Machine learning reveals reservoir regulation of riverine sedimentary organic matter and its climatic implications.

Reservoirs play a pivotal role in riverine water resource management, while their influence on the cycling processes of riverine carbon remains to be further explored, constraining our comprehension of ecological impacts of reservoirs on terrestrial ecosystems. Here, we developed a machine learning (ML) model to characterize the dynamic of sedimentary organic matter (SOM) in the Yangtze River, one of the world's largest rivers under the influence of the Three Gorges Reservoir (TGR), one of the world's largest reservoirs. The TGR exhibited lower organic carbon and dissolved organic carbon concentrations than the midstream of the Yangtze River (MYR), with mean values of 0.90 ± 0.20% and 1.58 ± 0.37 mg L-1, respectively, compared with 1.34 ± 0.45% and 2.60 ± 0.58 mg L-1 in the MYR. These differences suggested enhanced microbial processing and carbon elimination within the reservoir. Our ML model further revealed that the TGR promoted the in situ burial of heteroatom-poor, low-aromaticity organic matter (OM), while facilitating the downstream transport and subsequent deposition of microbially reworked, highly recalcitrant OM in the MYR. Combining with greenhouse gas (GHG) emission data from TGR to MYR, we found that this dynamic of SOM is involved severely in the decline of CO2 (∼60%) and CH4 (∼24%) emissions in the middle and lower reaches of the Yangtze River. This study highlights the potential of ML in the assessment of ecological impacts of reservoirs, and suggests that reservoirs contribute to GHG mitigation in downstream river systems.