由人类大量排放的碳引发的全球变化问题正在不断恶化，引发了许多灾害。在这一过程中，陆地生态系统发挥着至关重要的作用，它们在维持全球碳循环和减缓气候变化方面具有关键性的功能。净初级生产力（Net primary productivity, NPP）与总初级生产力（Gross primary productivity, GPP）的比值被视为陆地植被的碳利用效率（Carbon use efficiency, CUE），代表陆地植被从大气中吸收碳的能力，也是理解陆地生态系统碳循环和碳分配的重要参数。GPP代表了最初进入陆地生态系统的物质和能量，直接反映自然条件下陆地生态系统的生产力，是陆地碳循环的关键组成部分。准确量化GPP及其动态时空变化，从而提高CUE估测的准确性，不仅是生态系统功能评估和碳平衡研究的重要前提，还可作为评价陆地生态系统对人类可持续发展支撑能力的重要指标。

本文选取多源全球GPP和NPP数据产品作为主要研究目标，同时利用原位观测数据、卫星遥感数据和其他全球数据集进行辅助分析，使用数理统计学方法、生态过程模型、数据同化方法、机器学习算法等，分别在全球GPP数据产品的不确定性分析、融合多源GPP数据产品、陆地植被碳利用效率的时空特征分析、陆地植被碳利用效率对全球变化和植被状态的响应四个方面开展研究。本文的主要研究内容和结论包括：

（1）在总初级生产力GPP难以直接测量，地面观测和通量站点观测仅适用于有限的空间范围的背景下，无论是基于观测研究还是模型模拟研究，GPP估算结果总是受到环境因素、模型结构以及不同植被类型的影响，这意味着不同时空尺度下的估算结果存在很大的不确定性。本研究详细地分析了当前全球45套GPP数据产品，对不同时空尺度的GPP估算方法及其应用潜力进行对比分析，综合比较其特点，并总结了在不同研究中GPP产品的选择指南。同时本文对这45套GPP数据产品的不确定性进行了定量化的分析，结果显示机器学习产品总体上具有较低的不确定性，且GPP产品的不确定性相比于空间变异性更多地体现在空间模式上。最后进一步得到了土壤湿度和降水是当前全球GPP产品不确定性的主要来源。

（2）基于贝叶斯原理的三角帽方法，融合了多源的GPP数据产品，综合了当前主流GPP产品的特点，生产出一套全新的GPP数据集BTCH-GPP。在全球尺度逐网格针对各套GPP数据产品的不确定性，每年每月赋予每套GPP产品不同的权重，以此得到一套融合后的新GPP数据产品。该套GPP数据产品在与通量塔站点数据及日光诱导叶绿素荧光（Solar-induced Chlorophyll Fluorescence，SIF）的评估和验证下，被证明大大减小了随机误差，并能很好地再现GPP的全球分布格局和长期年际变化趋势。BTCH-GPP为未来从大尺度的遥感数据集中提取有效数据，整合模型模拟结果和实验数据提供了一个新思路，同时BTCH-GPP也为全球陆地植被CUE提供了新的独立组分，有利于提升CUE估计的准确度。

（3）基于拥有独立组分的BTCH-GLASS和BTCH-GIMMS的2套新CUE数据集以及多模型模拟的29套CUE数据集，开展了全球范围的陆地植被CUE的时空特征分析。总体而言，CUE年变化趋势不显著，空间分布存在纬度梯度，高纬度地区的CUE值要高于低纬度地区。季节上，CUE值夏季最高，秋冬季最低。另外对比多源CUE数据集后发现，31套全球CUE数据集的表现存在显著差异，主要体现在数值大小方面，范围从0.31±0.001到0.57±0.002，主要由NPP估计的差异造成。空间上的差异分布在北半球高纬度地区以及澳大利亚、非洲南部及沙漠边缘的差异较大，对于多年平均CUE值，误差可达0.3以上。

（4）在全球尺度和站点尺度分别对CUE的驱动机制进行分析，结果显示在全球范围内，CUE对气候变化、土地利用变化、大气CO₂浓度和氮沉降的时空响应各有不同。从时间上，全球陆地植被CUE的变化主要是由大气CO₂浓度（所有模型的平均贡献为28.2%）和土地利用变化（37.4%）的年变化趋势贡献，以及气候变化（51.8%）的年际变化所贡献的。从空间上，主要由气候变化的年际变异主导的CUE的增加，模型间空间占比从57.9%到99.2%。而站点尺度上，在气候数据、土地类型及植被特征等数据集的驱动下，利用原位观测数据和卫星遥感数据对碳水耦合过程模型进行参数的优化，进行了两组数据同化实验。发现植被状态信息能显著改变植被CUE的分布模式，并降低其估计值，主要是通过降低树干碳库的周转率和增加土壤碳对温度的适应性及模型对水分条件的敏感性。针对这两组实验的模型模拟结果，分别在CUE的时间和空间变异性上进行归因，得出在时间上影响CUE的因子主要是大气CO₂浓度、气温以及饱和水汽压差（VPD），在空间上，土壤属性与风速及短波辐射与光合光子通量密度的非线性增强交互作用对CUE的影响最大。

综上所述，本研究为完善定量总初级生产力的研究方法和模型选择过程提供参考，为理解和模拟全球变化下的陆地植被碳利用效率提供理论依据和科学支撑，从而为估算区域碳循环固碳总量、碳分配关键参数、进行碳中和核算以及制定和实施地方碳减排政策提供有效的研究方法。

The global challenge posed by the substantial human carbon emissions has intensified, resulting in numerous disasters. Terrestrial ecosystems play a pivotal role in maintaining the global carbon cycle and mitigating climate change. The ratio of net primary productivity (NPP) to gross primary productivity (GPP) is recognized as the carbon use efficiency (CUE) of terrestrial vegetation, symbolizing the capacity of terrestrial vegetation to sequester carbon from the atmosphere. GPP represents the initial influx of materials and energy into terrestrial ecosystems, directly reflecting their productivity under natural conditions and constituting a critical component of the terrestrial carbon cycle. Precise quantification of GPP, alongside its dynamic spatial and temporal variations, is essential not only as a fundamental requirement for assessing ecosystem functionality and researching carbon balances but also as a crucial metric for evaluating the capability of terrestrial ecosystems to support sustainable human development.

