Bayesian Sensor Placement for Multi-source Localization of Pathogens in Wastewater Networks

Abstract

Wastewater monitoring is an effective approach for the early detection of viral and bacterial disease outbreaks. It has recently been used to identify the presence of individuals infected with COVID-19. To monitor large communities and accurately localize buildings with infected individuals with a limited number of sensors, one must carefully choose the sampling locations in wastewater networks. We also have to account for concentration requirements on the collected wastewater samples to ensure reliable virus presence test results. We model this as a sensor placement problem. Although sensor placement for source localization arises in numerous problems, most approaches use application-specific heuristics and fail to consider multiple source scenarios. To address these limitations, we develop a novel approach that combines Bayesian networks and discrete optimization to efficiently identify informative sensor placements and accurately localize virus sources. Our approach also takes into account concentration requirements on wastewater samples during optimization. Our simulation experiments demonstrate the quality of our sensor placements and the accuracy of our source localization approach. Furthermore, we show the robustness of our approach to discrepancies between the virus outbreak model and the actual outbreak rates.