2.2 Modeling hyporheic respiration
We used previously published modeled estimates for aerobic respiration in the hyporheic zone at the scale of NHDPLUS reaches within the two study basins using the River Corridor Model (RCM) presented in (Son et al. 2022b). The model computes at hourly timesteps, using reactions, rate formulations and kinetic rate parameters described in (Fang et al. 2020). Key model inputs for each reach include hyporheic exchange fluxes (HEF) and hyporheic residence times estimated by the Networks with Exchange and Subsurface Storage (NEXSS) model (Gomez-Velez and Harvey 2014), and long-term (~10 year) average stream solute concentrations (dissolved organic carbon, dissolved oxygen, and nitrate) estimated from available observations. We gap-filled missing HEF and residence time values for ~13% of reaches, using a eXtreme Gradient Boosting (XGBoost, v1.5.0.2) machine learning model (Chen and He 2023) ingesting topography variables (watershed/catchment area, maximum/mean/minimum elevation), hydrology (mean annual streamflow/runoff/velocity and mean annual precipitation), and stream geometry (slope, width and length, sinuosity, bankfull depth, median particle size, roughness, total stream length) with high accuracy (R2>0.94). Detailed descriptions of the stream solute input data and our XGBoost approach are included in Supplemental Information.