P20 - Predictive Modeling of Aedes aegypti Abundance Using Environmental and Spatial Data in Southern California, 2017–2023

Abstract: Aedes aegypti, the primary vector of dengue, Zika, and chikungunya viruses, has rapidly expanded across Southern California, including semi-arid areas once considered unsuitable. This spread increases the risk of local arboviral transmission and highlights the need for timely, data-driven forecasting of mosquito populations to support targeted control.

We analyzed seven years (2017–2023) of Ae. aegypti surveillance data from the West Valley region of San Bernardino County, integrating environmental variables from remote sensing. Predictor variables included temperature, precipitation, surface water, elevation, built area, and seasonal time terms. Three predictive models were developed and compared: (1) a Generalized Additive Model (GAM) to capture nonlinear relationships while maintaining interpretability; (2) a Random Forest (RF) to model complex interactions; and (3) a spatial Random Forest (RFsp) incorporating a Gaussian spatial kernel to account for spatial autocorrelation. Models were trained on 2017–2022 data and validated using 2023 observations.

Mosquito abundance increased sharply from 2017 to 2023, with consistent seasonal peaks in late summer and early fall. Densely built urban areas supported higher abundance, emphasizing the role of human-modified environments in sustaining populations despite arid conditions. Among models, the RF achieved the best predictive performance (MAE = 15.57, RMSE = 29.44), while the GAM identified significant nonlinear effects of temperature, precipitation, and seasonality. The RFsp improved local hotspot detection; however, spatial prediction accuracy was limited by the shifting hotspots over time.

Overall, our results demonstrate that machine learning and spatially explicit models can complement traditional analyses to enhance forecasting and identifying mosquito hotspots. This modeling framework can help districts identify mosquito receptive areas, forecast seasonal surges, and optimize interventions such as sterile insect technique (SIT) releases or targeted IVM in high-risk areas.