，可以保证预报精度。如果残差大，就有理由怀疑包含有异常样地，剩余样地的预报精度小于85%，就不能满足实际需求。只有满足精度要求，才认为这次抽样符合要求。多次随机抽样，对满足精度要求的系数求平均值，得到平均系数，该系数就是最优样地求得的模型系数。 为验证抽样效率及精度，本文使用密云县森林资源一类调查样地数据和校正后的同期TM遥感影像，用软件从中提取波段和比值波段信息进行森林蓄积量估测。对三种抽样方法在最小二乘模型、偏最小二乘模型、稳健估计模型中进行检验，分层多次随机法的中误差最小，精度高，可用于森林蓄积量估测。

Along with the high resolution satellite launch, It，forest stock volume estimation， has become a hot research topic by the use of high resolution remote sensing images combined with a small amount of sample plot survey. After determining remote sensing and GIS factors that have a greater influence on forest stock volume estimate, establish the estimation equation to forecast the forest stock volume on the basis of a small amount of sample plot survey data corresponding to remote sensing and GIS factors, thereby furthest, reduce the forest survey work. The traditional sampling, generally establish equation by extracting partly, forecast accuracy of the remaining sample, by using a small amount of sample to setup estimate equation. After all, need how many sample, that is too little to guarantee the accuracy to meeting the requirements, and too many to demonstrate the superiority of the estimation because of manpower and material resources wastage. And sample quantity is proportional to the contained information, more number of samples make the equation more stable. Use all sample to establish equation can use furthest the information contained in samples. But also there are two problems. Firstly, the total sample contains the abnormal samples information, what is harmful for establishing equation. Second, if all samples used in establishing equation, it cannot test the forecast accuracy of the equation, and cannot know whether the equations satisfy the requirements of accuracy. So need a sampling method, can use farthest the favorable information without harmful information, and get the prediction precision of the equation. It, multiple random sampling average method, is proposed in this paper. For random sampling, the selective probability of every sample pumped are equal, and multiple sampling can guarantee that all samples are selected, meanwhile control each sampling accuracy, to reduce the abnormal sample information and guarantee forecast accuracy. If have a big residual error, there is reason to suspect that contain the abnormal samples, and if the forecast accuracy is less than 85%, it cannot satisfy the actual demand. Ultimately, meet all above, we think that this sample meets the requirements. Multiple random sample, get the average coefficient by average of coefficients. For verify efficiency and accuracy of the sample, this article use the A sample survey data and TM remote sensing image, that adjusted, in the same period of forest resource from Miyun county of Beijing, by software to extract the information of the wave band and the ratio of wave band to estimate the forest stock volume. For three sampling methods, by testing in least squares model, partial least squares model and the robust estimation model, mean square error of hierarchical multiple random method is the smallest and have a high accuracy, can be used in the forest stock volume estimation.