Artificial intelligence for PM2.5 estimation and forecasting in southeast Asia

Traffic and industrial emissions sustain background PM 2.5 pollution in Southeast Asia. Seasonal peatland and agricultural fires produce transboundary haze. Monsoon winds carry that pollution between countries. Rainfall removes particles from the atmosphere. Chemical reactions during transport can also form secondary aerosols before polluted air reaches downwind cities. This review examines how atmospheric processes affect model evaluation and assesses the reported accuracy. Data-driven algorithms track normal concentration changes, but their accuracy varies between studies and tests. Differences in predictors, validation setups and satellite data availability limit comparisons between studies. Some models show larger errors during extreme PM 2.5 events, at unseen stations or across different monsoon seasons. Feature attribution scores show how algorithms work internally, but they do not show actual physical emission sources. Evaluations should report separate scores for severe haze days, test geographically withheld stations and assess physical consistency against chemical transport models. Standardized data for Mainland and Maritime Southeast Asia can help assess whether new algorithms support public health warnings.