Reactive-Transport Prediction Uncertainty and Simulation Accuracy, Observation Errors, and Sensitivity Analysis

1. Introduction Prediction uncertainty is the likely discrepancy between model predictions and the actual, unrealized system responses. Contributions to uncertainty include solution error and limited capabilities of numerical models, error and deficiency of data, and errors in system conceptual models. Uncertainty can be reduced by improving numerical models and using numerical models, data, and conceptual models together. For example, using conceptual models to build simulations forces ideas about system behavior that are often vague and possibly wrong to be clarified and tested thoroughly against data. Problems with the numerical methods or constitutive relations, however, can obscure results. This article considers three issues important to this process: (1) a common numerical-methods issue for groundwater transport simulations; (2) the problem of matching data too closely; and (3) using a model to evaluate the importance of observations to parameters, parameters to predictions, and observations to predictions.