Risk Graph Neural Networks Improve Heat-Mortality Risk Calibration

Risk Graph Neural Networks (RGNNs) optimize Distributed Lag Non-linear Model (DLNM) coefficients using hierarchical GNN encoders and granular census features. In evaluations across England and Wales, RGNNs maintained lower point-errors and better uncertainty coverage during extreme heat events compared to baseline models.