By Saad Iqbal
Somewhere over your wellpad, a satellite the size of a microwave oven just measured a methane plume you didn’t know existed. It wasn’t looking for you specifically — GHGSat’s constellation scans facilities worldwide on a schedule, not a grudge — but if the plume is big enough, the math that follows is unforgiving. A compressor seal that’s been leaking quietly for three weeks isn’t just lost product anymore. It’s a data point in someone else’s regulatory system, and you found out about your own equipment from space before your own LDAR survey caught it.
That scenario is no longer hypothetical for upstream operators, and it’s why continuous methane monitoring has moved from “nice for ESG reporting” to “the difference between catching a leak in hours versus a quarter.” AI-driven leak detection at the wellpad doesn’t replace a field tech with a camera — it changes how long a leak gets to run before anyone notices.
What Actually Counts as a “Super-Emitter,” and Who’s Watching for One?
Under EPA’s Super-Emitter Program, a single detected event of 100 kilograms of methane per hour or more, at or near an owner’s or operator’s facility, qualifies as a reportable super-emitter event. That threshold isn’t academic. It’s roughly the kind of release rate you get from a stuck-open relief valve, a failed compressor rod packing, or a wellhead that’s been quietly weeping gas since the last workover — not a trace fugitive leak, but something a continuous monitor should catch fast and a quarterly walk-around might miss entirely.
The notifications don’t only come from EPA inspectors. The program allows EPA-certified third parties — contractors or NGOs using satellite, aerial platforms (drones, fixed-wing aircraft, helicopters), or ground-based mobile monitoring — to submit detections directly. Optical gas imaging doesn’t qualify as a third-party submission method under the program, and third parties can’t physically access your site. Once a notification lands, the clock starts: an operator has 5 calendar days to begin an investigation and 15 calendar days to submit a report back to EPA. That’s not a lot of runway if the first you’re hearing about your own leak is a notice from a party that scanned you from a plane or a satellite pass.
Wait — Didn’t Congress Kill the Methane Fee? What’s Actually Still in Force?
Here’s where a lot of operators are working from outdated information, and it’s worth being precise about it. The Inflation Reduction Act’s Waste Emissions Charge — the per-ton fee that was set to ramp from $900 in 2024 to $1,200 in 2025 and $1,500 for 2026 and later reporting years — was targeted for repeal by Congress under the Congressional Review Act in early 2025. The House and Senate both passed the repeal resolution, and it was expected to be signed into law. The practical upshot: don’t plan around that fee schedule as a live cost driver going into a 2026 budget.
What didn’t go away is the detection and notification machinery. EPA’s mid-2025 interim final rule delayed the Super-Emitter Program’s future implementation by roughly 18 months rather than eliminating it, alongside pushing back several other oil-and-gas methane rule deadlines. In other words: the fee that was supposed to make a leak expensive is gone, but the mechanism that makes a leak visible to regulators, investors, and insurers is still on the books and still scheduled to come back online. For an upstream engineer, that’s the detail that matters — continuous monitoring earns its budget line through leak response time and reputational exposure, not through dodging a per-ton charge that no longer exists.
How Does AI Actually Change What Gets Caught?
Three detection layers are doing the real work at most wellpads today, and each trades off coverage against resolution.
Satellite monitoring, like GHGSat’s dedicated constellation, measures methane concentration over a facility from orbit and can flag a large plume anywhere on the planet without a single sensor on your lease. The tradeoff is revisit frequency and plume-size floor — satellites are built to catch the big, costly releases, not a slowly degrading valve packing leaking at a few kilograms an hour.
Continuous ground-based monitoring, the category Project Canary and similar networks operate in, places sensors directly at the wellpad and fenceline, feeding data into a model continuously rather than on a flyover schedule. This is where AI does its most practical work: distinguishing a genuine rising-concentration event from wind-direction noise, background flaring, or a neighbouring site’s emissions, and doing it fast enough to alert a field tech the same shift instead of the same quarter.
Optical gas imaging cameras, such as Opgal’s EyeCGas line, remain the tool for pinpointing exactly which component is leaking once a broader system has told you something’s wrong at that site — walking a compressor skid with a thermal camera that renders a methane plume visible to the naked eye is still how you find the specific flange or seal, satellite and fenceline data can only tell you the neighbourhood.
None of these three layers is “the answer” on its own. The honest architecture most operators converge on is wide-area satellite screening for the catastrophic cases, continuous site-level monitoring for the routine drift that accumulates into real volume, and OGI cameras for the final walk-down. That layered approach is the same logic behind good virtual flow metering programs — you don’t trust one measurement point when the consequence of being wrong is expensive.
The modelling problem underneath all three layers is the same: raw concentration readings are noisy, wind-driven, and full of transient spikes that have nothing to do with a leak, so the useful work isn’t the sensor itself — it’s the layer that turns a concentration time series into a confident “yes, investigate this” or “no, that’s background.” That typically means building a site-specific baseline from historical readings segmented by wind speed and direction, flagging sustained deviations rather than single-sample spikes, and weighting recent readings against seasonal patterns so a hot, still August afternoon isn’t scored the same way as a windy spring day. Done well, this is what turns a continuous monitor from a device that cries wolf into one a field tech actually trusts enough to act on without double-checking first.
Where Do the False Alarms Actually Come From?
Continuous monitoring systems earn trust slowly and lose it fast, so it’s worth knowing where the noise comes from before you roll one out across a field.
- Wind direction dominates fenceline readings. A sensor downwind of a flare or a neighbouring lease on a given day can register a spike that has nothing to do with your equipment. Models need wind-corrected baselines, not raw concentration thresholds.
- Compressor stations are chronic, not acute, sources. Reciprocating compressor rod packing and valve leaks tend to leak continuously at a modest rate rather than in dramatic events — exactly the pattern a quarterly LDAR survey is bad at catching and continuous monitoring is built for. It’s also why compressor maintenance and emissions monitoring overlap more than most facility org charts reflect; see our piece on predictive maintenance for gas compressors for the mechanical side of that same leak path.
- Satellite and fenceline data disagree more often than vendors advertise. A satellite pass averages over a footprint that can include several facilities; your fenceline sensor sees only yours. When the two disagree, that’s not a system failure, it’s two different measurement scales telling you different parts of the truth — reconcile them, don’t pick whichever one you like better.
A Practical Rollout Checklist for Continuous Methane Monitoring
- Map your highest-consequence leak paths first — compressor seals, dehydration units, and tank vents typically dominate site-level methane intensity more than wellhead equipment does.
- Decide your detection layering deliberately: satellite for facility-wide screening, continuous ground sensors for trend and alerting, OGI cameras for component-level confirmation — don’t rely on just one.
- Build a wind-corrected baseline before you trust any alert threshold; a week of calm-wind data looks nothing like a week with a shifting prevailing breeze.
- Pre-stage your 5-day and 15-day response workflow now, before a third-party Super-Emitter notification arrives, so the investigation isn’t being designed under a deadline.
- Track every confirmed leak back to a root cause and a repair, not just a closed alert — the monitoring system’s value is measured in leaks fixed, not alerts generated.
That satellite pass over your wellpad will happen again, on its own schedule, whether or not you’ve done anything about the leak it might catch. The only variable you actually control is whether your own monitoring found the compressor seal first — in hours, not in someone else’s quarterly data release.
