植被光合有效辐射吸收比（Fraction of absorbed Photosynthetically Active Radiation, FPAR）定义为植被吸收400-700nm间光合有效辐射（Photosynthetically Active Radiation, PAR）占总入射光合有效辐射的比例，直接反映植被光合作用强弱，是陆地生态系统模型、碳循环模型中的主要输入参量。卫星遥感是获取大范围、时间连续FPAR的唯一有效手段。利用卫星遥感手段准确反演FPAR，一直以来是定量遥感研究中的热点问题，现已发展出多种FPAR估算方法。作为植被总初级生产力（Gross Primary Production, GPP）估算中唯一的遥感变量，FPAR估算的精度直接影响GPP的估算精度。但是，已有GPP估算研究大多直接采用某种算法，很少有研究系统地对FPAR不同估算方法及其在GPP估算中的应用进行比较。

本文基于全球共享地面观测网络数据集（VALERI和AmeriFLUX），选取8种在GPP估算中应用较为广泛的FPAR经验线性模型，利用Landsat数据对FPAR进行估算，同时选取MODIS FPAR产品作为FPAR物理算法的代表，对不同方法FPAR估算结果进行了精度验证与对比；在此基础上，利用改进的光能利用率模型对AmeriFLUX站点区域GPP进行了估算，对不同的GPP估算结果进行精度验证与对比，分析了FPAR反演精度对GPP估算的影响，为未来提高FPAR和GPP估算精度提供科学依据和参考。主要研究内容如下：

（1）选取8种分别基于归一化差值植被指数（Normalized Difference Vegetation Index, NDVI）、增强型植被指数（Enhanced Vegetation Index, EVI）、土壤调节植被指数（Soil-Adjusted Vegetation Index, SAVI）建立的经验模型，利用Landsat数据对11个VALERI站点以及6个AmeriFLUX站点范围内FPAR进行反演，并选取基于物理算法的MODIS FPAR产品，分别与实测数据进行对比验证。结果表明：基于NDVI的经验模型整体精度最高，其次是MODIS FPAR，最后是基于EVI与SAVI建立的经验模型。这说明基于NDVI的经验模型在小范围区域精度较高，且算法简单方便，基于物理算法的MODIS FPAR产品兼具普适性和精度较高的优点，更加适用于大范围长时序FPAR估算。

（2）基于上述不同方法反演得到的FPAR，结合站点实测PAR、温度等环境因子和不同地表类型的最大光能利用率数据，通过改进的光能利用率模型对4个同时具备FPAR与GPP数据的AmeriFLUX通量站点的GPP进行估算，利用站点实测GPP对结果进行验证与对比。结果表明：GPP估算精度在耕地类型站点较高、在林地上较低；综合不同地表类型结果，将EVI近似为FPAR的模型得到的GPP精度最高，其次是基于SAVI的经验模型和基于NDVI的经验模型，基于EVI线性变换的FPAR估算模型得到的GPP精度最差。而由MODISFPAR估算得到的GPP精度比使用NDVI进行估算的经验模型精度较低，且更适用于高植被覆盖区。因此，当需要使用长时间序列GPP数据时，可以直接使用MODIS FPAR进行估算，精度较高，此外基于NDVI的经验模型对FPAR和GPP的估算精度都较高，对耕地和林地类型均具有适用性。将EVI近似为FPAR和基于SAVI的经验模型在本文选取的站点上对FPAR存在低估，而其GPP估算精度较高，这说明GPP估算同样受到光能利用率模型精度和其它环境因子的影响。

（3）采用敏感性分析、站点实测GPP与相关因子分析、FPAR与GPP估算精度对比等方法，分析了FPAR反演精度对GPP估算的影响。结果表明：对于耕地类型，FPAR的总敏感度最高，而对林地类型，FPAR的总敏感度低于PAR和温度的总敏感度；在相关因子分析中，FPAR与GPP间的相关性较高，仅次于APAR和温度；在所有选取的FPAR估算模型中，FPAR反演精度与GPP估算精度直接相关，表现为FPAR反演精度高则GPP估算精度也较高，反之亦然。本文进一步分析了采用不同方法时FPAR估算不确定性（即不同方法估算FPAR的标准差）与GPP估算不确定性间的关系，发现FPAR估算的不确定性与GPP估算的不确定性基本呈正相关，但也与入射PAR等环境因子相关，说明提高FPAR估算精度对于进一步提高GPP估算精度具有重要意义。

