Using Sensitivity Analysis as a Tool to Determine the Need for Regeneration of Hydrological and Biogeochemical Predictions

Hydrological and biogeochemical predictions, such as predictions of dynamically-evolving groundwater plumes or carbon/nutrient fluxes in heterogeneous aquifers, require periodic improvements of conceptual and numerical models, as new field data or model parameters become available. The ingestion of new data in numerical simulations may help with reducing the uncertainty and improving the accuracy of predictions. In order to assure robustness of model results and obtain the desired accuracy, one needs to evaluate whether addition of new input data would improve the outcome of model predictions, and what minimum observations are needed to parameterize the model to capture the entire (relevant) range of hydrological and biogeochemical conditions. In this study, we quantified the sensitivity of a 2-D high-resolution transect model developed for the Rifle floodplain site in Colorado, as criteria for regeneration of hydrological and biogeochemical predictions. To determine when and where data regeneration is required, we used the 2-D transect model developed in the numerical code ToughReact, and tested it for different data addition/update scenarios. Our results indicate that adding certain microbial pathways, such as chemolithoautotrophic reactions, and temperature variations significantly impacted predictions of carbon dynamics. Similarly, we determined that the model predictions were highly sensitive to certain model parameters, such as porosity, hydraulic conductivity of the hydrostratigraphic units, oxygen and nitrate content, and inhibition constants. We also determined that the sensitivity of model predictions to model parameters was dependent on variations in geomorphological conditions. In particular, the sensitivity of predictions was higher at upslope regions, where recharge boundary conditions are applied, compared to the downslope area. The application of the sensitivity analysis is recommended to determine whether and when regeneration of model predictions is required.