Measuring CO2 emissions using deep learning and remote sensing
Abstract
Global emissions of carbon dioxide (CO2) are currently estimated using "bottom-up" accounting from national greenhouse gas inventories, which largely rely on self-reporting. Different inventories, however, can disagree, and the coarse resolution of emissions reporting make it difficult to target reduction strategies. Remote sensing of greenhouse gases has the potential to address existing gaps by providing "top-down" measurements that are high resolution, objective, and transparent. Multiple satellites, such as Greenhouse gases Observing SATellite (GOSAT), Orbiting Carbon Observatory-2 (OCO-2), and OCO-3, highlight carbon sources and sinks that fulfill this objective with additional missions launching in the near future. However, CO2 enhancements from anthropogenic sources tend to be small compared to background levels, making it difficult to attribute emissions to specific emitters. In this work, we explore the feasibility of scalable CO2 emissions monitoring using a deep learning approach, starting with electricity-generating facilities in the United States. Our method circumvents the need for a chemical transport model and enables the integration and investigation of relationships of multiple data sources. We compare our results with other emissions databases, discuss uncertainty estimations, examine strengths and limitations in our approach, and provide a roadmap for non-point source emissions mapping in the future.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.A15L1389M