Global methane emissions from rivers and streams

Freshwater ecosystems are responsible for nearly half of global CH⁴ emissions to the atmosphere^4.7. Yet, among freshwaters, the role of rivers and streams in the global CH4 cycle remains unclear although
current best estimates of global fluvial emissions^3.4are similar in magni-tude to other important CH₄ sources such as biomass burning and ricecultivation⁸. Fluvial ecosystems play key parts in connecting terrestrial,marine and atmospheric carbon pools⁸, and are unique in their poten-tialto produce CH₄ internally, while also receiving and emitting largeamounts of CH₄ generated externally in adjacent soilsand wetlands^9.10.
Thus, global CH₄ emissions from streamsandrivers maybe regulated by multiple environmental factors that operate across land–waterboundaries. Resolving these controls should improve our predictions
of riverine CH4 emissions and our broader understanding of how run-ning waters process and deliver carbon to downstream ecosystems inresponse to climate warming and other global environmental changes. Despite their potential as an important atmospheric source, cur-rent syntheses of riverine CH₄ emissions highlight extreme spatial and temporal variability, with measured rates spanning seven orders of magnitude^3.4, as well as strong fine-scale controls over CH₄ dynamics^10.11 .
Thus, efforts to generate global estimates have been basedona simpleaveraging of measured CH₄ emissions, which has resulted in massiveuncertainty^3.4.7.12, unknown global patterns³ and large discrepanciesbetween bottom-up inventories and top-down estimates^4.7. Furthercomplications arise from the fact that aquatic CH₄ emissions occur by diffusion and by the even-more variable process of ebullition, in which CH₄-rich bubbles are released from sediments. To address these uncer-taintiesand advance our understanding of CH₄ dynamics in runningwaters, we leveraged aCH₄ database¹³(Global River Methane database(GRiMeDB)) containing more than 24,000 observations of CH₄ con-centration and more than 8,000 observations of CH₄ fluxes (ExtendedData Fig. 1) to model CH₄ concentrations globally using random forestmachine-learning models. From these models, we can explain a sub-
stantial fraction of the total variability inCH₄ concentrations (R² fromlog-transformed modelled versus withheld observations of 0.45–0.68;Extended Data Fig. 2) and produce a seasonally and spatially explicit
global estimate of CH₄ emissions from rivers and streams. More impor-tantly, using this database and model outputs, we are able to identifythe main drivers of CH₄ concentrationsand fluxes from running watersacross the globe.