Emission-trend-constrained quality assessment model for remote sensing data of vehicle NO emissions

Remote sensing (RS) can monitor in-use vehicle emissions but suffers from variable data quality. This study obtained high-quality RS (HQRS) data via a controlled campaign and developed a machine learning framework (RDEV) to improve RS data usability. RDEV exploits the characteristic NO emission pattern (higher in older vehicles than in newer ones), to assess the reliability of remote sensing records, based on the accuracy of classifying vehicles of known-age. RDEV model achieved an accuracy of 86.95 % on the HQRS test subset. When applied to routine monitoring data, RDEV reduced variability in repeated vehicle measurements and aligned gasoline vehicle emission trends more closely with established literature, demonstrating its effectiveness in selecting high-credibility data. This study presents an alternative machine learning application that learns established emission patterns to screen massive RS datasets, offering a method to improve data quality and unlock the value of historical RS data.