森林生态系统是陆地生态系统碳固存的重要组成部分，其总初级生产力固定的碳（GPP）约占陆地生态系统GPP的一半。因此，森林光合作用的微小变化都对减轻大气中人为排放CO₂浓度的升高有重要的科学意义。直到目前，基于植被动态过程的陆地碳循环模型一直被认为是估算和预测森林光合作用的主要工具。尽管此类碳循环模型已广泛用于计算森林生态系统中的GPP，但模拟结果仍存在很大的不确定性。对森林生态系统年内和年际的光合生产力进行准确地量化与建模仍然是目前碳循环模拟研究的一个重要目标。

本文基于“热时间”与“冷度日”的方法，开发、验证和应用新的叶物候模型，并将此模型与叶生长过程以及森林碳循环模型FORCCHN进行耦合，进而发展出第二代基于植被动态的森林碳循环模型FORCCHN2。在站点尺度，不仅验证了该模型对模拟森林生态系统叶物候时间和GPP的适用性，还测试了FORCCHN2模型相对于其他植被动态碳循环模型对GPP模拟的改善效果。在区域尺度，利用FORCCHN2估算了1985-2016年北美森林生态系统叶物候时间和GPP。此外，将FORCCHN2模型应用至全球森林生态系统，并利用其模拟的GPP与动态的森林覆盖率数据相结合，探究了森林覆盖率对估算全球森林生态系统GPP的影响。主要结论如下：

（1）FORCCHN2中的物候模型可以再现北美森林生态系统叶物候的时空变化特征。在114个物候相机的森林观测站点，模拟与观测的展叶时间、叶着色时间以及叶生长季长度显著正相关，相关系数r分别为0.89、0.76和0.82；模型效率E分别为0.80、0.58和0.68；相对均方根误差RMSE分别为9.22、17.63和24.63天。模型能较好地捕捉落叶林的叶物候变化，而对于捕捉常绿林的叶物候变化的能力较弱。将该物候模式被首次应用至整个北美0.5°×0.5°的森林区域叶物候研究中，结果表明：1985-2016年，北美森林生态系统的平均展叶日序数为第125.1±3.8天，平均叶着色日序数为第252±24.5天，平均叶生长季长度为126.4±25.8天。此外，整个北美的展叶时间的变化趋势表现为每十年提早0.3天，叶着色时间的变化趋势表现为每十年推迟3.7天，叶生长季长度的变化趋势表现为每十年延长4.0天，该变化趋势与前人研究具有可比性。模拟的变化趋势结果表明在最近几十年中，北美森林的叶生长季长度的延长主要受到了叶着色时间的推迟影响，而叶着色时间的推迟归因于北美大部分地区夏季和秋季显著增温的结果。

（2）FORCCHN2模型能提高北美森林光合作用有关物候模拟能力，并能改善年内及年际GPP的模拟表现。在18个碳通量观测站点，FORCCHN2模型相比于其他6个植被动态碳循环模型（3个诊断性模型和3个预测性模型）而言，模拟与观测的光合作用开始时间、光合作用停止时间的相关系数r增加了0.17-0.73、0.03-0.54；模型效率E增加了0.39-2.57、0.07-1.32。对于逐日GPP模拟的表现而言，FORCCHN2模型在所有预测性模型中表现最佳，但略差于诊断性模型。对于年际GPP模拟的表现而言，FORCCHN2模型在所有模型中具有最好的模拟效果。FORCCHN2被应用到整个北美森林生态系统光合作用的研究中。结果表明：1985-2016年，占北美森林72.4%的北部地区的GPP呈增长的趋势，占27.6%的南部地区的GPP呈减少的趋势。对整个北美而言，森林区域的GPP年均总量为7.45±0.77 Pg C yr^-1，并且整个森林生态系统的GPP在近几十年来的气候变化背景下呈显著增长的趋势。对于北美区域的模拟结果，FORCCHN2模型与其他基于植被动态的模型模拟的GPP年际变化显著相关。其中，基于遥感反演植被动态的诊断性模型CASA-GFEDv2与FORCCHN2相关性最高，相关系数达到0.90。以上结果表明：FORCCHN2模型可以提高光合作用的模拟能力，并能应用于北美森林生态系统GPP的评估中。

（3）森林覆盖数据能有效改善FORCCHN2对全球森林生态系统GPP的估算。首先，将FORCCHN2模型应用至全球森林生态系统中，利用其模拟的结果与基于通量观测外推的FLUXCOM全球数据集作为对比，两者空间分布显著相关，决定系数（R²）为0.81，相对均方根误差为0.37 kg C m^-2 yr^-1。相对而言，FORCCHN2模型在北半球模拟效果最佳，在南半球模拟效果略差。结果表明：FORCCHN2模型能再现全球森林生态系统GPP的空间变化特征。其次，将动态的森林覆盖率资料与FORCCHN2模拟的GPP相结合，结果显示：1985-2013年，森林全覆盖与计入覆盖率模拟的GPP总量之差为14.34±0.75 Pg C yr^-1，计入覆盖率结果的GPP增加趋势低于森林全覆盖结果的增加趋势。就常绿阔叶林（EBF）、落叶阔叶林（DBF）、混交林（MF）、常绿针叶林（ENF）及落叶针叶林（DNF）等五种森林类型而言，上述两者的GPP总量差值分别为5.48±0.20、1.25±0.08、2.12±0.17、4.00±0.23、1.50±0.13 Pg C yr^-1。

Given that GPP products by forest ecosystems account for about half of terrestrial ecosystems, forest ecosystems are the important parts of terrestrial carbon sequestration. A small change in forest photosynthesis could potentially mitigate anthropogenic carbon dioxide (CO₂） emissions. The terrestrial carbon cycle model based on the dynamic process of vegetation has always been considered as the essential tool for estimating and predicting forest GPP. Although the carbon models have widely used in forest ecosystems, there are still large uncertainties. Currently, accurate prediction and simulation of photosynthetic productivity from forests is critical for modeling carbon cycle.

