Optimizing LTM Networks: A Comparison of Approaches at Three Sites (Extended Abstract)

Abstract

The Air Force spends millions of dollars annually on groundwater Long-Term Monitoring (LTM) networks at installations around the country. Because LTM is costly and its well networks so extensive, the Air Force is actively pursuing testing and implementation of optimization strategies for its LTM networks at a number of sites. One such strategy developed in coordination with AFCEE is a decision-logic statistical optimization scheme named the Geostatistical Temporal-Spatial algorithm or GTS. GTS uses known statistical and geostatistical techniques in a novel manner to answer two questions: given an existing LTM network, 1) what is the optimum number and placement of wells in that network (i.e., is there spatial redundancy and/or is there `undercoverage' within the spatial network)?, and 2) what is the optimal sampling frequency for wells in the network (i.e., is there temporal redundancy)? Optimized networks in typical applications of GTS have resulted in an estimated cost savings off the total LTM budget of over 30%. To measure spatial redundancy, GTS has heretofore applied kriging and robust spatial modeling techniques in an iterative fashion in order to identify optimal subsets of the existing monitoring network. Optimality is measured by balancing the costs of sampling, analyzing, and maintaining the `optimal' network against 1) deterioration in estimated site maps compared to the baseline, 2) increases in global uncertainty associated with map estimates, and 3) increases in localized areas of uncertainty. To maximize the algorithm's flexibility, numerous changes and improvements to GTS have been tested. Some of these changes have been applied at LTM networks on three different Air Force bases in California, New Hampshire, and Maine. More specifically, comparisons have been made between the iterative mapping approach — where increasingly optimal well location subsets are identified via changes in the global estimation weights — and a newer approach based on the use of genetic algorithms. The specific genetic algorithms developed for GTS are designed to perform the overall spatial optimization more efficiently and in fewer analysis steps.