从陆地至河网并最终到达湖泊或海洋的溶解有机碳(Dissolved Organic Carbon, DOC)横向运输是全球碳循环的重要组成部分。DOC的输移代表了陆地生态系统中碳储量的再分配，这个过程受到多个因素的共同影响，这些扰动加速了陆地碳沿着陆地-内陆水连续体-海洋的周转。忽视这部分碳的横向流动会低估陆地生态系统碳汇，水生垂直和横向碳通量可能会有明显的偏差。河流DOC浓度有很强的空间变异性，然而很少有研究预测全球河流DOC浓度，仅有研究只预测流域或区域平均值，忽视了流域内不同河段的差异，这导致了特殊或热点河段的重要性被忽略，进而产生错估。因此，亟需对全球河段DOC浓度及河流DOC输移通量进行准确模拟。

本研究汇编了不同等级河流DOC浓度的数据集，筛选出12个主要控制因子，结合大数据分别构建传统统计模型与随机森林(Random Forest, RF)机器学习模型，在年尺度与月尺度分别对两种模型预测能力进行评估对比，选择最优模型预测每月290万个河段的DOC浓度，阐明河流浓度、流域产率以及通量的时空分布格局，量化特殊类型流域的通量及对大陆与近海碳预算的贡献，并探索其控制机制。主要研究结论如下：

1. 影响河流DOC浓度最直接的控制因子不是土壤有机碳储量的大小，而是输移过程的地理地形特征。其中，坡度的解释度最高，平缓的坡度有利于延长水的停留时间，促进有机质浸出。其次为地下水水位，高的地下水水位与浅层腐殖质层相交，而地下水水位低时流经矿物层。此外，高海拔河流DOC浓度比平原河流低，大河平均浓度比小河低。

2. 利用RStudio4.0.3程序分别构建逐步回归模型与随机森林模型。年尺度的对比结果显示逐步回归模型的决定系数(R²)为0.25，模型残差为0.31；RF模型的R²为0.71，模型残差为0.21。因此，RF模型具有更高的R²与更低的模型残差，性能更优。月尺度上逐步回归模型各月R²范围为0-0.62，均方根误差(Root Mean Square Error, RMSE)范围为2.5-11.2；RF模型R²范围为0.64-0.84，RMSE范围为1-10。在各个月份中RF具有更高的R²与更低的RMSE，更为合理地再现了DOC的季节性。因此选择RF模型进行后续预测。

3. 由RF模型预测的全球河流DOC平均浓度为5.1±2.5 mg L^-1，北极河流DOC浓度最高，北温带最低，高DOC浓度主要分布在亚马逊流域、北欧、西西伯利亚和北美高纬度地区。流域DOC产率变化范围为0.003-22.61 g C/m²/yr，平均为1.77±2.62 g C/m²/yr。产率最高的是10° S-10° N热带流域，高产率从流域输出表明更多的有机质从陆地流失。世界内流河输出的DOC总量为10.15 Tg C yr^-1，向海洋输出的DOC总量为241.96 ± 61 Tg C yr^-1，DOC入海通量约占大陆边缘净初级生产力(Net Primary Productivity, NPP)的2.69%。向近海输出最多的流域是亚马逊河和刚果河，分别占全球河网DOC输出总量的20.0%和7.7%，显示了大河在向海洋输出DOC过程中发挥主导作用。

4. DOC产率占陆地NPP的比例从可忽略不计(<0.01%)到高达26.88%。平均而言，约0.48%±1.27%的陆地净初级生产力以DOC的形式输送到河流。具有较高输出百分比的区域主要为在北部泥炭地盆地和热带潮湿地区，在这些地区，碳储量大或/和河流网络流量大。从空间上看，全球流域表现出运输限制而非来源限制，即区域河网输出的DOC随着降水量或流量的增加而增加。流域水文过程主导了DOC的空间分布，陆源有机质输出在潮湿地区高，在干燥地区低。

5. 极地、北温带、热带、南温带河流DOC浓度月平均变异系数分别为9.6%、7.6%、7.6%、5.2%。比起其他气候带，极地河流具有更大季节变化，5-7月DOC浓度迅速升高，其DOC输出对全球通量具有短暂但强烈的影响。热带贡献了全球河流DOC通量的一半以上，峰值接近14.6 Tg C/月。高流量和丰富的土壤有机碳储量是主要原因。本研究发现出口量最高的月份是出口量最低月份的1.45倍。

本研究强调了流域水文循环在驱动陆地生产力运输至内陆水域和海洋的重要性。这项研究对于了解并优化全球碳循环、评估人类活动对碳平衡的影响以及应对气候变化具有重要意义。通过深入研究DOC从陆地-河流-海洋的运输，可以更好地了解碳的迁移和转化。

The lateral transport of dissolved organic carbon (DOC) from land to river networks and ultimately to lakes or oceans is an important component of the global carbon cycle. The transport of DOC represents the redistribution of carbon storage in terrestrial ecosystems, which is influenced by multiple factors. These disturbances accelerate the turnover of terrestrial carbon along the land inland water continuum ocean. Neglecting the lateral flow of this part of carbon will underestimate the carbon sink of terrestrial ecosystems, and there may be significant deviations in the vertical and horizontal carbon fluxes of aquatic systems. The concentration of riverine DOC has strong spatial variability, but few studies have predicted the global DOC concentration in rivers. Only studies have predicted the average value of the basin or region, ignoring the differences in different river sections within the basin. This has led to the neglect of the importance of special or hot river sections, resulting in misestimation. Therefore, it is urgent to accurately simulate the global DOC concentration in river sections and the transport flux of DOC in rivers.