This study focuses on multi-source global GPP and NPP data products as its primary research objectives. In addition, it leverages in-situ observational data, satellite remote sensing data, and other global datasets for supplementary analysis. Various analytical approaches such as mathematical statistics, ecological process models, data assimilation methods, and machine learning algorithms are employed. The research is conducted in four main areas, namely, the uncertainty analysis of global GPP data products, the integration of multiple GPP data products, the uncertainty analysis of terrestrial vegetation carbon use efficiency, and the response of terrestrial vegetation carbon use efficiency to climate change and vegetation status.

The primary research findings and conclusions of this study are as follows:

 (1) Given the challenge of directly measuring gross primary productivity (GPP) and the limitations of ground observations and flux tower measurements, which are only applicable within limited spatial scales, GPP estimation results are invariably influenced by environmental factors, model structures, and variations in vegetation types in both ground observations and model simulation studies. This implies that there is significant uncertainty in estimation results across various spatial and temporal scales. This study thoroughly analyzes 45 global GPP data products currently available, comparing the methods used for GPP estimation at different spatial and temporal scales and assessing their potential applications. It comprehensively summarizes their characteristics and provides guidelines for selecting GPP products in various research contexts. Additionally, this study quantitatively analyzed the uncertainty of 45 global GPP data products. The results show that machine learning products, in general, have lower uncertainty, and the uncertainty of GPP products is primarily manifested in spatial patterns compared to spatial variability. Furthermore, soil moisture and precipitation are the main sources of uncertainty in current global GPP products.

(2) The the Bayesian three-cornered hat approach (BTCH), based on the principles of Bayesian theory, integrates multiple sources of GPP data products, combining the characteristics of current mainstream GPP products to generate a novel GPP dataset, BTCH-GPP. At a global scale, it assigns different weights to each set of GPP products on a grid-by-grid basis to account for their uncertainties on a monthly and yearly basis. This process results in a fused GPP data product. BTCH-GPP has been validated and evaluated against eddy covariance tower data and Solar-induced Chlorophyll Fluorescence (SIF), demonstrating a significant reduction in random errors and the ability to effectively replicate the global distribution patterns and long-term interannual trends of GPP. BTCH-GPP not only offers a novel approach for extracting valuable data from large-scale remote sensing datasets, integrating model simulation results and experimental data, but also provides a new independent component for global land vegetation carbon use efficiency (CUE) estimation, contributing to enhanced accuracy in CUE estimation for future applications.

(3) Based on the two new CUE datasets, BTCH-GLASS and BTCH-GIMMS, each incorporating independent components, as well as 29 sets of CUE datasets derived from multiple model simulations, a global-scale spatiotemporal analysis of terrestrial vegetation CUE was conducted. In general, the annual trend of CUE does not show significant changes. There is a latitudinal gradient in its spatial distribution, with higher CUE values in high-latitude regions compared to low-latitude regions. Seasonally, CUE values are highest in summer and lowest in autumn and winter. Furthermore, upon comparing CUE datasets, it was found that there are significant differences in the performance of 31 global CUE datasets, primarily in terms of numerical values, ranging from 0.31±0.001 to 0.57±0.002, mainly attributed to differences in NPP estimates. Spatial differences are more pronounced in high-latitude regions of the Northern Hemisphere, as well as in Australia, southern Africa, and desert margins, with errors in multi-year average CUE values exceeding 0.3.

(4) The driving mechanisms of CUE were analyzed at both the global and site scales. The results show that globally, CUE exhibits diverse spatiotemporal responses to climate change, land-use change, atmospheric CO₂ concentration, and nitrogen deposition. Temporally, changes in global terrestrial vegetation CUE are primarily attributed to the annual trends in atmospheric CO₂ (contributing 28.2% on average across all models) and land-use change (37.4%), as well as the interannual variability in climate (51.8%). Spatially, the increase in CUE is mainly dominated by interannual variability in climate change, ranging from 57.9% to 99.2% among models. At the site scale, two sets of data assimilation experiments were conducted using in-situ observations and satellite remote sensing data to optimize parameters of carbon-water coupled process models, driven by climate data, land type, and vegetation characteristics. It was found that vegetation state information significantly changes the distribution pattern of vegetation CUE and reduces its estimates, primarily by reducing the turnover rate of stem carbon pools, increasing soil carbon adaptability to temperature, and decreasing the model's sensitivity to moisture conditions. For the model simulation results from these two sets of experiments, attributions were made to the temporal and spatial variability of CUE. The main factors influencing CUE over time were found to be atmospheric CO₂ concentration, temperature, and VPD (Vapor Pressure Deficit). Spatially, the nonlinear enhancement of interactions between soil properties and wind speed, as well as shortwave radiation and light quantum flux density, had the greatest impact on CUE.

In summary, this study provides a reference for improving the research methods and model selection processes for quantifying GPP. It offers a solid theoretical foundation and robust scientific support for comprehending and simulating land vegetation carbon use efficiency in the context of global changes. Consequently, it serves as an effective research methodology for estimating regional carbon cycle sequestration, determining crucial carbon allocation parameters, conducting carbon neutrality assessments, and formulating and implementing local carbon emission reduction policies.