P-12 - Predicting Aedes vexans Populations Using Artificial Intelligence and Hyperlocal Weather Observations in Saginaw Michigan

Abstract: Accurate prediction of mosquito populations is crucial for effective mosquito control operations. In this study, we employed the XGBoost machine learning model to predict populations of Aedes vexans using mosquito trap data from the Saginaw County Mosquito Abatement Commission (SCMAC) and hyperlocal weather observations provided by Precip. The trap data, collected from 25 unique locations, included 669,015 observations spanning from 2012 to 2023. Weather observations were derived from Precip’s ultra-high-resolution weather models, offering daily precipitation and temperature data at an 800-meter resolution.



Our model integrated key environmental variables, including daily precipitation totals, minimum and maximum temperatures, and species-specific data. The predictive model was trained using a temporal domain approach, incorporating cross-validation to ensure accuracy. Results demonstrated that the model achieved a correlation coefficient (R²) of 0.36 for Ae. vexans, identifying precipitation and temperature as primary drivers of mosquito population dynamics.



The study highlights the potential for AI-based models to enhance mosquito surveillance programs by predicting species-specific population trends. By leveraging hyperlocal weather data, agencies can optimize intervention strategies and resource allocation, ultimately improving public health outcomes.