Persistent emitter or episodic leak: reading a methane plume time series
A single satellite pass over a pad tells you one thing: there was a plume there on that overpass, at that time, at roughly that rate. It does not tell you whether you just caught a stuck pneumatic controller that has been bleeding for eight months, or a blowdown that started an hour before the satellite flew over and will be closed out by dinner. Both show up in the pixel the same way. The difference only shows up over time.
This is the question that actually matters for an MRV program or an emissions-weighted portfolio screen: is this site a repeat offender, or did we catch it mid-event. Get that wrong and you either chase a phantom leak that closed itself, or you write off a chronic emitter as "noise from one bad pass."
Why one overpass can't settle it
Most regional methane advisories and annual flyovers give you exactly one data point per site per visit. A single detection records that gas was in the air at that moment. Confirming a recurring source takes more than one look. Liquids unloading, a planned blowdown, tank flashing during a hot afternoon, or a compressor startup can all produce a plume that is gone an hour later and never comes back. If your only look at that site is the one overpass that happened to catch it, you have no way to separate that from a seal that has been leaking since spring.
More images of the same site boundary, close enough together in time, tell you whether the plume is still there next time you look. That's the only way to turn "we saw a plume once" into "this site has a chronic source."
What the two patterns look like in the data
A persistent emitter shows a plume at or near the same spot on the site, pass after pass, over days or weeks. The rate can drift a bit between images (wind, source pressure, partial repairs), but the plume keeps reappearing from the same equipment location. If you've got five consecutive clear-sky detections off the same compressor skid, you're not looking at coincidence. You're looking at a source that needs a work order.
An episodic emitter looks different: a plume shows up once, maybe twice in a short window, and then the site goes quiet on every subsequent clear pass. Sometimes you can tie the timing to a known activity, like a liquids-unloading cycle or a scheduled blowdown logged in ops records. Sometimes you can't tie it to anything and you just have to trust the pattern: one hit surrounded by clean passes is a different story than a hit surrounded by more hits.
Classification runs on revisit count, not plume strength. A big plume seen once is still just one data point. A smaller plume seen on four of five passes is a persistent source, full stop.
Attribution to the site, not the basin
None of this works if the detection only tells you "somewhere in this county." Regional advisories that name a basin or a county are built for a different purpose; they can't support a pass-by-pass time series because they were never resolving individual site boundaries to begin with. Classifying persistent versus episodic requires imaging that resolves the SWIR absorption signature down to a specific plume and ties it back to a specific pad, so that the second, third, and fourth detections can be checked against the first one at the same site boundary.
That's the gap a per-site plume map that updates daily where tasking allows is built to close: instead of waiting on the next annual flyover or a coarse regional bulletin to name a basin, you get a time series on the specific site you're already watching, so a repeat detection reads as a repeat detection and not as a coincidence.
If you're trying to sort your watchlist into "fix this" and "already closed," that's the view worth pointing at your highest-risk sites first.