For both WNV meningitis and encephalitis incidence rates and total WNV disease incidence rates, we found regional differences in the associations between the proportion of specific land cover types and human WNV disease incidence. In contrast to the Northeast, where we found the proportion of developed land cover to be positively associated with human WNV disease incidence, in the Western part of the country, human WNV disease incidence was positively associated with the proportion of grassland and crop land cover. Associations at the national extent were disproportionately affected by associations in the Western part of the country; this is shown by similarities in the associations in the Western regions and the national extent in contrast with the Eastern regions and the national extent.
Disease vectors are often associated with landscape because the abundance and distribution of vectors are determined by the environment.27
In this study, we used 14 land cover classes15
to characterize the landscape in different regions of the United States. West of the Mississippi River, the main mosquito vector of WNV is Cx. tarsalis
, which breeds in standing water that receives ample sunlight, such as in savannas or grasslands.28–30
The prominent WNV vector that is abundant on the East coast is Cx. pipiens
, which is found more commonly in suburban to urban areas, mostly because it breeds in artificial containers that are often polluted or eutrophic. This species can reproduce in storm sewers and sewage treatment plants as well as in ditches and other drainage facilities that are nutrient-heavy.28–30 Cx. pipiens
is also found in Western urban areas; however, the majority of infections in the Western United States are in predominantly rural counties.
When we compared the preferred habitat of these mosquito species with the WNV disease-promoting land cover types in the regions in which the species are located, we found striking similarities: Cx. pipiens prefers urban polluted habitats and is located in regions that showed positive associations between human WNV disease incidence and urban land cover, whereas Cx. tarsalis prefers natural ground pools and is located in regions that showed positive associations between incidence and agricultural land cover. Because of the similarities between vector habitat and WNV disease-promoting land covers in these regions, we speculated that a regional difference in the abundance of these species is the determining factor behind the associations that we found. This conjecture was confirmed by the results of our GAMs, which show only a 0.03% difference in the deviance explained between the two models. When we included both land cover type and mosquito species presence as predictors, the model explained 80.4% of the variation in our data. However, when we include only mosquito species presence as a predictor, the model still explains 80.1% of our data, indicating that the effect of land cover type on human disease risk is primarily mediated by its effect on the vector community.
To our knowledge, no previous studies have considered land cover correlations with WNV disease incidence between regions. However, some studies that looked at specific cities or regions have yielded results similar to ours. For instance, a positive association with agricultural land and negative associations with wetlands, forests, and developed land have been found previously in the Upper Plains.5
When looking at equine WNV, deciduous forest was found to reduce the predicted infection rate in horses in the South Central region, and an increased human disease incidence and risk of death attributed to WNV was found to be associated with crop land cover in the same region.6,9
A study in the Great Lakes region also found negative associations with agricultural land, wetland, and forest cover and a positive association with fractionated habitat, which would be comparable with our developed land cover classifications.4
In the Mid-South, a model identified the ideal conditions for WNV vector mosquitoes to be at a Normalized Difference Vegetation Index (NDVI) value > 0.30 (indicating moderate vegetation cover).31
This supports the positive correlation between crop land cover and human WNV disease incidence that we found in the same region, because WNV vector mosquitoes are projected to be found in the same land cover type. A study in New England also reported a positive association between urban land cover and human WNV disease incidence and a negative association between forested land cover and human disease incidence.3
In the Mid-Atlantic, it was found that WNV antibodies in several mammal species increased as urbanization increased, consistent with the positive correlation that we report between WNV incidence and urban land cover in this region.8
In addition to findings concerning the association between WNV incidence and land cover, many studies also report the presence or abundance of Cx. pipiens
in Eastern regions or Cx. tarsalis
in Western regions.5,6,10,11,31–38
Supplemental Figure S2 has a range map of these species.24
A few studies have reported results that were contradictory to our findings. In the Northwest, a study found that the abundance of potential WNV vector species increased with increasing urbanization.36
This study concentrated on the urban center of Seattle, and of the 26 sites sampled, only 4 were not classified as urban or suburban. This uneven distribution of sites along an urban gradient could contribute to the apparent contradiction with our results. In the Southwest, higher infection rates of primary mosquito vectors were found in urban areas.11
However, high mosquito infection rates do not necessarily lead to high human disease infection rates, especially if high densities of other appealing hosts are also present in urban areas and therefore, mosquitoes are not feeding on humans (the dilution effect).13
Another possible explanation is that increased mosquito infection rates are masked by even larger increases in the human population. When looking at equine WNV disease, a negative correlation with crops was reported in the South Central region. This study used a different land cover dataset than was used for our analyses, which could have resulted in differences in land cover classifications between the two datasets.9
In the Deep South, two different studies found increased seroprevalence in birds as urbanization increased.7,10
These two studies did not use a land cover dataset to classify urban or suburban land use in their respective studies: one used an urban score based on human population density and forest cover,7
whereas the other used field observations to make land use classifications.10
Therefore, it is possible that what they classified as urban on a small scale may not have been classified as urban in the NLCD 2001 dataset, which provides uniform land cover classifications across the United States.
In summary, we found strong evidence that human WNV disease associations with land cover differ at the regional extent. Our study shows that urban land cover is positively associated with human WNV disease incidence in Northeastern regions, whereas agricultural land, such as grassland and crops, is positively associated with disease incidence in the Western part of the country. Analyses incorporating peak year of human disease infection, as well as WNV meningitis and encephalitis disease incidence, supported these associations. Because of the different habitats present in these positive associations, we suspect that regional differences in the community of WNV-competent mosquitoes are the probable drivers behind the correlations that we report. Research aimed at succinctly defining the ranges of these mosquito species in the United States would be beneficial in further determining the validity of our speculation.
Our results suggest that the relationship between land cover composition and human WNV disease incidence should not be generalized at the national extent, and they provide supporting evidence for regional vector control programs. In short, WNV disease is not simply urban or suburban but is a disease whose transmission processes are embedded in an ecological–epidemiological system that varies across large-scale environmental gradients.