Identifying the determinants of the spatial patterns and temporal fluctuation characteristics of riverine pCO2 of the largest subtropical river using machine learning methods

the Yangtze River Basin (YRB), China Carbon dioxide (CO 2 ) exchange at the riverine water–air interface is critical for global carbon cycle and is controlled by the partial pressure of CO 2 ( p CO 2 ). However, the spatiotemporal characteristics of riverine p CO 2 and quantitative contributions of its determinants at large scales are largely unexplored. This study identified the spatial and temporal characteristics of riverine p CO 2 in the YRB, and quantified the relative contributions of determinants using a machine learning method (boost regression tree, BRT). This study highlights the differences in determinants of the spatial and temporal characteristics of riverine p CO 2 , which are important for understanding the riverine carbon cycle. Riverine p CO 2 showed an increasing trend from upstream to downstream and its monthly fluctuations gradually evolved from smooth (upstream) to bimodal mode (downstream), which were identified by the k -Shape clustering algorithm. BRT model with nine determinants of climate, vegetation, and water quality effectively simulated the spatiotemporal characteristics of riverine p CO 2 . Climate and vegetation factors dominated the spatial distribution of yearly riverine p CO 2 , with total relative contributions exceeding 92.3 ± 3.3 %. Determinants varied with different fluctuation modes, whereas climate factors still played a major role. As the monthly fluctuations evolved from smooth to bimodal mode, vegetation and water quality factors became more important, with total relative contribution increasing from 32.8 ± 0.7 % to 53.1 ± 4.6 %. • Monthly fluctuation pattern of riverine p CO 2 showed spatial variation in the YRB. • Determinants of riverine p CO 2 were identified using machine learning methods. • Determinants and their contributions of riverine p CO 2 varied spatiotemporally. • Climate factors dominated both spatial and temporal variations of riverine p CO 2 . • Vegetation and water quality had highest contributions in downstream bimodal type.