Hyperspectral satellite data is routinely used to monitor greenhouse gas concentrations worldwide. The standard processing of taking averages of these images over time produces rough estimates of emission rates, but atmospheric transport of the gases prevents this simple processing from pinpointing sources of emissions. However, very recent advances in the use of Optimal Transport have shown that using Wasserstein barycenters coupled with weather data can better concentrate mass around significant sources. Obtaining gas concentrations in the atmosphere from hyperspectral images is a low-cost method requiring no additional deployment of ground sensors. Moreover, remote sensing can cover large areas around natural gas storage facilities out of reach from ground sensors. Because light gases such as methane rapidly migrate from the leak area, the newly developed ability to invert measured gas concentrations back to their source of emission opens up an effective and dramatically cheaper Underground Gas Storage monitoring tool. While it is standard to measure gas concentration from hyperspectral images, identifying the gas source requires assimilating weather data to understand the gas migration. Because they solve the NP-hard problem of the Earth Movers Distance, recent demonstrations of the use of inverse transport theory for gas migration (shown in Fig. 1) do not scale. However, the recent discovery of the Sinkhorn distance as an approximation of Optimal Transport (OT) enables its rapid computation. There- fore, Geolabe LLC will develop and adapt methods based on scalable optimal computational transport to trace back the origin of gas emissions in order to identify gas emissions linked with gas migration and leakage. Geolabe LLC proposes to (i) adapt recent developments in Optimal Transport, inverse transport using Wasserstein barycenters and weather data assimilation that (ii) leverage the recent discovery of computationally efficient approximations of Optimal Transport; in order to propose a turnkey software for monitoring possible gas leaks from Underground Gas Storage settings, that can directly be used on freely available Sentinel 5 methane data; and (iii) apply our method to the 2015 Aliso Canyon Gas Storage facility incident to demonstrate our technology. The cost of gas losses resulting from leaks is partially transferred to US consumers. It is estimated that US consumers paid an extra $20 billion in natural gas between 2000 and 2011 solely from leaks in the distribution pipeline network. Geolabe LLC will deliver an automatic subsurface gas leak detection software in satellite hyperspectral and multispectral images, that will monitor the UGS network integrity. Our product purely relies on state- of-the-art algorithmic advances in AI and relies on freely available satellite data. The cost of our solution is negligible compared with the cost of existing solutions that do not scale or rely on repetitive expensive aircraft flights or commercial satellites.
