Developing Machine Learning Models for Air Quality Forecasting and Decision Support in Non-Attainment Cities of India

Context & Background

Air pollution is a major environmental threat in Indian cities. Over 100 cities are classified as 'non-attainment' under the National Clean Air Programme (NCAP). Municipal authorities lack real-time predictive tools to forecast pollution levels and execute preventive measures.

Problems to be Addressed

Forecasting PM2.5 and PM10 levels requires fusing meteorology, local emissions, and satellite data, which is highly non-linear. Legacy models fail to predict short-term spikes in toxic air.

Aims and Objectives

1. Formulate predictive machine learning models for PM2.5/PM10.
2. Integrate air quality sensors with real-time forecasting dashboards.
3. Build a decision support system for municipal action plans.

Methodology

The team collects weather, traffic, and industrial emission datasets. LSTM and recurrent neural networks (RNNs) learn temporal trends. The output is integrated into a public dashboard, giving municipal agencies 48-hour advance forecasts of pollution spikes to manage traffic and industries.

Expected Outcomes

An air quality forecasting web portal, decision dashboards for partner cities, and publications in leading atmospheric science journals.