Semi-Automated Methane Detection in Multispectral Imagery using Machine Learning

Colombia has a strategic position that allows it to exploit resources for electricity generation. By 2012 the country had installed capacity mainly from hydroelectric and natural gas. The latter consists of between 80% and 90% methane. At the environmental level, it is crucial to monitor the infrastructure that transports the gas, detecting leaks and preventing this powerful gas from affecting the atmosphere. This research proposes a methodology for the detection of methane from multispectral images captured on Colombian territory by MAXAR’s Worldview-3 satellites, which operate in a spectral range of 400 nm to 2365 nm, reaching up to 31 cm spatial resolution with the Pansharpened process. Automated comparisons were made between image values and the spectral signature of methane for hydrocarbon identificación in the area. The methodology presented proved to be useful in identifying methane accumulations in multispectral images, avoiding in situ measurements. This method contributes to more efficient methane management in the energy transition and environmental protection. The algorithm developed shows optimal performance in open areas, but in urban areas, there may be interference with certain materials. In this case study, a complementary analysis was carried out that purifies the result, but to achieve a better yield it is recommended to use it in rural areas.