ESTGCN: Extreme Spatiotemporal Graph Convolutional Networks for Air Quality Forecasting.
Journal of the Royal Statistical Society: Series A., 2026
Accurate air quality forecasting is challenging due to strong nonlinearity, nonstationarity, spatiotemporal dependencies, and the presence of extreme pollution events. In this work, we propose E-STGCN, an EVT-guided spatiotemporal graph neural network that explicitly models extreme pollutant behavior using a generalized Pareto distribution. The framework combines graph convolutions for spatial structure with LSTM-based temporal dynamics.
Overview of the Extreme Spatiotemporal Graph Convolutional Networks.