Keira Johnson Kathi Jo Jankowski Joanna C. Carey Lienne R. Sethna Nicholas Lyon Pamela L. Sullivan Sidney A. Bush 2026 <p><span>Riverine dissolved silicon (DSi) dynamics reflect integrated geologic, hydrologic, climatic, and ecological controls. We compiled annual DSi data for 337 rivers across four continents and trained interpretable machine-learning models to predict concentrations and yields from 28 watershed variables. Both models reproduced testing data (</span><i>R</i>²<span>&nbsp;=&nbsp;0.85 for concentration and 0.96 for yield) and withheld-site validation (</span><i>R</i>²<span>&nbsp;=&nbsp;0.91 and 0.93). Lithology, especially volcanic rock fraction, strongly controlled DSi while subsurface storage, topography, and land cover further shaped DSi dynamics. DSi concentrations and yields exhibited nonlinear responses to basin slope, recession-curve slope, proportion of open-water cover, and nutrient availability. Concentrations showed sharper threshold responses to hydrologic and biotic variables, whereas yields varied more gradually with climate and lithology. These results provide a framework for forecasting DSi under land cover and climate change and for embedding realistic, nonlinear processes in mechanistic models.</span></p> application/pdf 10.1029/2025GL118853 en American Geophysical Union Thinking outside the rocks: Subsurface water storage, topography, and land cover are key modulators of large-scale riverine dissolved silicon dynamics article