Evidence note: This article rests on layered evidence. The work by Google Research and the akmobius/mapping-methane project is described from primary sources the subjects published themselves — a Google Research blog post and a public code repository — and is stated as fact with that attribution kept explicit. Claims about Descartes Labs, Kayrros, and the GeoWaste Database Repository come from single secondary, tertiary, or self-published sources that have not been independently corroborated here; those are presented with uncertainty and are not treated as settled. Where a source names who reported something, that attribution is preserved throughout.
Methane is a short-lived but powerful greenhouse gas, and finding out where it escapes — from oil and gas infrastructure, landfills, or agriculture — is a measurement problem before it is a policy one. Satellites help because methane absorbs shortwave infrared light, so instruments in orbit can estimate how much of it sits in a column of atmosphere below them. The harder part is turning noisy, cloud-interrupted, globe-spanning readings into something a person can act on: a map that shows concentrations and, ideally, pinpoints the plumes. That last step is where machine learning and open tooling increasingly come in. The projects below show the range of approaches, from a large research lab to a single public repository.
Google Research: deep learning applied to methane mapping
Google Research published a blog post describing work on mapping global methane emissions from space using deep learning (research.google). The framing is worth stating plainly, because it is attested by Google’s own account rather than inferred: the effort applies deep-learning models to satellite observations to produce global methane maps.
Why reach for deep learning at all? Raw satellite retrievals are messy — clouds, surface reflectance, and instrument noise all sit on top of the methane signal. Learned models can help separate the signal of interest from that background and flag likely emission sources at scale, which is difficult to do by hand across the entire planet. The specifics of the model architecture and any accuracy figures are not restated here, because they would need to be read directly from the source rather than paraphrased into numbers.
Descartes Labs and its methane work
A post published on Medium under the Descartes Labs team account, titled "Sharpening the Focus on Methane Emissions," discusses the company’s attention to methane (medium.com). That Descartes Labs has published work on mapping global methane emissions is reported by this self-published post but is not independently corroborated here (evidence: a single vendor-authored Medium article). Readers wanting the concrete methods should treat the post as the company’s own description of its work rather than as third-party verification of it.
Kayrros: global-scale methane mapping from satellite data
According to a Wikipedia entry, Kayrros is an analytics firm associated with satellite-based methane monitoring (en.wikipedia.org). The specific claim that Kayrros maps methane emissions on a global scale from satellite data is reported by that entry but unverified here (evidence: a tertiary encyclopedia source, not confirmed against Kayrros’s primary materials). It is a plausible description of the type of work such firms do, but this article stops short of stating it as an established fact.
Kayrros satellite data and the COP26 methane pledge
The same Wikipedia entry is the basis for connecting Kayrros satellite data to the COP26 pledge to reduce methane emissions (en.wikipedia.org). That a link was drawn between Kayrros’s data and the COP26 methane commitment is reported, not confirmed (evidence: the same tertiary source; the COP26 association has not been checked against primary reporting). The broader context — that COP26 produced a high-profile methane-reduction pledge — is widely discussed, but whether and how Kayrros data figured into it remains an open point here rather than a documented one.
akmobius/mapping-methane: TROPOMI data to SVG maps
The clearest end-to-end "how" among these examples comes from a small open project. The akmobius/mapping-methane repository gathers methane data from the Copernicus Sentinel-5P satellite’s TROPOMI instrument into a database and generates SVG image maps of global methane concentrations, published on a website — this is stated by the project’s own repository (github.com) and is reported here as fact on that basis.
The pipeline is instructive because each stage is visible: Sentinel-5P/TROPOMI supplies the atmospheric methane readings; those readings are collected into a database; and the stored values are rendered as SVG maps for the web. For a reader trying to understand the mechanics of satellite methane mapping, a public repository like this is a practical entry point precisely because the ingestion-to-visualization path is laid out in code rather than described at a distance.
GeoWaste Database Repository: open tools for dumpsites and methane
The GeoWaste_Database_Repository on GitHub is described as providing open-source tools for global mapping and characterisation of major dumpsites using AI, geospatial analysis, and methane emissions modelling (github.com). That description is drawn from the repository’s own materials and is not independently corroborated here (evidence: the project’s self-description). If accurate, it points at a distinct slice of the problem — waste sites as methane sources — and complements the atmospheric-column approach by modelling emissions from specific land features.
Where a reader can go next
Taken together, these efforts span the pipeline that turns orbital measurements into usable maps: a shared upstream data source in Sentinel-5P/TROPOMI, machine-learning methods to extract signal, and different downstream products — global concentration maps, source attribution, and site-level modelling. The two examples grounded in primary sources here (Google Research’s deep-learning post and the akmobius repository) are the firmest starting points; the Descartes Labs, Kayrros, and GeoWaste descriptions are best treated as leads to verify against each project’s own primary materials before relying on them. Because several of these tools and datasets are public, a curious reader can inspect the code and, in the TROPOMI case, the underlying satellite data directly rather than taking any single account at its word.