Machine learning based aerosol classification over South and East Asia using MODIS top of atmosphere reflectance and...

Aerosols play a vital role in atmospheric processes, climate, air quality, and human health. South and East Asia, experience variable aerosol conditions driven by natural and anthropogenic emissions, strongly influenced by seasonal meteorology. Ground network AERONET provide accurate observations but is spatially sparse, while satellite sensors like MODIS offer broader coverage but existing classification methods often rely on thresholds or retrieval products, limiting robustness and seasonal representation. This research develops a scalable aerosol classification framework using MODIS (MOD02SSH) top-of-atmosphere reflectance (TAR), validated with AERONET-derived clusters. Unsupervised clustering of 22 AERONET parameters identified four dominant aerosol types (Urban/Industrial, coarse desert dust, fine absorbing, and fine non-absorbing), which served as reference labels for supervised machine learning (ML) model, i.e. Light Gradient Boosting Machine (LGBM). The model is trained on MODIS spectral bands, angular geometry, and derived indices, with performance assessed using multiple metrics, alongside Area Under the Receiver Operating Curve (AUROC). The framework was also applied to Multi-Angle Implementation of Atmospheric Correction (MAIAC) surface reflectance (MCD19A1) and compared with MOD02SSH TAR through pixel-wise resampling, diverging plots, and per-class statistics. The model achieved high accuracy (74 %), spatial patterns highlighted regional aerosol type dominance, while cross-product comparison confirmed overall robustness but revealed localized atmospheric correction effects. By integrating unsupervised clustering with supervised ML, this study demonstrates that TAR-based classification can bypass retrieval uncertainties, capture seasonal variability, and deliver high quality aerosol maps. The framework provides a transferable tool for climate modeling, transboundary pollution monitoring, and air quality management.