This study compiled data sets of DOC concentration in rivers of different levels, screened out 12 main control factors, combined with big data to build traditional statistical model and Random Forest (RF) machine learning model, and evaluated and compared the prediction ability of the two models at annual and monthly scales. The optimal model is selected to predict DOC concentrations in 2.9 million river sections per month, elucidates the spatial and temporal distribution patterns of river concentrations, basin yields, and fluxes, quantifies fluxes in special types of basins and their contributions to continental and offshore carbon budgets, and explores their control mechanisms. The main conclusions are as follows:

1. The most direct control factor affecting the concentration of riverine DOC is not the size of soil organic carbon storage, but the geographical and topographic characteristics of the transport process. Among them, the slope has the highest interpretation, and the gentle slope is conducive to extending the residence time of water and promoting the leaching of organic matter. The second is the groundwater level, the high groundwater level intersects with the shallow humus layer, while the low groundwater level flows through the mineral layer. In addition, the DOC concentration in high-altitude rivers is lower than that in plain rivers, and the average concentration in large rivers is lower than that in small rivers.

2. Use RStudio4.0.3 program to build stepwise regression model and random forest model respectively. The annual comparison results show that the coefficient of determination (R²) of the stepwise regression model is 0.25, and the residual is 0.31. The R² of the RF model is 0.71, and the model residual is 0.21. Therefore, the RF model has higher R² and lower model residuals, and the performance is better. On the monthly scale, the monthly R² range of stepwise regression model was 0-0.62, and the Root Mean Square Error (RMSE) range was 2.5-11.2. The RF model R² ranges from 0.64 to 0.84 and the RMSE ranges from 1 to 10. The RF model has higher R² and lower RMSE in each month, which more reasonably reproduces the seasonality of DOC. Therefore, RF model was selected for subsequent prediction.

3. The average concentration of DOC in global rivers predicted by RF model is 5.1±2.5 mg L^-1, with the highest DOC concentration in Arctic rivers and the lowest in the Northern temperate zone, and the high DOC concentration is mainly distributed in the Amazon basin, Northern Europe, Western Siberia and the high latitudes of North America. The DOC yield in the basin varied from 0.003 to 22.61 g C/m²/yr, with an average of 1.77±2.62 g C/m²/yr. The highest yields are found in the 10° S-10° N tropical basin, where the output of high yields from the basin indicates more organic matter loss from the land. The total DOC output of the world's internal rivers is 10.15 Tg C yr^-1, and the total DOC output to the ocean is 241.96 ± 61 Tg C yr^-1. The DOC flux to the sea accounts for about 2.69% of the Net Primary Productivity (NPP) of the continental margin. The Amazon River and the Congo River accounted for 20.0% and 7.7% of the total DOC output of the global river network, respectively, indicating that large rivers play a leading role in the process of DOC output to the sea.

4. The proportion of DOC yield to NPP on land ranges from negligible (<0.01%) to as high as 26.88%. On average, about 0.48%±1.27% of net terrestrial primary productivity is transported to rivers in the form of DOC. The regions with a higher percentage of output are mainly in northern peatland basins and tropical humid regions, where carbon stocks are large or/and river network flows are large. Spatially, global watersheds exhibit transport constraints rather than source constraints, i.e. DOC output from regional river networks increases as precipitation or discharge increases. The spatial distribution of DOC is dominated by hydrological processes in the basin, and the output of terrigenous organic matter is high in wet areas and low in dry areas.

5. The average monthly coefficients of variation of DOC concentration in polar, northern temperate, tropical and southern temperate rivers were 9.6%, 7.6%, 7.6% and 5.2%, respectively. Polar rivers have greater seasonal variability than other climatic zones, and DOC concentrations increase rapidly from May to July. Its DOC output has a brief but strong effect on global flux. The tropics contribute more than half of the global river DOC flux, peaking at nearly 14.6 Tg C/month. High discharge and abundant soil organic carbon storage are the main reasons. The study found that the month with the highest export volume was 1.45 times that of the month with the lowest export volume.

This study emphasizes the importance of watershed hydrological cycles in driving land productivity transport to inland waters and oceans. This study is of great significance for understanding and optimizing the global carbon cycle, assessing the impact of human activities on carbon balance, and addressing climate change. By conducting in-depth research on the transportation of DOC from land to river to ocean, we can better understand the migration and transformation of carbon.