The natural reservoir for West Nile Virus (WNV) is in songbirds. Bird species vary in their competence for transmitting the virus to uninfected mosquitoes, but the American robin is known to be highly competent. Culex pipiens, the primary vector for WNV in Connecticut, has been shown to feed almost exclusively on birds, demonstrating a preference for robins over other songbirds (Molaei et al. 468-474; Ngo and Kramer 215-222). This fact combined with robins’ high competence as a host makes them a crucial element in the amplification of West Nile Virus in suburban and urban environments.
Robin abundance may be a useful predictor variable for WNV risk. Bird species composition varies with respect to land use. Robins are a dominant species in suburban areas, while more urban areas are dominated by invasive omnivorous species such as the European house sparrow and the European starling (Clergeau et al. 413-425). It would be useful for us to understand the point at which the dominant species switches, and what factors are responsible for it.
Preliminary data suggest a negative correlation between robin abundance and WNV prevalence, which seems counterintuitive at first glance. We hypothesize that along the increasing urbanization gradient, although robin population size decreases, the few remaining robins are heavily concentrated together within small undisturbed forested areas. The mosquitoes that feed on birds are also condensed into the small vegetation patches, resulting in higher “encounter rates” despite the fact that robin populations are lower. This could partially explain why WNV incidence in humans is higher in urban areas than in the suburbs.
Methods
I will use 15m resolution ASTER images focused on southwestern Connecticut to conduct several unsupervised classifications, and use IKONOS aerial photos for ground truthing-and will conduct some ground truth visits manually as time allows. These classifications will be compared to the 1992 NLCD land use classification and also compared to one another. I will look for classes that are correlated with robin abundance as determined by on our bird count data. I also want to find a meaningful way to consider the texture of developed areas in the model. Several texture filters will be applied and tested for predictive value.