Bayesian estimation of CO2 flux divergence maps using joint (CO2M-like) NO2 and CO2 images

To support the ambition of national and EU legislators to substantially lower greenhouse gas (GHG) emissions as ratified in the Paris Agreement on Climate Change, an observation-based "top-down" GHG monitoring system is needed to complement and support the legally binding "bottom-up" reporting in national inventories. For this purpose, the European Commission is establishing an operational anthropogenic GHG emissions Monitoring and Verification Support (MVS) capacity as part of its Copernicus Earth observation programme. A constellation of up to three CO₂, NO₂, and CH₄ monitoring satellites (CO2M) will be at the core of this MVS system. The satellites, to be launched from 2026, will provide images of CO₂, NO₂, and CH₄ at a resolution of about 2 km  2 km along a 250-km wide swath. This will not only allow observing the large-scale distribution of the two most important GHGs (CO₂ and CH₄), but also capturing the plumes of individual large point sources and cities.

The divergence method can be used to estimate point source emissions from satellite images, using fewer assumptions than other light-weight plume quantification methods (e.g., no a-priori source locations have to be known). However, the method only uses a few pixels near a point source, while pixels downstream of the source are implicitly excluded. Combined with the high noise of, in particular, CO₂ satellite images, the divergence computed from a single overpass image is usually too noisy for CO₂ emission quantification. In order to improve the information content in divergence maps, it is therefore common to average the map over many images (e.g., computing monthly or yearly averages) to get better emission estimates. As a result, the temporal resolution of the divergence method is limited.

In this work, we present a novel approach to improve the information content in CO₂ divergence maps, by exploiting the joint information content present in the simultaneously acquired CO₂ and NO₂ images. The purpose of this approach is to allow us to compute accurate divergence maps based on fewer images than typically required for the divergence method, and thus to obtain a finer temporal resolution of emission estimates. The method assumes that the signal-to-noise ratio of NO₂ images is better than that of CO₂ images, while both images contain a similar set of plumes related to emission point sources. Based on the signal-to-noise ratio in the CO₂ and NO₂ images and their covariances, we can estimate the optimal CO₂ image and optimal CO₂ divergence map in the minimum mean square error (MMSE) sense using a linear MMSE estimator. We demonstrate the effectiveness of this estimator on examples from the SMARTCARB dataset (Kuhlmann et al., 2020), and show that an about +20 dB boost in the peak signal-to-noise ratio (PSNR) can be achieved for individual overpass CO₂ divergence maps, which is roughly equivalent to the PSNR improvement otherwise obtained by averaging 10 images. Our approach therefore allows us to estimate CO₂ emissions over shorter observation periods and increases the emission estimation accuracy of weaker sources.