SRM Institute of Science & Technology
Spatiotemporal GNN for PM2.5 Forecasting
Graph neural networks forecasting air quality across 184 cities.
Jan 2022 – May 2022
A spatiotemporal GNN that forecasts air quality across 184 cities
PM2.5 is spatially coupled (a city's air quality depends on its neighbors') which purely temporal baselines structurally cannot represent. I modeled 184 cities as a graph and combined Graph Isomorphism Networks with GRUs in PyTorch Geometric, matching the architecture to the shape of the problem, and beat the baselines by 7.2% RMSE and 12.4% CSI at four-day horizons.
- −7.2%
- RMSE vs baseline
- +12.4%
- CSI vs baseline
- 184
- cities modeled
- PyTorch Geometric
- GNN
- GIN
- GRU
- Python
The problem
PM2.5 concentration is spatially coupled (a city's air quality depends on its neighbors'), which purely temporal baseline models cannot represent.
What I built
- 01Engineered a spatiotemporal architecture combining Graph Isomorphism Networks with GRUs in PyTorch Geometric.
- 02Modeled 184 cities as a graph so spatial coupling was learned rather than ignored.
- 03Targeted four-day prediction horizons.
Outcome
7.2% lower RMSE and 12.4% higher CSI than baseline models, at four-day prediction horizons across 184 cities.