The Auditor Who Certifies Satellite Math

A regulator approving a detection system is signing off on math that lives 500 kilometres up and targets a facility they have never entered. The distance ought to make it easier. The distance makes it worse.

Carbon Mapper was approved by the EPA as third-party notifier TPN-0001 on 5 November 2024, which means that its Tanager-1 satellite, launched in August 2024, can now submit super-emitter notifications directly to the federal Super Emitter Program Portal. On 19 September 2024, Tanager-1 captured a methane plume 4 kilometres long over a Karachi landfill with a preliminary emission rate of 1,200 kilograms per hour. On 24 September, a plume south of Midland, Texas, in the Permian Basin was estimated at 400 kilograms per hour. A scene over Turkmenistan showed 13 plumes with a preliminary combined emission rate of 78 tonnes per hour, twice the average total methane output from the Los Angeles Basin. The operator at the Permian site has five days to initiate an investigation once EPA forwards the notification, 15 to file a report. The data is public, the site is not named until the operator confirms ownership, and the line between a detection and an enforcement action runs through an EPA certification process that most regulators had no part in writing.

I sit on the other side of that certification process often enough to know the shape of the problem. The regulator signs off on a method, the method produces a number, the number triggers a compliance clock, and the facility it points at has exactly one move: challenge the method. Not the emission, not the time of day, not whether the plume was theirs. The method. Because if the method was approved by a process the regulator cannot fully reconstruct, the number it produces is not a finding. It is a liability.

The EPA's 2024 methane rule defines super-emitter events as releases exceeding 100 kilograms per hour and established the Methane Super Emitter Program, allowing certified third parties using EPA-approved technology including satellite detection, aircraft sensors, or mobile platforms to submit data through the Super Emitter Portal. The process has the shape of rigour. EPA anticipates a 270-day review for applications combining technology approval and third-party certification, 60 days if the technology is already approved. But what the rule does not contain is a framework for how the regulator explains to the board of a midstream gas processor why a nonprofit with a hyperspectral imager is more authoritative than the continuous monitoring system that cost them $400,000 and passed the last audit.

The satellites work. A September 2026 study published in Atmospheric Measurement Techniques found satellite-derived methane emissions were 1.5 to 3 times higher than the global EDGAR inventory for London, 1.3 to 3.1 times higher for Los Angeles, and 7 to 10.2 times higher for New York; the EPA inventory underestimated emissions by 56% for Los Angeles and 84% for New York. UNEP's International Methane Emissions Observatory reported in July 2026 that AI-assisted workflows processed 12 to 15 times more satellite data while maintaining scientific rigour, with AI identifying 80 to 85% of confirmed methane detections before expert review. The detection layer is no longer experimental. It is operational, and it is feeding a compliance programme faster than the inspection capacity underneath it has caught up.

What the regulator inherits is not a technical question. It is an institutional mismatch. The third-party notifier is certified at federal level, the detection happens without notice, the emission estimate carries no plant-level context, and the enforcement expectation falls to a state or regional authority whose last methane training course covered Method 21 handheld analysers. TROPOMI operates at 7 kilometre by 3.5 kilometre spatial resolution, while Tanager-1 achieves approximately 30 metres by 30 metres, allowing attribution to individual facilities. The precision is new. The process for disputing it is borrowed from an analogue era.

I have been in the room when a regulator receives the first satellite-flagged notification for a facility in their jurisdiction. The first question is not whether the emission happened. It is whether the method used to calculate it is defensible in the administrative process that will follow. AI models in continuous emissions monitoring systems must produce outputs consistent with EPA Method 21 and 40 CFR Part 75 accuracy requirements, and exceedance detection logic must trigger automated regulatory notification workflows within permit timeframes. Those requirements exist for onsite systems. They do not yet exist in a form that maps cleanly onto orbital hyperspectral imaging with an AI plume-segmentation layer in the detection pipeline.

Google partnered with NASA's Jet Propulsion Laboratory to develop the MAPL-EMIT deep learning model, which detects 50% more emissions than human experts and has identified over 23,000 previously unmapped methane plumes, including 24 of the world's 25 largest-emitting landfills. NASA's EMIT satellite system has operated a fully automated Daily Methane Digest since 1 January 2026, sending predictions with normalised scores under 0.3 to analysts for catalogue consideration. The detection infrastructure is maturing in one direction. The regulatory apparatus is still anchored in quarterly manual surveys with geolocation accuracy measured in site visits, not pixels.

The gap is not scientific. It is procedural. The certified third-party submits a detection, EPA posts it to the public portal with geolocation but no operator name, and the operator has 15 days to file an investigation report. Third parties must submit notifications within 15 calendar days of detection, EPA reviews for completeness and accuracy before notifying facility owners, and subsequently posts the notification excluding operator attribution on the public Methane Super Emitter Program webpage. The pathway from plume to notification is clear. The pathway from notification to a control decision at plant level is still under construction. The operator receiving the notice does not get the raw spectral data, does not get the retrieval algorithm, and in most cases does not get enough information to align the satellite overpass time with the maintenance log or process upset that caused the release.

The regulator signing off on the certification is not signing off on one detection. They are signing off on a notification stream that will generate investigation obligations faster than most regional offices can schedule the follow-up inspections. IMEO has analysed over 1.3 million satellite observations since 2023, delivering a scalable methane monitoring approach as environmental data volumes grow rapidly. Every one of those observations that crosses the 100 kilogram per hour threshold and falls inside a certified third party's coverage area is a potential notification. Every notification is an investigation clock. And every investigation that cannot definitively link the plume to a process unit becomes an evidentiary problem when the operator challenges it.

I do not write this to discredit the satellites or the third parties operating them. The detection layer is the most significant methane transparency gain in a decade, and the facilities receiving notifications are, more often than not, the ones that need them. The problem is that the regulatory architecture underneath assumes the regulator can verify what the satellite saw. In practice, the regulator can verify that the method was certified. They cannot verify the emission without access to the same spectral inversion model, the same atmospheric correction, the same wind field data, and the same pixel-to-facility attribution logic that the third party used. Those layers are not secret. They are simply not included in the compliance file the operator receives.

MethaneSAT lost contact on 20 June 2025 and by 2 July was announced unrecoverable after likely running out of power, leaving Carbon Mapper's Tanager-1 and the anticipated Tanager-2 as the primary nonprofit satellites feeding the EPA programme. Tanager-1 completed two years of continuous operations in August 2026; Tanager-2 is expected to enter orbit at 405 kilometres altitude. The constellation is growing, the detection threshold is dropping, and the notification volume will follow. The regulatory question is not whether the satellites are accurate. It is whether the regulator can defend the number when it lands on the facility's desk, and whether the investigation process can close the loop between a detected plume and a verified source in time for the result to matter.

The facilities that can afford the appeal will appeal. The ones that cannot will comply, or they will shut the unit until the regulator moves on. Neither outcome serves the objective, which is to stop the emission and verify it stopped. That requires a procedural bridge between orbital detection and ground truth that does not yet exist in most jurisdictions, and building it is the regulator's problem, not the satellite operator's. I know, because I am often the one explaining to a facility why a number derived from space carries the same evidential weight as a reading taken at the flange, and the short answer is that it does not. The long answer is that it will, once the administrative process catches up. Until then, the regulator certifies the method, the method produces the notification, and the loop stays open until someone writes the guidance that closes it.

Tarry Singh is the founder and CEO of Real AI (realai.eu), an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan (earthscan.io) for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.