---

A Foundational Methane Detection Dataset: Transparent Access to Cloud-Optimized Spatio-Temporal Datasets (TACO)

PI and co-PIs: Luis Gómez-Chova (University of Valencia, Spain); Cesar Aybar (University of Valencia, Spain); Julio Contreras (University of Valencia, Spain); Luis Guanter (Polytechnic University of Valencia, Spain); David Montero (Leipzig University, Germany); Miguel Mahecha (Leipzig University, Germany)

Funding amount: $150,000

Project overview: For decades, oil and gas companies, landfills, and industrial facilities self-reported methane emissions with no independent verification. Methane monitoring through Earth observation satellites has revealed actual emissions 50 to 200 percent higher than reported. This discovery is urgently important because methane accounts for 45 percent of recent warming, is 86 times more potent than CO2, and its 12-year atmospheric lifetime means that reducing emissions leads to measurable cooling. Machine learning can automatically detect methane plumes in satellite imagery, yet building robust models requires training on comprehensive datasets spanning multiple sensors. Currently, methane data remain scattered across incompatible formats, forcing researchers to spend months on manual harmonization before any model training begins. To address this gap, the project will create TACO, a toolbox designed to facilitate the creation of FAIR-compliant Earth observation datasets, including the first unified AI-ready dataset for methane detection, standardizing observations across multiple satellite sensors. This foundational methane detection dataset will unlock diverse training data, enabling improved models and more accurate estimates to drive targeted mitigation, helping governments and operators to curb methane emissions.

Earth Observation & Monitoring Computer Vision & Remote Sensing Data Mining