Here, we developed, validated and applied a new leaf phenology model based on the concepts of “Thermal time” and “Cold Degree Day”. Then, we revised the forest carbon cycle model FORCCHN to the second version, FORCCHN2. This model couples the leaf phenology model with leaf growth process. At site-level, our goals are to apply FORCCHN2 model to simulate leaf phenology and GPP and to test the improvement of FORCCHN2 when against other dynamic carbon models. At region-level, we estimate the forest GPP at the scale of all of North America between 1985-2016 by using FORCCHN2 model. Moreover, we added the data of dynamic forest coverage into the GPP simulation and explored the impact of forest coverage on estimating GPP of global forest ecosystems. The main conclusions of this study are presented as follows:

1. The phenology model in FORCHCHN2 could reproduce the temporal and spatial changes of leaf phenology in the forest ecosystems of North America. At the forest observed sites of 114 phenological cameras, the FORCCHN2 model simulated onset time of leaf growth, cease time leaf growth, and length of leaf growing season with correlation of 0.89, 0.76, and 0.82, model efficiencies of 0.80, 0.58, and 0.68, and RMSE of 9.22, 17.63 and 24.63 days, respectively. The model can better in capturing the phenology time in deciduous forests than in evergreen forests. We applied this model to estimate leaf phenology in the North America. The results showed that the mean leaf unfolding time in the forest ecosystems of North America was DOY 125.1±3.8 during 1985-2016; mean leaf coloring time was DOY 252±24.5; and the mean length of growing season was DOY 126.4±25.8. Moreover, the leaf unfolding time in North America had advanced by 0.3 days per decade, the leaf coloring time had delayed by 3.7 days per decade, and the length of leaf growth season had increased by 4.0 days per decade. These change trends were comparable to previous studies. The results showed that the length of leaf growing season in forests was mainly affected by the delay of leaf coloring time in recent years. The delay of leaf coloring time was attributed to the significance increment of temperature in summer and autumn.

2. The FORCCHN2 could improve the predicted performance of photosynthesis related time, intra- and inter-annual GPP. The FORCCHN2 model demonstrated improvements compared to other six models (three prescribed models and three prognostic models）, photosynthesis onset time and cease time with up to 0.17-0.73, 0.03-0.54 increase in correlation and up to 0.39-2.57, 0.07-1.32 increase in model efficiency at 18 observed sites. For daily GPP simulations, the FORCHCH2 model showed the best performance among all of prognostic models, but its predicted performance showed slightly worse than the prescribed models. For inter-annual GPP simulations, the FORCCHN2 model showed the best performance among all of the models. We then applied the FORCCHN2 model in the forest ecosystems of North America. The results showed that 72.4% of forests (north region） had the increasing trend, while 27.6% of forests (south region） had the decreasing trend. For the whole North America, the mean annual GPP in the forests was 7.45±0.77 Pg C yr^-1, and the GPP has shown a significant increase trend under the background of climate change in recent decades. For the regional result inter-annual GPP, the FORCCHN2 model was significantly correlated to other carbon models which based on vegetation dynamics. CASA-GFEDv2, a prescribed model based on remote sensing, had the highest correlation with FORCCHN2 model (r=0.90）. The results indicate that the FORCCHN2 model can improve the predicted performance of photosynthesis and can be used in the assessment of forest GPP in North America.

3. Forest cover data could improve GPP estimate of FORCCHN2 in the global forest ecosystems. We applied this model in the global forest ecosystems. The simulated results of FORCCHN2 model were compared with the FLUXCOM global dataset based on carbon flux observation. The spatial distribution of the two results had a high correlation with R square of 0.81 and RMSE of 0.37 kg C m^-2 yr^-1. Relatively, the FORCCHN2 model had the best predicted performance in the northern hemisphere and slightly worse predicted performance in the southern hemisphere. The results showed that the FORCCHN2 model could reproduce the GPP spatial variation of the global forest ecosystems. Second, we added the data of dynamic forest coverage into the GPP simulation. The results showed that the GPP difference between the 100% forest coverage and the observed forest coverage was 14.34±0.75 Pg C yr^-1during 1985-2013. The increase trend of GPP with the observed coverage was less than that of the 100% forest coverage. For the five forest types such as EBF, DBF, MF, ENF and DNF, the GPP difference between the 100% forest coverage and the observed forest coverage was 5.48±0.20, 1.25±0.08, 2.12±0.17, 4.00±0.23, 1.50±0.13 Pg C yr^-1, respectively.