森林蓄积量作为评估森林数量的关键指标之一，是全国森林资源调查的一项重要内容。由于技术水平和资金的限制，传统的蓄积量估测方法存在着精度低，时效性差等不足。随着GIS与遥感技术的快速发展和普及应用，采用高分辨率遥感影像结合少量地面调查样地进行森林蓄积量估测，已成为众多林业科技工作者研究的热点方向。目前国内基于GIS和遥感影像进行蓄积量估测的研究主要是使用传统的多元线性回归方法，取得了一定成果,但是当估测因子间存在多重复共线性时，会严重影响最小二乘法等传统回归方法的稳定性，最终导致估测精度下降。 本文采用近年来发展较快的偏最小二乘回归方法建立蓄积量估测模型，用所得模型估测森林蓄积量,并且从非参数化角度使用机器学习经典算法之一的K-近邻法进行蓄积量估测，以从估测精度和算法效率上与偏最小二乘法进行对照。此外，本文还对一种基于偏最小二乘的Bootstrap变量向后筛选方法和遥感影像分类在森林蓄积量估测中的应用进行了探讨。最后使用C#语言和ArcGIS Engine开发组件对所研究的蓄积估测主要算法进行开发实现。 为验证算法效率及精度，本文使用北京市密云县第七次全国森林资源一类清查样地数据和经过几何校正的同期TM遥感影像及30米分辨率DEM数据作为基础数据，用开发的软件从中提取相关信息进行密云县森林蓄积量估测。在样地较少的情况下使用偏最小二乘回归建模计算得到的蓄积量预报偏差均方根误差为6.853m3/hm2,相对均方根误差为18.289%；使用K-近邻法进行蓄积估测时，采用交叉验证方法计算其样本偏差，所得均方根误差为7.369 m3/hm2，相对均方根误差为19.009%，整体效果较好，表明这两种方法均可用于森林蓄积量估测。

Forest stock volume is a key factor in forest quantity evaluation, and it occupies an important position in the National Forest Investigation. Due to the limitation of technology and capital, the traditional forest stock volume estimation method cannot provide high-precision and high real-time result for policy makers. With the rapid development and application of Geographic Information System (GIS) and Remote Sensing (RS), estimating forest stock volume with high-resolution remote sensing image and a small amount of sample plots has become a hot research direction. In recent years, most of the GIS and RS based studies about stock volume estimation interiorly are focused on classical Multiple Linear Regression (MLR) and have made great progress. However there is one problem that the classical Multiple Linear Regression such as Least Square Regression cannot get good result when multicollinearity exists among factors. This paper introduces the Partial Least Squares Regression (PLS-Regression) method to build models to estimate forest stock volume, and use a non-parametric method called K-Nearest Neighbor as a contrast. A PLS based variable selection method—Bootstrap arithmetic and the use of Remote Sensing image classification in forest stock volume estimation are also discussed in this paper. The main algorithms were realized by C# program with the ArcGIS Engine development component. In order to compare the efficiency and the precision of the PLS-Regression and the K-Nearest Neighbor method, this paper use the seventh National Forest Inventory sample plots ,Landsat TM image in same period and 30m cell size DEM of Miyun County of Beijing to estimate its forest stock volume. With a small amount of sample plots, the RMSE of PLS-Regression estimation is 6.853m3/hm2,the Relative Standard Deviation is 18.289%,while the RMSE of K-Nearest Neighbor estimation is 7.369 m3/hm2, the Relative Standard Deviation is 19.009%.The results show that both the PLS-Regression and K-Nearest Neighbor method can be used in forest stock volume quantitative estimation.