Application of an entropy-based Bayesian optimization technique to the redesign of an existing monitoring network for single air pollutants.

We apply the entropy-based Bayesian optimizing approach of Le and Zidek to the spatial redesign of the extensive air pollution monitoring network operated by Metro Vancouver, in the Lower Fraser Valley, British Columbia. This method is chosen because of its statistical sophistication, relative to other possible approaches, and because of the very rich, two-decade long data record available from this network. The redesign analysis is applied to ozone, carbon monoxide and PM(2.5) pollutants. The analysis provides guidance with regard to stations monitoring the three pollutants. For both ozone and PM(2.5), the analysis indicates a need for more stations in the eastern part of the monitoring domain. A parallel analysis indicates that stations may be removed from the more central parts of the domain. An analysis of the carbon monoxide network produces results that are not nearly as clearly defined as those for the other two pollutants, presumably because carbon monoxide is a primary pollutant with many locally important sources. The work demonstrates the great utility of the analysis technique, and also provides statistically defensible guidance on the spatial redesign of this important monitoring network.