叶面积指数（Leaf Area Index, LAI）定义为地表单位面积上植被单面叶片面积的总和，是用于表示植被生长状态的重要参数。由于地表覆盖的植被具有空间分布范围广、生长时间呈周期变化的特点，叶面积指数常被作为生态、气候、环境演变研究的关键驱动变量。因此，定量获取时间序列连续、空间覆盖完整、检验精度高的叶面积指数遥感产品，在研究植被生长状态、覆盖变化等相关工作中具有重要理论意义和实际应用价值。现有遥感估算LAI的方法主要包括基于经验模型的估算方法，基于物理模型的反演方法，以及非参数的估算方法。基于统计拟合遥感和地面观测数据的经验模型估算植被参数的方法，实现方式便捷，但往往缺少对模型较完善的物理解释，特别是高维数据集的出现，导致其应用有局限。遥感物理模型建立传感器获得的观测值与植被参数之间的关系，具有较明确的物理解释，但模型复杂、反演困难。非参数估算方法可提取对植被状态的遥感观测值同目标参数的映射关系，但参数估算精度常受限于映射关系所含信息量，如查找表数据容量。由于时间序列遥感观测数据可提供植被时序生长过程中辐射信号的动态变化信息、地表植被空间分布特征，如何用动态模型表达其时空变化特征，进而应用于改进LAI反演精度，是遥感反演研究的热点。所以，本文研究基于多角度遥感观测数据的时空动态信息估算叶面积指数的方法，主要研究内容和结论如下：1、使用时序多角度遥感观测数据估算植被LAI使用时序多角度遥感观测（MOD09GA）数据，发展了一种时序叶面积指数遥感反演算法。在基于数据机理（Data-based mechanistic, DBM）的时序数据建模和反演方法的支持下，使用基于辐射传输理论的核驱动模型及SAILH模型对植被冠层主平面热点、冷点及天顶观测方向的反射率数据进行了计算和模拟，引入各向异性指数(ANIX)作为表示植被冠层二向反射分布的特征信息，发展了一种时序LAI建模和估算的参数化模型（LAI_DBM），最终实现了LAI反演的动态建模和估算。对实验站、研究区时序LAI估算结果的分析表明：1）在LAI_DBM建模中采用时序天顶观测表达植被生长状态变化的信息，可以降低由直接使用时序多角度数据因观测几何变化所引入的反演噪声。2）在时序LAI动态建模和估算过程引入各向异性指数，可以有利于补充植被冠层二向反射可用的观测信息，有效改进了LAI估算精度。3）本研究实验区LAI估算结果的时间连续性和数值稳定性优于MODIS LAI产品，算法有望用于更大空间范围的时序LAI估算。2、贝叶斯最大熵方法的时空LAI缺失值填补使用MODIS LAI产品数据（MOD15A2），及与其匹配的归一化各向异性指数（Normalized red and NIR ANIX, NDAX），发展了一种时空LAI缺失值填补的后处理方法（LAI_BME）。研究分析了LAI与归一化各向异性指数（NDAX）的关联特征，将NDAX作为指示植被生长状态的先验信息，使用Bayesian Maximum Entropy (BME)参数化模型实现时空LAI缺失值填补的处理，得到时空分布信息完整的LAI数据。对研究结果的分析发现：1）归一化各向异性指数可表示植被冠层的二向反射分布特征，时间序列各向异性指数产品可表示植被生长状态或LAI的变化信息。2）贝叶斯最大熵方法可以将与LAI相关的不同数据的统计信息进行融合，通过后验概率更新的计算形式实现空间LAI缺失数据的估算。该统计方法具备将多维数据信息融合的潜力。3、不同叶面积指数的组合计算由于不同反演模型估算的LAI数据具备其各自的优势特征，研究实现了将两种LAI数据进行组合计算，以此改进时序LAI估算结果的精度。使用GLASS和 LAI_DBM 两种模型估算的LAI数据，从多年积累的遥感和地面测量数据中提取背景参考值，用于构建代价函数, 用Levenberg-Marquardt algorithm (LMA)非线性寻优算法求解代价函数，得到两种LAI数据的组合计算权重系数，实现了对两种LAI数据的组合计算（LAI_COMB）。研究同时计算了LAI时序数值之间的相关性并对相关性数值做特征函数的拟合，该特征函数被用作构建代价函数并估算组合权重，同样可以获得LAI数据的组合计算结果。对组合计算结果的检验和分析可知，不同LAI数据的组合计算可改进时序LAI的估算结果精度，同时也为不同反演模型的组合计算提供了有价值的研究方向。

Leaf area index (LAI) defined as the one-side green leaf area per unit ground surface area. It is an essential parameter to represent both the vegetation land cover characteristics and vegetation growth status. In general the LAI is a key variable in ecosystem, climate, and environment dynamic changing research, it can provide a spatial distribution for the land surface vegetation coverage, and a temporal sequence continuous growth information for the vegetation canopy. Then spatiotemporal high accuracy LAI retrieval is important for the remote sensing theoretical research and application.This paper firstly discussed the retrieval methodology characteristics and shortcomings of the current LAI data, such as the empirical estimation model using the vegetation index, the physical model using the satellite observations to retrieve vegetation LAI. Generally the empirical model can be applied in a simple structure, however it has the shortage in explaning the retrieval results. The physical model can initialize the relationship between vegetation canopy observations and LAI data, it can provide explanations for the retrieval procedure, but it has a complex model structure. The nonparameteric method can establish the relationship between vegetation canopy observations and LAI by using the reference data set, but the calculation efficiency is influenced by the size of the reference data set (e.g. the capacity of the lookup table).The new trend of the LAI retrieval research focus on considering the vegetation canopy bidirectional reflectance characteristics and introduce the canopy