Abstract
<jats:p> Rivers transfer organic carbon from land to inland waters, but sparse and uneven observations have limited our ability to determine where, when, and how this lateral pathway matters for terrestrial carbon accounting. We integrated more than 95,000 measurements of total and dissolved organic carbon using a multitask deep learning model that leverages complementary carbon observations with broader spatial coverage to inform the reconstruction of sparsely observed total organic carbon. This produced an observation-constrained reconstruction of daily riverine total organic carbon concentration and transport at 0.25° resolution across the contiguous United States from 1984 to 2023, resolving where lateral carbon transport arises across the landscape. Annual transport averaged 13.8 ± 1.9 Tg C yr <jats:sup>–1</jats:sup> . Relative to the absolute magnitude of vertical net ecosystem carbon exchange, riverine transport averaged 1.9 ± 0.3% across the continent, exceeded 10% in parts of the lower Mississippi Basin and southeastern Coastal Plains, and was most important near the transition between water-limited and wetter climates. Wet years expanded the area where riverine transport was quantitatively important, while the largest year-to-year increases occurred when wet conditions followed dry periods. It arose mainly from renewed water throughput, reinforced by higher carbon concentration, and was strongest where carbon-source and hydrological-mobilization influences were both strong. Riverine organic carbon transport is therefore a spatially structured, episodic, and history-dependent component of terrestrial carbon accounting. As hydroclimatic variability intensifies, representing this lateral pathway dynamically will become increasingly important for carbon budgets and Earth system models. </jats:p>