Air Quality Forecasting with a Spatio-Temporal Graph Neural Network

A three-day PM2.5 forecast for every continuous monitoring station of Delhi-NCR, built with a spatio-temporal graph neural network in PyTorch Geometric on a laptop CPU. The 56 stations become the nodes of a graph; a graph-convolutional GRU with learned, wind-aware edge weights reads 14 days of PM2.5, wind, temperature and humidity and forecasts day+1 to day+3. The forecasts are shown on station maps in the India National AQI bands by the AirCast India web application.

Category: Artificial Intelligence / Graph Neural Networks / Environmental Data Science / Spatio-Temporal Forecasting

CSE Project 25 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.