Development and comparison of different methods of monitoring network design based on co-kriging

This study develops and compares several network design methods for groundwater monitoring based on the geostatistics theory and optimization schemes. Depending on the optimization structure, those methods are classified as sequential design model, branch and bound model, and non-linear programming model. The sequential design model extends the variance reduction method proposed by Rouhani in 1985. Theoretically, it is not a truly optimal design scheme. On the other hand, the branch and bound model, as proposed by Ben-Jemaa et al. in 1994 can compute a global optimal solution; however, it requires much more computation than the sequential design model. The non-linear programming method, extending the methods proposed by Lin in 1995, can search the optimal locations of the monitoring wells within the planning area without pre-defining a set of available sites, in addition, the solution is local optimal. In this study, we applied those methods to the network planning in the Ping-Tung Plain of southern Taiwan. According to our results, the differences of the solutions for different models decrease with an increase of all the monitoring wells. Owing to its efficiency and flexibility, the sequential design model is the most practical method in general application, although the solution is not necessarily optimal.