Dengue fever is one of the most common and rapidly spreading arboviral diseases in the world, with major public health and economic consequences, especially in tropical and sub-tropical regions. The World Health Organization (WHO) reports that the global burden of Dengue increased eight-fold over the last two decades. The highest number of Dengue cases was recorded in 2023, affecting over 80 countries with over 6.5 million cases and more than 7,300 dengue-related deaths reported. While Dengue cases occur worldwide, the recent acceleration is especially prevalent in Latin America and Asia, exposing an increasing amount of the world's population.

Early warning systems at regional and micro scales are critical for emergency managers and decision makers to facilitate data-driven distribution of resources and disease intervention strategies to minimize the impact on the population. Artificial Intelligence (AI) and Machine Learning (ML) techniques can be leveraged to improve our ability to predict outbreaks and manage and prioritize public health interventions. Furthermore, remote sensing and other geospatial data hold immense potential in monitoring environmental conditions conducive for dengue incidence. Temperature, rainfall, humidity, elevation, and land use/ land cover all play major roles in the mosquito population dynamics that influence dengue transmission. Geospatial analytics and ML modeling can combine these environmental data with population characteristics and dengue monitoring to provide predictive assistance to local disease managers.

Our team has developed the Disease Incidence and Resource Estimator (DIRE), which provides decision-makers a geospatial predictive tool for imminent dengue epidemics, as well as a recommendation of health resources required to control and treat diseases. We employ free and open-source geospatial tools to ingest and analyze the environmental, climatological and demographic conditions conducive to dengue development, construct ML models to predict the next month of dengue cases at a variety of administrative levels, and serve emergency managers with a web application to map and analyze dengue predictions and resource needs. Funded by the Wellcome Trust, this tool is a combined effort between The University of California San Diego, New Light Technologies, and UNICEF, and leverages an ensemble ML approach developed by the European Space Agency (ESA) and UNICEF.

Employing open-source geospatial, modeling, and visualization tools has been critical as we construct this into a stable long-term solution for the public and nonprofit sectors. Our pipeline leverages urllib, BeautifulSoup, and cdsapi to acquire environmental data from NASA and ESA, and employs geopandas, rasterio, and rasterstats to analyze these data at each administrative level. The tensorflow, sklearn, and catboost libraries power a ML model for dengue incidence that includes CNN, LSTM, CatBoost, and SVM elements in a RandomForest ensemble. The web application is built using React, MapLibre, deck.gl, and Turf.js to allow emergency managers to interrogate both the spatial and temporal patterns in emerging dengue outbreaks.