Automated identification, industrial attribution and conditional emission-rate quantification of recurrent CO enhancement source regions from TROPOMI observations

Haoyi Wu, Renwei Meng

The TROPOspheric Monitoring Instrument (TROPOMI) repeatedly measures global carbon monoxide (CO) total columns, but industrial CO studies generally begin with known cities or facilities; a general method is therefore needed to discover where enhancements recur globally. We developed an automated workflow that searches without facility coordinates. A weakly supervised model screens scenes for localized CO enhancements; corrected CO pixels then delineate their footprints, duplicate detections within an overpass are merged, and nearby enhancements on different dates are grouped into candidate source regions. Only afterward are industrial facilities and active fires introduced to assess industrial context. Polygon-CSF estimates instantaneous emission rates only when enhancement geometry and winds permit cross-sectional calculation. Applied to 20,279 TROPOMI Level-2 CO product files from 2020 to 2023, the workflow identified 25,587 independent enhancement observations and 13,363 candidate source regions. Of these, 4,986 were observed at least twice and 168 at least 10 times; 134 of the latter were detected in all four years. The proportion with a nearby industrial facility record rose from 38.1% among single-observation regions to 81.0% among regions observed at least 10 times. Among 180 regions with the strongest industrial evidence, 136 contained multiple industrial facility types. Under qualifying conditions, polygon-CSF yielded 512 instantaneous emission-rate estimates across 177 regions, with a region-level median of 35.0 Mg h⁻¹. By linking automated global CO-enhancement discovery, industrial attribution and conditional emission-rate quantification, the workflow prioritizes industrial areas for verification and continued monitoring.