极化干涉合成孔径雷达（Polarimetric Interferometry Synthetic Aperture Radar，PolInSAR）已被广泛用于森林高度的反演，正确评估模型输入参数、模型假设、林分结构、立地条件等引起的不确定性是提高基于PolInSAR技术森林高度反演精度及准确性的关键之一。本文以贝叶斯模型为基础，以模拟的L波段PolInSAR数据为数据源，首先基于贝叶斯模型确定了随机体散射（Random Volume over Ground，RVoG）模型输入参数引起的不确定性，在此基础上使用先验知识（成像中森林高度的值）对RVoG模型的消光系数进行“固定”，并反演了森林高度；然后基于RVoG模型反演结果及贝叶斯后验采样分析，讨论了树种、森林密度、地面粗糙度及土壤含水量四个因子变化引起的森林高度反演结果的不确定性。研究结果表明：对于L波段的PolInSAR模拟数据，采用RVoG模型进行森林高度反演时，使用先验知识对消光值进行固定可大大降低森林高度反演的不确定性；四个因子中，树种和森林密度引起的不确定性较显著，然后为地面粗糙度，最后为土壤含水量。阔叶林反演结果的不确定性明显高于针叶林；森林密度从150株/hm2增至1200株/hm2时，其标准误最高可下降67.5%；在针叶林纯林和阔叶林纯林中，地面粗糙度与反演结果的标准误呈现明显的正相关关系；土壤含水量引起的不确定性最小，几乎可以忽略不计。

Polarimetric Interferometry Synthetic Aperture Radar (PolInSAR) has been widely used in forest height inversion. Correct evaluation of the uncertainty caused by model input parameters, model assumptions, stand structure, and site conditions is one of the keys to improve the accuracy of forest height inversion with PolInSAR technology. In practical application, the study of uncertainty of forest height inversion is as important as the study of forest height estimation methods. Quantification of global carbon stocks based on forest biomass calculations usually requires reducing the error in biomass estimates through forest height. The uncertainty of forest height may come from model input parameters, model assumptions, observed data and forest scene factors. However, there are few comprehensive collaborative impact analyses on the uncertainty of forest height inversion results. In view of this, it is necessary to study the uncertainty of forest height inversion using PolInSAR technique. In this paper, based on the simulated L-band full PolInSAR data, we first analyze the uncertainty caused by the input parameters of the RVoG (Random Volume over Ground) model based on the Bayesian model, and then uses prior knowledge(the value of the forest height in the imaging) to fix the extinction of the RVoG model，next we inversed the forest height. The results shows that a priori knowledge can thus greatly reduce canopy height uncertainties in some cases. Based on these, we combine RVoG model and Bayesian framework, use L-band simulated PolInSAR data, comprehensively explored the uncertainties resulted from input parameters of RVoG model, the model hypothesis, observation value, the changes of forest tree species, forest density, surface properties, ground moisture content and other factors in the process of forest height inversion. With the research, we concluded than: (1) The prior knowledge can reduce the uncertainty of the forest height inversion (by fix the extinction value) with RVoG model and L-band PolInSAR data. (2) The forest height inversion results are greatly affected by forest tree species, the inversion results uncertainty of coniferous forest is lower than that of broad-leaved forest. (3) The change of forest stand density has a significant influence on the uncertainty of the forest height inversion results. The higher the density, the lower the uncertainty, especially in the pure coniferous forest. When the forest density is small, the uncertainty of the forest height retrieved by RVoG model is large. When the forest stand density increases from 150 plants/hm2 to 1200 plants/hm2, the uncertainty decreased down about 67.5%. (4) The change of surface roughness has a positive correlation with the uncertainty of forest height inversion results, the greater of the roughness, the higher of the uncertainty. (5) Compared with the other factors, the uncertainty caused by ground moisture content is very small and can be ignored.