VIIRS Nightfire Super-Resolution Method for Multiyear Cataloging of Natural Gas Flaring Sites: 2012–2025

We present a new method for mapping of gas flares in the multiyear spatio-temporal database of the VIIRS Nightfire (VNF) nighttime infrared heat source detections from the three satellites, Suomi NPP, NOAA-20 and NOAA-21. The algorithm is composed of several steps: (i) 2D histogram binning of the high temperature (>1200 K) detection counts into 15 arcsec latitude-longitude grid, (ii) segmenting of the counts histogram into oil-field sized watershed features that serve as a guide where to search for the VNF detection clusters, (iii) super-resolution clustering the cloud of detections within each feature into a Dirichlet process variational Bayesian Gaussian mixture of compact clusters centered at the location of individual flare stacks, (iv) post-processing of the detected flares to avoid over-splitting and to find flare attraction contours with Voronoi/Apollonius geometry, (v) classification of the detection clusters into a pre-defined categories such as upstream, midstream, LNG, etc. with provenance from the earlier flare catalogs and multimodal LLM reasoning. The AI-assisted classifier uses reverse geocoding of the IR-emitter coordinates, high-definition daytime satellite imagery and time history profiles of the detections inside attraction contours to hint the expert with a probable category of the emitter together with the short summary of reasoning. Compared to the annual catalogs used for the country-level estimates of flared gas volume, the new algorithm is robust to atmospheric glow from large flares (higher selectivity) results in twice the number of the active flares (higher sensitivity), located with subpixels precision ~50 m and separable within ~400-700 m. For the well-defined class of downstream flares at export LNG locations the catalog demonstrates near-complete detectability.