<p>This study develops an appropriate model for forecasting age-specific mortality rates at regional and prefectural-levels in Japan. We modified existing mortality models, including the Lee–Carter (LC) and Li–Lee (LL) models, to incorporate spatial sensitivity within a Bayesian framework. We also addressed sudden changes in mortality rates caused by disasters and pandemics, which is particularly relevant to some of the considered models. Our findings highlight that incorporating spatial effects enhances the model’s fit and provides robust predictions in during abrupt mortality changes. This improved performance is crucial when dealing with unexpected events that can significantly impact mortality rates. Furthermore, our analysis underscores the importance of spatial modelling in capturing regional disparities and improving the accuracy of mortality forecasts. The study’s results demonstrate the advantages of using models with spatial effects when analysing limited multi-population data. By accounting for geographical variations, these models offer a more nuanced and precise approach to mortality rate prediction. This advancement in modelling techniques is invaluable for demographic research, public health planning, and policy-making at both regional and prefectural-levels in Japan.</p>