A Comparison of Groundwater Fluxes Computed with MODFLOW and a Stable Isotope Mixing Model: Application to the Eastern Nevada Test Site and Vicinity

The primary objective of this study was to compare groundwater flows in the vicinity of the eastern Nevada Test Site (NTS) computed with a hydraulically defined groundwater flow model against those predicted with an isotopically based mixing model. The US Geological Survey (USGS) Death Valley Regional Flow Model (DVRFM) is a transient, three-dimensional, groundwater model that uses the public domain, finite difference code MODFLOW. The second model (Discrete-State Compartment Model-Shuffled Complex Evolution; DSCM-SCE) is a recently developed code that auto-calibrates groundwater flows (both magnitude and direction) using a steady- state mixing algorithm to best match observed conservative tracer concentrations in groundwater. The model can be calibrated based on data for multiple tracers with tracer concentrations considered either independent or fully covariant. To compare modeling approaches, DVRFM boundary conditions and cell-to-cell interactions were used in the DSCM-SCE's 15-cell eastern NTS model. Analysis of δD and δ18O data conducted throughout the model domain suggests recharge and mixing may be the dominant mechanisms for groundwater isotopic enrichment in the down gradient direction. Model results show that DVRFM boundary fluxes and cell outflows match representative groundwater isotopic values relatively well. However, optimization on data for an individual isotope lowered the objective function, while combining δD and δ18O independently produced the least error. Large uncertainty in the estimated covariant relationships between δD and δ18O prohibited development of a unique DSCM-SCE solution. Results suggest error exists in estimated DVRFM boundary conditions associated with three of the fifteen modeled basins, while the lack of isotopic data in several basins defining the eastern edge of the model prevents certitude of results pertaining to fluxes in this portion of the model. Future work will look at the uncertainty associated with boundary conditions and what data will better constrain the model for improved fit.