Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{208210,
        author = {Rahul Raut and Karan Khonde and Mohit Jangale and Varsha Hole},
        title = {Evaluation of Threshold-Based and Hybrid Aerosol Classification Frameworks Using AERONET Observations},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {4},
        pages = {1039-1057},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=208210},
        abstract = {Accurate classification of aerosol types is crucial for understanding their optical properties and their effect on the Earth–atmosphere system. Aerosol classification is typically performed either by fixed thresholding of optical parameters or, more recently, by data-driven probabilistic approaches such as Gaussian kernel density estimation (GKDE). Previous studies have applied the two approaches independently, and it remains poorly understood how the choice of method affects aerosol-type attribution, representative aerosol characteristics, and subsequent optical interpretation. In this study, two aerosol classification methods are investigated using long-term Aerosol Robotic Network (AERONET) inversion products at 47 global sites. A complete hybrid aerosol classification framework is implemented, comprising Gaussian kernel density-based preliminary classification, selection of representative aerosol observations, generation of a consistent optical database using the Mie scattering model, and label refinement through machine learning with ensemble classifiers. Separately, a traditional threshold-based classification is applied to the same AERONET dataset following fixed-parameter decision criteria; the threshold-based labels do not undergo representative-observation selection, Mie-based optical modelling, or machine learning, and serve only as a reference for comparison. Results from both pathways are analysed to investigate differences in aerosol-type distributions, the redistribution of observations between aerosol categories, variations in representative aerosol observations and their refractive indices, and the resulting effect on the interpretation of aerosol optical characteristics. Despite their fundamentally different design, the two methods produce highly consistent aerosol-type assignments, with agreement exceeding 98 % for all four aerosol types (dust, mixed, urban/industrial, and biomass-burning). The differences that occur are confined to a small fraction of observations near class boundaries, where the fixed-threshold scheme and the density-based hybrid framework treat transitional cases differently. Because label assignments largely coincide, the representative refractive indices and optical properties derived under the two schemes are also closely comparable. The hybrid framework therefore reproduces the established threshold-based classification while adding a physically grounded, probabilistic description of aerosol types, providing a quantitative basis for understanding how method choice affects aerosol-type attribution in climate-relevant analyses.},
        keywords = {Aerosol-type classification, Gaussian kernel density estimation, Mie scattering model, optical properties, random forest, threshold-based classification.},
        month = {September},
        }