Fugitive emissions, such as leaking components, are important sources of methane and other pollutants within the oil and gas sector. These gas emission sources are difficult to mitigate because their existence, location, and magnitude are unknown, especially within facilities which may have thousands of potential source components. Improving the speed at which these sources are characterized (located and quantified) can improve these mitigation efforts.

Using sparse concentration information (from finite sensor locations) and estimated wind fields, a scalar transport adjoint-based optimization approach was developed to characterize both single and multiple simultaneous gas releases representative of the fugitive emissions problem. This approach was tested using data from an open-field gas release to determine a single source location within 5 m and source magnitude within 13%. Simulated simultaneous releases over a complex 3D geometry based on an Alberta gas plant were also characterized using both detailed transient wind and wind fields approximated with a series of steady-state wind simulations more readily implemented in a field application. Magnitudes were predicted within 10% and major regions located.

By extending the method with a database of pre-computed retro-tracers generated on simplified steady-state wind fields, the required computational time for the solution optimization was reduced by a factor of 200-600, making computations feasible on a desktop machine and raising the possibility of near-continuous fugitive emissions quantification in the future. Major sources were successfully located, and magnitudes estimated within -75 to -32%, even with limited wind coverage (60° direction variation).

Turbulent Schmidt number selection (which scales diffusivity) had little effect on estimated source locations, but strongly influenced estimated magnitudes. For a tested turbulent Schmidt number range of 0.33 (diffuse) to 2.0 (highly advective), the predicted emission rates for the open field release varied between -35 to +128% of actual. Buildings dampened this effect, suggesting that open-field estimates can act as an error bound.