A supervised machine learning approach to classify traffic-derived PM sources based on their magnetic properties

Area of Science:

• Environmental Science
• Geophysics
• Analytical Chemistry

Background:

• Particulate pollutants pose environmental and health risks.
• Environmental magnetism offers a method for mapping pollutant deposition.
• Tracing pollutant sources to sinks is crucial for effective management.

Purpose of the Study:

• To evaluate the k-nearest neighbors (kNN) algorithm for classifying traffic-related particulate matter sources.
• To explore the source-to-sink traceability of pollutants using magnetic properties.
• To assess the potential of magnetic mapping for complementing air quality monitoring.

Main Methods:

• Characterizing magnetic properties of traffic-related particulate matter sources (tire, brake pads, exhaust).
• Training a kNN classification model with source magnetic parameters.
• Measuring magnetic parameters on accumulating surfaces (plant leaves, filters).
• Confronting measured magnetic data with the trained kNN model for source identification.

Main Results:

• The kNN algorithm successfully predicted dominant traffic-related sources for various accumulating surfaces.
• Model predictions showed consistency across different sampling locations.
• The method achieved adequate resolution, distinguishing sources within a single street.
• Demonstrated the ability to trace traffic-derived pollutants from source to sink using magnetic signatures.

Conclusions:

• kNN classification based on magnetic properties enables effective tracing of traffic-derived pollutants.
• Magnetic mapping provides a high-resolution complement to conventional air quality assessment methods.
• This approach can enhance numerical models for pollutant dispersion and source apportionment.