anisotropy information into the retrieval model. This paper retrieval the vegetation LAI using spatiotemporal information generated from the multi-angular remote-sensing data. The new developed model was tested using different types of vegetation study area and the retrieval results were validated using ground measurements. The research conclusions are summarized as follows:1. LAI estimation using multi-angular remote-sensing observation time seriesA temporally continuous LAI estimation approach was developed using Moderate Resolution Imaging Spectroradiometer (MODIS) multi-angular observations. The RossThick-LiSparse-Reciprocal (RTLSR) model, and Scattering by Arbitrarily Inclined Leaves with Hotspot (SAILH) model are used to generate the hotspot, dark spot, and nadir viewing reflectance. Anisotropic index (ANIX) time series are used as an auxiliary variable to represent the bidirectional reflectance anisotropy of the vegetation canopy. Based on data-based mechanistic (DBM) mothod, a dynamic LAI modeling and retrieval procedure was developed (LAI_DBM). The preliminary results show that: 1) the LAI_DBM approach using nadir viewing reflectance observation and anisotropic index time series can be used to improve the continuity of estimated LAI time series. 2) An anisotropic index time series can represent the vegetation-canopy bidirectional reflectance anisotropy information and its dynamic changes. It works well in the retrieval procedure for improving LAI time-series estimation. 3) The preliminary retrieval results demonstrate that the estimated LAIs can achieve better time-series continuity than the original MODIS LAI product.2. Spatial LAI missing data gap filling use bayesian maximum entropy methodAn LAI spatial missing data gap filling post processing method was developed using MODIS LAI and Normalized red and NIR ANIX (NDAX) product. The research analyzed the scatter dots distribution between the LAI and NDAX data, then the NDAX data was used as the LAI spatial distribution prior-knowledge. The bayesian maximum entropy method (LAI_BME) calculation can generate the post processed LAI in pixel scale and it contains a complete spatial information. Based on the research: 1) the post processed LAI demostrated that the NDAX anisotropy information can be introduced into vegetation LAI estimation, and it can provide auxiliary information to indicate the the vegetation canopy bidirectional reflectance distribution characteristics. 2) Bayesian maximum entropy method has the potential to assemble the different prior information and estimate the post-prior LAI data.3. LAI combinate calculation using different LAI dataA combinate calculation using GLASS and LAI_DBM LAI data was examined. The paper analyzed the LAI temporal covariance can be used to represent the variations of the LAI time series, and the covariance characteristics can be fitted by using empirical model. Levenberg-Marquardt algorithm (LMA) non-liner optimization algorithm was used to minimum the cost function which was initialized by reference and combinated LAI. Then the combination weight generated from the cost function was used to combinate the GLASS and LAI_DBM LAI. According to the preliminary calculation, the combination of the different LAI can provide a satisfactory input data for the LAI retrieval model, it also implemented some meaningful information for the combination of the different LAI retrieval model.