Identifying the spatiotemporal dynamics and determinants of riverine DOC and pCO2 in the Yangtze River using machine learning methods

Rivers are the link among terrestrial, oceanic, and atmospheric carbon pools, with longitudinal transport and vertical emission processes serving as critical pathways for riverine carbon export from river ecosystems. However, the spatiotemporal dynamics and determinants of riverine carbon export processes in large river systems remain poorly understood due to limited quantification of riverine carbon fluxes. In this study, the spatiotemporal patterns and determinants of longitudinal transport and vertical emissions of riverine carbon (i.e., dissolved organic carbon concentration (C_DOC) and partial pressure of carbon dioxide (pCO₂), respectively) were revealed using machine learning methods in the world’s third-largest river (the Yangtze River). Long-term, monthly, and river-reach scale estimation of riverine C_DOC derived from Random Forest and Recursive Feature Elimination methods identified upstream hotspots of annual variation (approximately 23.6% of all basin reaches) and a spatial pattern of higher tributary concentrations, which corresponded to an annual DOC export of approximately 0.80 to 1.55 Tg to the ocean. Riverine pCO₂ was higher than the atmospheric level, exhibited an increasing trend from upstream to downstream, and showed monthly fluctuations that, as identified by the k-Shape clustering algorithm, gradually evolved from smooth (upstream) to bimodal mode (downstream). Both riverine C_DOC and pCO₂ exhibited a seasonal pattern with high values in summer and autumn, whereas a distinct springtime peak in pCO₂ was observed in the downstream. Climate and vegetation served as major determinants of the spatiotemporal patterns of riverine C_DOC and pCO₂. Precipitation, air temperature, and cumulative gross primary productivity exhibited significant and nonlinear increasing effects on riverine C_DOC, with their importance second only to elevation. Air temperature was the most important determinant for riverine pCO₂, with a relative contribution ranging from 17.8±1.3% to 40.0±0.9%. Vegetation factors exerted stronger influences on riverine pCO₂ with strong fluctuation than on pCO₂ with a smooth mode. This suggested that both longitudinal transport and vertical emission processes in the Yangtze River system would strongly respond to global warming, wetting and greening trends. Consequently, the DOC-enriched Yangtze River (with approximately 45.9% of river reaches being significantly transport-limited and only 0.6% being significantly source-limited) might export more DOC to the ocean, and the peak time of riverine CO₂ emissions might vary in the future. In summary, this study revealed the spatiotemporal dynamics and determinants of riverine carbon longitudinal transport and vertical emission processes in the Yangtze River, emphasizing that riverine carbon export processes need to be further concerned under the global change.