Machine learning deciphers CO2 sequestration and subsurface flowpaths from stream chemistry

Abstract

Abstract. Endmember mixing analysis (EMMA) is often used by hydrogeochemists\nto interpret the sources of stream solutes, but variations in stream\nconcentrations and discharges remain difficult to explain. We discovered\nthat machine learning can be used to highlight patterns in stream chemistry\nthat reveal information about sources of solutes and subsurface groundwater\nflowpaths. The investigation has implications, in turn, for the balance of\nCO 2 in the atmosphere. For example, CO 2 -driven weathering of\nsilicate minerals removes carbon from the atmosphere over ∼ 10 6 -year timescales. Weathering of another common mineral, pyrite, releases sulfuric\nacid that in turn causes dissolution of carbonates. In that process,\nhowever, CO 2 is released instead of sequestered from the atmosphere. Thus, understanding long-term global CO 2 sequestration by weathering\nrequires quantification of CO 2 - versus H 2 SO 4 -driven\nreactions. Most researchers estimate such weathering fluxes from stream\nchemistry, but interpreting the reactant minerals and acids dissolved in streams has been fraught with difficulty. We apply a machine-learning\ntechnique to EMMA in three watersheds to determine the extent of mineral\ndissolution by each acid, without pre-defining the endmembers. The results\nshow that the watersheds continuously or intermittently sequester CO 2 , but the extent of CO 2 drawdown is diminished in areas heavily affected\nby acid rain. Prior to applying the new algorithm, CO 2 drawdown was\noverestimated. The new technique, which elucidates the importance of\ndifferent subsurface flowpaths and long-timescale changes in the watersheds,\nshould have utility as a new EMMA for investigating water resources\nworldwide.