Large Scale Parameter Estimation through the Inverse Procedure and Uncertainty Propagation in the Columbus Basin, New Mexico

Estimates of the aquifer parameters storativity, transmissivity, boundary heads and fluxes were made for the Columbus Basin, New Mexico. A two-dimensional groundwater flow code called CERT was used. CERT employs a non-linear unconstrained generalized weighted least squares minimization algorithm to obtain optimal aquifer parameters from historic observations of hydraulic head. An initial or prior estimate of the aquifer parameters was obtained using kriging and other geostatistical tools. The record of historic observed head data from 1930 to 1975 was used for the least squares fit. Optimal parameter estimates for transmissivity, storativity and fluxes were reasonable. Nodal and zoned values of the parameters were estimated. The algorithm had difficulty estimating realistic boundary heads for conceptualization and data for the Columbus Basin.

A validation study of CERT was conducted by estimating transmissivity and/or nodal fluxes using portions of the observed head data set divided in space or time. A posterior linearized estimate of the parameter covariance matrices was also conducted. These results were used as input to the uncertainty propagation algorithm coded into CERT, and predictions of heads and head standard deviations were made over the time period 1930 to 1975. Model heads matched well with observed heads when the model input parameters were conditioned only on a portion of the observed head data. Values of the standard deviation on heads were reasonable once boundary node correlation was accounted for.