Forecasting of dengue importation risk in Brazil using deep learning and mobility networks

Timely anticipation of dengue spread is essential for public health planning, particularly in large endemic countries where transmission is shaped by both local conditions and human mobility. While most early warning systems focus on local forecasts, they often neglect how cases in one location may influence risk elsewhere through spatial movement. Here, we present a generalizable framework that integrates short-term dengue incidence forecasts with a multimodal mobility network to estimate the relative importation pressure index between all 5,570 municipalities in Brazil. By coupling a climate-informed long short-term memory (LSTM) model with a composite mobility matrix spanning road, river, and air transport, we generate dynamic, city-level surfaces of predicted importation risk for 2024. Results reveal spatially structured corridors of transmission, strong asymmetries between source and sink cities, and wide variation in states’ dependence on internal versus external seeding. This framework advances dengue surveillance by capturing spatial spillover effects in real time and offers a modular architecture for forecasting importation risk across diseases and settings.