Optimal placement of sampling locations for identification of a two-dimensional space

ABSTRACT

In recent years, substantial attention has been given to the optimal placement and planning of sampling or monitoring based on the Value of Information (VoI). The drawback of the VoI approach is its typically high computation cost. The method proposed here can significantly reduce the computation cost of VoI by applying the theory of Gaussian Process Regression. Due to the low computation cost, global optimisation algorithms such as particle swarm optimisation can be used to optimise a set of sampling locations. The proposed method was applied to a soil contamination problem. For comparison purposes, twenty-one subjects (persons) were asked to follow a three-step procedure. In the first step, contamination levels at five fixed sampling locations are given. Based on this information, the contaminated area is estimated by the subject, who is then asked to place five additional sampling sites in order to gather more information. The procedure is repeated three times. The subject is ultimately asked to produce a final estimate of the contaminated area. It was found that the proposed method generally produced better agreement with the true contaminated area than did the participants’ empirical human judgment.

KEYWORDS:

• Gaussian random field
• Gaussian process regression
• optimal placement
• value of information

Disclosure statement

No potential conflict of interest was reported by the author(s).