Exploring AI potential in optical analysis of organic and elemental carbons: A review of emerging applications in airborne particulate characterization

Abstract
Carbonaceous aerosols, comprising organic carbon (OC) and elemental carbon (EC), constitute up to 70% of atmospheric particulate matter (PM) and significantly impact human health, climate forcing, and air quality. Despite their significance, accurate characterization of OC and EC fractions remains challenging due to aerosols’ complex and variable composition and the constraints in conventional analytical techniques. This review aims to assess how artificial intelligence (AI) can advance OC/EC characterization by enhancing data interpretation, automation, and measurement reliability. Recent developments in photo-optical methods for OC/EC quantification, including real-time monitoring, destructive, and non-destructive techniques, are comprehensively analyzed. The review further examines emerging efforts to integrate AI into conventional analytical frameworks, critically evaluating the potential of semi-supervised learning, active learning, incremental learning, TinyML, explainable AI, and large language models (LLMs) for improving OC/EC analysis. Notably, the underexplored potential of these AI approaches in source apportionment and real-time monitoring is emphasized, alongside a proposed roadmap for implementing AI-driven strategies in carbonaceous aerosol analysis and air quality assessments. AI integration with existing OC/EC analytical techniques can minimize measurement uncertainties and enable real-time, cost-effective monitoring systems. This approach facilitates high-resolution source apportionment with broad applications in environmental monitoring research, ultimately improving urban air quality diagnostics and evidence-based pollution control strategies.