An automated approach to screening water level observations from driller's logs is presented for calibrating regional groundwater models. Standard methods for model calibration were not converging on a solution using a large water level data set. Many of the observations cannot be matched for various reasons. Manual screening was not practical so the following approach was implemented.
Each observation is considered every time a new parameter value is tested. If the difference between observed and modeled head exceeds a specified value, the data point is temporarily excluded from the calibration by artificially setting this difference to zero. In this way, large residuals do not influence the parameter upgrades calculated using the non-linear least squares method. This pragmatic approach boils down to “fitting what you can and ignoring the noise”.
Two criteria were added to the objective function of the inverse model: 1) an observation that applies a penalty for not fitting all of the head observations, and 2) an observation that applies a penalty for not fitting heads across the observed range of values in the model domain. Methods for balancing the weights between the observations and additional criteria are discussed.