Validation of aquifer parameter determination by extrapolation fitting and treating thickness as an unknown

Abstract A genetic algorithm is used here to guess-estimate a close-to-true set of trial values as input to a three-staged quasi-linear inverse modeling scheme for the determination of aquifer parameters. To validate the parameter determination, in addition to the conventional measures of misfit root mean squares (rms) and distribution, the aquifer thickness is treated as an unknown parameter and the model parameters are further evaluated by comparing the expected drawdown with the observed drawdown at wells which are not used for parameter determination (extrapolation fitting). The method is tested with synthetic and observed drawdown data from five partially screened monitoring wells in a water-table aquifer. Test results for synthetic data doped with random errors indicate that modeling based on two or more well data can yield satisfactory parameter values and extrapolation misfits in an ideal aquifer. For field data, the results indicate that a model misfit on par with the standard error of the data is achievable for each individual well or a combination of two wells but the extrapolation misfit distributions are generally biased and their rms are far greater—possibly due to aquifer heterogeneity. Consistent parameter values can be obtained from the geometric means for multiple runs of the genetic-inverse modeling of one-, two-, three-, and four-well data. Our test aquifer can be represented by a set of parameters with 10 to 15% consistency, including transmissivity, storativity, vertical-to-horizontal conductivity ratio, and storativity-to-specific yield ratio, as affirmed by model aquifer thicknesses that deviate less than 10% from the actual thickness.