FPAR (Fraction of absorbed Photosynthetically Active Radiation) is defined as the proportion of thetically Active Radiation absorbed by the vegetation at 400-700nm as a total incident PAR(Photosynthetically Active Radiation), which directly reflects the intensity of vegetation photosynthesis. It is the main input parameter in terrestrial ecosystem model and carbon cycle model. Satellite remote sensing is the only effective means to acquire large scale, long time series FPAR. Accurate FPAR inversion by satellite remote sensing has always been a hot issue in quantitative remote sensing research, and a variety of FPAR estimation methods have been developed. As the only remote sensing variable in the estimation of GPP(Gross Primary Production), the accuracy of FPAR estimation directly affects the estimation accuracy of GPP. However, most of the existing GPP estimation studies directly use an algorithm, and few studies systematically compare different FPAR estimation methods and their applications in GPP estimation.

Based on the global shared Ground Observation network data set (VALERI and AmeriFLUX), eight empirical linear models of FPAR that are widely used in GPP estimation are selected in this paper. Landsat data is used to estimate FPAR, and MODIS FPAR product is selected as the representative of FPAR physical algorithm. The accuracy of FPAR estimation results of different methods was verified and compared. On this basis, the improved light energy utilization model was used to estimate the GPP of AmeriFLUX site area, the accuracy of different GPP estimation results was verified and compared, and the influence of FPAR inversion accuracy on GPP estimation was analyzed, providing scientific basis and reference for improving the accuracy of FPAR and GPP estimation in the future. The main research contents are as follows:

(1) Eight kinds of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (Enhanced Vegetation Index, NDVI) were selected. EVI and Soil-Adjusted Vegetation Index (SAVI) were used to invert FPAR at 11 VALERI sites and 6 AmeriFLUX sites. The MODIS FPAR product based on the physical algorithm is selected for comparison and verification with the measured data. The results show that the overall accuracy of the empirical model based on NDVI is the highest, followed by MODIS FPAR, and finally the empirical model based on EVI and SAVI. This shows that the empirical model based on NDVI has high accuracy in a small area, and the algorithm is simple and convenient. The MODIS FPAR product based on the physical algorithm has the advantages of both universality and high accuracy, and is more suitable for large-scale long time series FPAR estimation.

(2) Based on the FPAR obtained by the above inversion methods, combined with the measured site PAR, temperature and other environmental factors as well as the maximum light utilization data of different surface types, the GPP of AmeriFLUX flux sites with FPAR and GPP data was estimated by using the improved light utilization model. The measured GPP was used to verify and compare the results. The results showed that the accuracy of GPP estimation was higher in cultivated land type sites and lower in forest land. Combining the results of different land surface types, the model approximating EVI as FPAR has the highest GPP accuracy, followed by the empirical model based on SAVI and the empirical model based on NDVI, and the FPAR estimation model based on EVI linear transformation has the worst GPP accuracy. The accuracy of the GPP estimates from MODIS FPAR is lower than that of the empirical model using NDVI, and it is more suitable for high vegetation cover area. Therefore, when long time series GPP data is needed, MODIS FPAR can be directly used for estimation with higher accuracy. In addition, the empirical model based on NDVI can estimate both FPAR and GPP with higher accuracy, and is applicable to both cultivated land and forest land types. The approximation of EVI to FPAR and SAVI based empirical models underestimate FPAR at the sites selected in this paper, while the accuracy of GPP estimates is high, suggesting that GPP estimates are also affected by the accuracy of light utilization models and other environmental factors.

(3) The influence of FPAR inversion accuracy on GPP estimation was analyzed by using sensitivity analysis, site measured GPP and correlation factor analysis, and comparison of FPAR and GPP estimation accuracy. The results showed that the sensitivity of FPAR to cultivated land was the highest, but that of FPAR to forest land was lower than that of APAR and temperature. In all the selected FPAR estimation models, the accuracy of FPAR inversion is directly related to the accuracy of GPP estimation. The higher the accuracy of FPAR inversion is, the higher the accuracy of GPP estimation is, and vice versa. This paper further analyzed the relationship between the uncertainty of FPAR estimation (the standard deviation of FPAR estimation by different methods) and the uncertainty of GPP estimation. It was found that the uncertainty of FPAR estimation was positively correlated with the uncertainty of GPP estimation, but was also correlated with incident PAR and other environmental factors. It is important to improve the accuracy of FPAR estimation for GPP estimation.