Point Sensor Network Detects Short Releases Under Favorable Wind Conditions

In this study, we apply a recursive Bayesian state updater algorithm to assimilate meteorological data with point-sensor measurements of methane concentrations to infer timeseries of methane emission rates at three operating oil and gas facilities. These calculations are performed over a timeframe with known numerous short (~ 30 minute) controlled releases, allowing for "ground truth'' data to compare our emission estimates against. The highly-varying and unknown operational emissions pose challenges in analyzing quantification results when trying to determine whether there is evidence in the emission estimates of a given controlled release. Ultimately, we find that despite the non-ideal conditions at these sites (poor sensor placement and the presence of large obstructions that the quantification model does not account for) that the site-level emission estimates show evidence for 31 out of the 60 controlled releases and that the majority of nondetections were due to the wind simply not pointing from the source to any sensor in the network during a short release event.