Knowledge distillation enables prediction of ring-class polycyclic aromatic hydrocarbons concentration using...

Area of Science:

• Environmental Chemistry
• Environmental Science
• Sensor Technology

Background:

• Polycyclic aromatic hydrocarbons (PAHs) concentrations shift significantly during rainfall in urban estuaries.
• Current monitoring methods lack either temporal frequency (chromatography) or molecular specificity (drones).

Purpose of the Study:

• To bridge the gap between laboratory specificity and field-scale monitoring frequency for PAHs.
• To develop a scalable method for assessing PAHs dynamics in urban estuaries during rainfall events.

Main Methods:

• Knowledge distillation was used to transfer laboratory spectral information into sensor-based field models.
• A variational autoencoder augmented limited sample data under rainfall conditions.
• The integrated framework was validated using extensive autonomous underwater drone measurements.

Main Results:

• The developed framework achieved a high predictive performance (R² = 0.92 for ΣPAHs), improving accuracy by 28%.
• Large-scale drone data revealed rainfall-induced surges of high-molecular-weight PAHs and shifting pollution hotspots.
• Interpretable analyses showed how spectral signatures change under hydrological perturbation.

Conclusions:

• This approach successfully couples laboratory specificity with autonomous sensing for effective PAHs assessment.
• It provides a scalable strategy for understanding pollutant dynamics in urban waters affected by rainfall.
• The method enhances process-resolving pollutant monitoring in complex aquatic environments.