Multivariate Geostatistical Design of Ground‐Water Monitoring Networks

Abstract

A multivariate approach for the design of ground‐water monitoring networks is presented. The proposed technique is based on the geostatistical method of cokriging. The network design problem is posed as an optimization model in which the variance of estimation is minimized. The multivariate approach used herein accounts not only for autocorrelations but also for variable cross correlations. By considering a network design in which several variables are jointly monitored, lower monitoring costs are achieved; also, and more importantly, a better estimation of the monitored parameters is yielded given the fact that lower estimation variances are reached. The method used for selecting the optimal monitoring sites is based on a simultaneous search technique using a branch and bound algorithm guaranteeing optimality of the solution. The methodology is applied for the design of a monitoring network to observe aquifer transmissivity (T) and the specific capacity (SC) in the Yolo County Basin, Calif. The results show the superiority of the multivariate geostatistical approach and how it provides a better and more economical network design than the univariate approach.