Generative diffusion models for spatiotemporal influenza forecasting

Forecasting infectious disease incidence can provide important information to guide public health planning, yet it is difficult because epidemic dynamics are complex. Current mechanistic and statistical approaches often struggle to capture multimodal uncertainty or emergent trends. Influpaint adapts denoising diffusion probabilistic models to epidemic forecasting. By encoding influenza seasons as spatiotemporal images in which pixel intensity represents incidence, Influpaint learns a rich distribution of disease dynamics from a hybrid dataset of surveillance and simulated trajectories. Forecasting is formulated as a conditional generation (inpainting) task from partial observations. We show that Influpaint generates realistic, diverse epidemic trajectories and achieves forecast accuracy that is competitive with leading ensemble methods in retrospective evaluation. In real-time evaluation during the 2022–2025 U.S. CDC FluSight challenges, performance improved substantially across seasons, with highly accurate but somewhat overconfident projections in 2024–2025. The best performance was achieved with a training dataset containing 30% surveillance and 70% simulated trajectories. These results show that diffusion models can capture important spatiotemporal structure in influenza dynamics and provide a flexible framework for probabilistic infectious disease forecasting.Author summary: Accurately forecasting influenza activity is difficult, which complicates public health preparedness and response. We propose an influenza forecasting approach, named Influpaint, that treats an influenza season like an image, with weeks on one axis, locations on the other, and color intensity representing reported hospitalizations. This framing lets us use a type of generative artificial intelligence, the diffusion model, to learn the patterns of influenza activity and then predict (inpaint) future weeks from the part of the season already observed. We show that Influpaint generates plausible influenza seasons and produces forecasts that are competitive with established ensemble methods. The approach is also highly flexible and can be used to impute missing data without additional training. We also find that combining real surveillance data with simulated epidemic trajectories improves performance.