2026 年 41 巻 1 号 p. 49-54
Identifying appropriate spatial structures in epidemic models is challenging when direct mobility data are limited, whereas incidence time series are routinely available. Using COVID-19 data from five Japanese prefectures, we examined whether the temporal shape of incidence curves can reveal inter-regional diffusion patterns. We compared three spatial SIR formulations—no diffusion, uniform diffusion, and heterogeneous region-specific diffusion—across five epidemic waves. Model selection using AIC revealed strong wave dependence: heterogeneous region-specific diffusion was favored for Waves I and III, uniform diffusion for Waves II and IV, and no diffusion for Wave V. Temporal descriptors including multi-modality, skewness, kurtosis, and peak-timing variability accounted for these differences, suggesting that incidence-curve morphology may serve as a practical diagnostic for selecting spatial epidemic models.
