A hybrid AI framework for aerosol type classification and 10-year forecasting using composition and source-based features

Aerosol behavior is critical for atmospheric research, environmental monitoring, and climate modeling, necessitating a comprehensive understanding of its composition, sources, and temporal dynamics to enhance air quality assessment and climate predictions. This study introduces a hybrid artificial intelligence framework designed for aerosol classification and long-term forecasting, specifically applied to the Dibrugarh region of India. The methodology leverages a multiannual dataset encompassing key aerosol parameters, including Aerosol Optical Depth (AOD), Angstrom Exponent (AE), Fine Mode Fraction (FMF), and Single Scattering Albedo (SSA). Various machine learning and deep learning models, such as Random Forest, XGBoost, CatBoost, LSTM, and Transformer architectures, are employed. The framework integrates composition-based, source-based, and seasonal classification, complemented by

correlation analyses using PCA, t-SNE, and Chi-square tests. Data imbalance is addressed through techniques like SMOTE and ADASYN, with long-term forecasting executed via a BiLSTM-Attention model. The proposed models demonstrate high performance across multiple tasks, with the BiLSTM-Attention model achieving approximately 90% accuracy in aerosol type forecasting over a 10-year horizon. Findings indicate a strong dependency between aerosol composition and source types, alongside significant seasonal variations in aerosol behavior.

Message: Hi FE, I need help with “A hybrid AI framework for aerosol type classification and 10-year forecasting using composition and source-based features” in “Environmental Engineering”