Satellite portals, UN notifications, and press reports now publish operator-linked methane attributions on a rolling cadence. Advect Labs re-derives the attribution from independent public data and delivers a scorable confidence on the claim.
Carbon Mapper, the UN's Methane Alert and Response System, and the EPA Super Emitter Response Program all publish operator-linked methane attributions on a rolling cadence. Journalists run their own analyses against the same feeds.
OGCI's Satellite Methane Detection Response Playbook, the de facto response reference, states that no best practices exist for evaluating the accuracy of a plume's origin. When an attribution lands on your assets, there is no accepted basis for judging whether it is correct. A field survey that finds nothing cannot close the question if the source was intermittent, and most detected sources are.
Re-derive the attribution from inputs the original accuser did not use: NOAA wind reanalysis, public well registries, and a forward Gaussian dispersion model run against the published plume centroid.
Assign a calibrated posterior over the wind-cone candidate set: a distribution, not a point estimate, that can be checked against ground truth and recalibrated when a gap appears.
A per-claim structured record covering source identification, uncertainty analysis, attribution rationale, response-workflow alignment, and complete data provenance. Filed alongside the operator's response to a notification.
Advect Labs does not own the detection layer. It re-derives and evaluates attributions that Carbon Mapper, UN MARS, EPA SERP, and other third parties publish. Independence comes from the inputs: wind, asset registries, and transport physics the original accuser did not use.
Methodology tested end-to-end on eleven Delaware Basin plumes at airborne (AVIRIS-NG) resolution against Cusworth et al. 2021 ground truth. That dataset publishes both the observed plume and an independently established true source, so each re-derivation can be scored. Nine plumes scored; two excluded under a documented low-wind cutoff. Confidence is mean-unbiased at the 150 m band on this sample (stated 0.88 vs. empirical 0.89, a −0.01 gap). Contact us to review the full methodology documentation.
Two-layer re-derivation: nearest-pad geometry for location, forward Gaussian dispersion for confidence. Nearest-pad recall 9/9; median truth distance 104 m. Confidence mean-unbiased at the 150 m band on this sample (stated 0.88 vs. empirical 0.89, −0.01 gap). Calibration degrades into measurable, disclosed overconfidence at satellite-scale input noise. The method measures this rather than asserting calibration it does not have.
Extension of the CS-01 pipeline to satellite-resolution detections, with a resolution-aware prior and abstention policy. Tests whether the confidence stays honest (or honestly abstains) on real Carbon Mapper and EMIT attributions where ground truth can be independently established.
For regulatory affairs, ESG, sustainability, and measurement leads at operators whose assets have been, or could be, named in a public methane attribution. No pitch decks.
DIRECT CONTACT
info@advectlabs.com