Forecasting Dengue Importation Risk in Brazil Using Deep Learning and Multimodal Mobility Networks

Dengue continues to expand across Brazil under increasingly variable climatic conditions, and anticipating where infections may spread is essential for effective public health preparedness. However, most existing early warning systems focus on local case trajectories alone and overlook the spatial redistribution of infection risk driven by human mobility. This gap leaves planners without the ability to foresee where cases are likely to be imported before local transmission accelerates.

In this study, we develop a generalizable forecasting framework that couples climate-informed dengue incidence predictions with a multimodal mobility network covering all 5,570 Brazilian municipalities. Weekly dengue cases are forecasted using a long short-term memory (LSTM) model that incorporates temperature and humidity dynamics. These forecasts are combined with a composite mobility matrix spanning road, river, and air flows, allowing us to estimate the expected volume of imported infections between cities for every epidemiological week of 2024.

The resulting importation-risk surfaces reveal well-defined corridors of movement-mediated dengue spread, including strong directional asymmetries between major source regions (e.g., large urban hubs with intense outbound flows) and peripheral sink municipalities that depend heavily on external seeding. We find that high importation risk often precedes subsequent local increases in incidence, highlighting the added value of capturing human mobility in early warning systems.

This framework advances dengue surveillance by integrating climate variability, human mobility, and short-term predictive modeling into a unified pipeline. Beyond dengue and Brazil, the approach is modular and transferable to other climate-sensitive infectious diseases and mobility-rich settings. By quantifying how infections may spread through movement pathways before they emerge locally, this work provides a scalable tool for proactive, spatially targeted public health response in an era of intensifying climate-health risks.