E‑STGCN: Extreme Spatiotemporal Graph Convolutional Networks for Air Quality Forecasting

Authors: Madhurima Panja, Tanujit Chakraborty, Anubhab Biswas, Soudeep Deb

This study introduces E‑STGCN, a neural architecture that merges spatiotemporal graph convolutions with extreme value theory (using generalized Pareto distributions) to forecast extreme and rare air pollution events (PM2.5, PM10, NO2) over 37 stations in Delhi.

The model is evaluated against benchmark temporal and spatiotemporal models, and can also produce probabilistic prediction intervals using conformal prediction techniques.

Read the full paper on arXiv.

Recommended citation: Panja, M., Chakraborty, T., Biswas, A., & Deb, S. (2024). "E‑STGCN: Extreme Spatiotemporal Graph Convolutional Networks for Air Quality Forecasting." arXiv preprint arXiv:2411.12258.
Download Paper | Download Bibtex