Traffic-driven spatio-temporal prediction for fine-scale epidemic outbreaks in China

As the high connectivity creates favorable conditions for the infectious diseases, understanding the dynamics of epidemic spread across regions is significant. However, existing models often focus on single dimensions and specific factors under strict controls, but generally overlook regional differences and interactions.

In this research, we constructed a spatio-temporal forecasting model based on multiple modeling architectures, specifically designed to predict the post-lockdown epidemic's transmission at a fine scale across prefecture-level cities nationwide. The model integrates comprehensive factors and constructs a structured graph based on transportation patterns and characteristic variables across municipal regions in China. To further enhance its ability to capture spatial interactions among regions, the model integrates human mobility into edge weight calculations to optimize the adjacency matrix.

We analyzed the prediction errors across different urban clusters, provinces, folds, and forecast durations, revealing spatial variations and consistent error reductions. The results demonstrate that our model can accurately predict the spatio-temporal spread of the epidemic across 309 Chinese prefecture-level cities with a high correlation coefficient (r = 0.94) validated through extensive cross-validation.

Our proposed approach integrates transportation patterns and human mobility into edge weight calculations to enhance spatial connectivity. This post-lockdown simulation across cities in China offers a fine-grained analytical scale previously unexplored, and the comprehensive analysis provides enhanced insights into epidemic dynamics and transmission patterns, supporting the future public health strategies.