A comprehensive artificial intelligence approach for multi-site temporal forecasting of particulate matter in the non-attainment cities of arid and semi-arid Western India

Rajasthan is home to several non-attainment cities where ambient air quality standards are persistently violated. Forecasting particulate matter (PM) is essential for protecting public health where air quality standards are persistently exceeded. This study develops a comprehensive artificial intelligence (AI) framework whose primary objective is the forward-looking, day-ahead forecasting of PM 2.5 concentrations in the five non-attainment cities of Rajasthan, namely Alwar, Jaipur, Jodhpur, Kota, and Udaipur, from 2018 to 2024, as reported by the Central Pollution Control Board (CPCB). The framework investigates eight air pollutants (PM 2.5 , PM 10 , nitric oxide (NO), nitrogen dioxide (NO 2 ), oxides of nitrogen (NOx), sulphur dioxide (SO 2 ), carbon monoxide (CO), and ozone (O 3 )) and three meteorological parameters - relative humidity (RH), wind speed (WS), and wind direction (WD). Air Quality Index (AQI) trend analysis showed that winter months recorded the worst deterioration across all cities, with PM 2.5 and PM 10 consistently the predominant contributors. Kendall rank correlation confirmed strong associations between PM fractions and AQI and identified WS and RH as the key meteorological drivers. Two deep-learning (DL) models - Long Short-Term Memory (LSTM) and Transformer, and two machine-learning (ML) models - Random Forest (RF) and Extreme Gradient Boosting (XGBoost), were trained on a chronologically ordered 80–10–10 split to forecast the next-day PM 2.5 from a seven-day look-back window of historical observations, and a city-specific best-fit model was selected for each city. The Transformer performed best for Alwar, LSTM for Jaipur, Jodhpur, and Kota, and RF for Udaipur. SHapley Additive exPlanations (SHAP) provided feature attribution for each model, confirming that lagged PM fractions were the most influential predictors. The selected models were then deployed to generate independent, year-long (2025) PM 2.5 forecasts, which were validated against CPCB ground observations. This study presents a comprehensive, meteorology-integrated approach to year-round PM 2.5 forecasting across all non-attainment cities of Rajasthan.