Predicting mosquito population with high-resolution inundation in California: implications for West Nile Virus

West Nile virus (WNV), a globally distributed mosquito-borne pathogen, infects hundreds of people per year in California, posing a significant public health threat to the state and its residents. Vectors of WNV in California depend on aquatic habitats for reproduction, refuge, and food resources, suggesting a potentially strong linkage between aquatic landscapes and vector abundance that affects the risk of WNV transmission. However, the contribution of surface inundation (i.e., standing water on the land surface) that could serve as larval habitat to WNV vector abundance has been challenging to characterize because spatially continuous, high-resolution, and long-term inundation data are currently unavailable. Taking the Central Valley of California as an example, we introduce a novel data fusion approach that leverages observations (Global Surface Water Dataset of the Joint Research Center, NHDPlus-HR, CropScape, National Elevation Dataset) and model output (Variable Infiltration Capacity model irrigation module) to generate monthly, high-resolution (10 m) estimates of inundation in the Central Valley from 2007 to 2016. We then apply a random forest modeling framework using the inundation product and other environmental covariates (temperature, precipitation, and land cover types) to predict the WNV vector abundance across the Central Valley. Using this random forest modeling framework, we further quantify the effect of irrigation on vector abundance for different hydrologic conditions (dry, moderate, and wet years) by comparing the original prediction with a counterfactual prediction after removing the effect of inundation in irrigated fields and ditches. We show that inundation is a stronger predictor of WNV vector abundance than precipitation, and the presence of nearby irrigation infrastructure, such as canals and ditches, substantially increases vector abundance in the Central Valley.