【摘要】 聚集指数CI (Clumping Index)是植被冠层的一个重要结构参数，对植被冠层的辐射截获，以及全球碳、水循环的研究均有重要作用。现有星载CI产品的估算主要是基于CI-NDHD (Normalized Difference between Hotspot and Dark spot)线性模型方法，由于针叶林和阔叶林在叶片尺度上存在聚集层级的差异，该模型对它们分别采用了不同的模型系数。但是，该模型对中粗分辨率的针阔混交林像元通常采用阔叶林的CI反演系数，因此，理论上会导致该类型CI的高估。为此，本文提出了一种动态选取混交林像元端元CI组分的方法，以改进针阔混交林植被聚集指数的估算精度。首先，通过国际地圈—生物圈计划（IGBP）的地表类型和描述二向性反射分布函数BRDF (Bidirectional Reflectance Distribution Function)特征的地表各向异性平整指数AFX (Anisotropic Flat Index)进行双重约束，逐像元地计算端元CI值；然后，结合高分辨率的土地覆盖分类数据确定端元在像元中的面积比例，并估算MODIS针阔混交林像元的聚集指数MFCI (Mixed Forest CI)；最后，将方法应用于研究区MODIS数据的MFCI估算，并通过地面实测数据进行精度评价。结果表明：目前的MODIS产品算法高估了针阔混交林像元的CI值，而MFCI估算方法在CI-NDHD算法的基础上，可以较显著地改善该类型聚集指数的估算精度，当针叶林树种成数达到60%时，精度改善可达28.03%，其中，改进结果的均方根误差（RMSE）和偏差（Bias）各降低约84%和175%。研究表明，MFCI方法对针阔混合像元的端元组分的变化敏感，在高分辨率地表分类已知的条件下，MFCI方法为针阔混交林CI产品生产和精度提高提供了可行的解决方案。更多还原

【Abstract】 The foliage Clumping Index(CI) is an important structural parameter of vegetation canopies. The CI influences radiation interception within canopies and plays an important role in the study of global carbon and water cycles. Currently, the widely used method for deriving satellite-borne CI products is based on a linear model constructed on the basis of the CI and the Normalized Difference between the Hotspot and Dark spot(NDHD) angular indices. As coniferous and broadleaf forests exhibit aggregate differences at the leaf scale, the CI inversion model can be applied to a variety of coefficients to generate different CI-NDHD models. Modelers typically use CI inversion coefficients of broadleaf forests to estimate the CI of coniferous-broadleaf mixed forests for medium-coarse resolution pixels, but this approach can theoretically cause a CI overestimation for this landcover type. Thus, in this study, we propose a novel coniferous-broadleaf Mixed Forest CI(MFCI) estimation method to dynamically select the endmember CIs of mixed forests pixel by pixel. The proposed method was successfully applied to satellite-borne MODIS data. The MFCI of the tree-farm study area on Saihanba was estimated, and the accuracy of the results was validated using ground-measured CIs.The MFCI was estimated by considering land cover classes and the Anisotropy Flatness Index(AFX), which describes the basic Bidirectional Reflectance Distribution Function(BRDF) variation. First, the prior values of the endmember NDHD were extracted pixel by pixel by imposing double constraints on the landcover type of the International Geosphere–Biosphere Program and the surface AFX, which characterize the shape of the BRDF. Then, the high-resolution land cover classification data were used to obtain the proportions of the endmembers in the coniferous-broadleaf mixed forest pixels. An optimization factor f was introduced to eliminate the differences between the NDHD of mixed forest pixels and the NDHD prior values of different vegetation cover types based on the NDHD linear mixing assumption. Then, the endmember CIs were calculated. Finally, the endmember CIs, combined with endmember abundance, were used to estimate the MFCIs based on Beer’s law.First, the existing MODIS CI product algorithm does not consider coniferous-broadleaf mixed forest pixels within mixed forest pixels,which leads to overestimation of coniferous – broadleaf mixed forest CIs. When the proportion of coniferous species reaches 60% in a mixed forest pixel, the overestimation of the CI can exceed 35%. Second, the proposed MFCI estimation method based on the CI-NDHD algorithm can significantly improve the CI estimation accuracy of coniferous-broadleaf mixed forest pixels. When the proportion of coniferous forest in the mixed forest pixels reached 60%, the accuracy improved by 28.03%. The root mean-square error and bias for the enhanced results were reduced by approximately 84% and 175%, respectively. Third, the MFCI method is more sensitive than the current MODIS CI products to changes in coniferous and broadleaf forest structures in mixed forest pixels.The current satellite CI products for mixed forest pixels typically use the modeled coefficients of broadleaf forests in the CI-NDHD model, which theoretically implies increased uncertainty in CI products. In this study, the proposed MFCI estimation method was used for coniferous-broadleaf forest mixed pixels. The CI endmembers were dynamically adjusted. The validation based on ground-measured CIs showed that the proposed method was significantly more accurate than the current MODIS CI products in terms of estimating the CI of mixed coniferous and broadleaved forests. In summary, the MFCI estimation method improved the estimation accuracy of mixed forest CI products in the selected study area. The proposed method is a promising technique for further improving the accuracy of global CI products.更多还原