A calibration capture–recapture model for inferring natural gas leak population characteristics using data from Google Street View cars

While new mobile monitoring technology has revolutionized our ability to measure pollutant levels over large regions, statistical methods for making inferences from data collected by these mobile systems are still being developed. We introduce a new capture–recapture model to answer key inferential questions from data collected by mobile monitoring systems. We apply our new method to characterize populations of natural gas (NG) leaks in urban areas using data collected by atmospheric methane analyzers placed on Google Street View cars. Leaks in urban NG distribution systems correspond to an economic loss, are a potential safety hazard, and are climate altering because NG is primarily composed of methane, a potent greenhouse gas. The new calibration capture–recapture (CCR) model combines data from controlled methane release experiments and data collected from mobile air monitors to enable inference for several NG leak population characteristics, including the number of undetected leaks and the total methane output rate in a surveyed region. Our methodology is a novel application of capture–recapture modeling. The CCR model addresses challenges associated with using a capture–recapture model to analyze data collected by a mobile monitoring system such as a variable sampling effort. We develop a Markov chain Monte Carlo algorithm for parameter estimation and apply the CCR model to data collected in two U.S. cities. The CCR model provides a new framework for inferring the total number of leaks in NG distribution systems and offers critical insights for informing intelligent infrastructure repair policy that is both cost effective and environmentally friendly.