Evaluation of the Effect of Initial Conditions on Parameter Estimation in a Groundwater Flow and Transport Model

Abstract

Parameter estimation techniques are effective at efficiently adjusting model parameters (e.g. hydraulic conductivity) to achieve reductions in error between observed conditions (e.g. water levels) and corresponding model-simulated values. These errors are typically aggregated using statistics such as sum of squared errors (SSE) to quantify overall model error. Calibration is iterative, non-linear, and can be sensitive to the values assigned to model parameter prior to calibration, such that the characteristics of the calibrated model can depend on those of the pre-calibration model. In order to illustrate this, seven focused calibrations of a groundwater flow and transport model were conducted, each with a unique set of initial parameter values, resulting in seven unique sets of parameter values. Additionally, though the model SSE was relatively invariant across most of the resulting models, further inspection revealed substantial differences in model quality across the calibrated models. Results of the analysis suggest that the derivation of initial model parameter values is as important as the calibration process, itself. So, too, is the inspection of model errors associated with the calibrated model in greater detail than is provided by model-wide statistics such as SSE.