Local impact of temperature and precipitation on West Nile virus infection in Culex species mosquitoes in northeast Illinois, USA

Abstract

Background: Models of the effects of environmental factors on West Nile virus disease risk have yielded conflicting outcomes. The role of precipitation has been especially difficult to discern from existing studies, due in part to habitat and behavior characteristics of specific vector species and because of differences in the temporal and spatial scales of the published studies. We used spatial and statistical modeling techniques to analyze and forecast fine scale spatial (2000 m grid) and temporal (weekly) patterns of West Nile virus mosquito infection relative to changing weather conditions in the urban landscape of the greater Chicago, Illinois, region for the years from 2004 to 2008.

Results: Increased air temperature was the strongest temporal predictor of increased infection in Culex pipiens and Culex restuans mosquitoes, with cumulative high temperature differences being a key factor distinguishing years with higher mosquito infection and higher human illness rates from those with lower rates. Drier conditions in the spring followed by wetter conditions just prior to an increase in infection were factors in some but not all years. Overall, 80% of the weekly variation in mosquito infection was explained by prior weather conditions. Spatially, lower precipitation was the most important variable predicting stronger mosquito infection; precipitation and temperature alone could explain the pattern of spatial variability better than could other environmental variables (79% explained in the best model). Variables related to impervious surfaces and elevation differences were of modest importance in the spatial model.

Conclusion: Finely grained temporal and spatial patterns of precipitation and air temperature have a consistent and significant impact on the timing and location of increased mosquito infection in the northeastern Illinois study area. The use of local weather data at multiple monitoring locations and the integration of mosquito infection data from numerous sources across several years are important to the strength of the models presented. The other spatial environmental factors that tended to be important, including impervious surfaces and elevation measures, would mediate the effect of rainfall on soils and in urban catch basins. Changes in weather patterns with global climate change make it especially important to improve our ability to predict how inter-related local weather and environmental factors affect vectors and vector-borne disease risk.Local impact of temperature and precipitation on West Nile virus infection in Culex species mosquitoes in northeast Illinois, USA.

Figure 1

Study area in the greater Chicago, Illinois. General area of mosquito trap locations for 2004 to 2007 (grey shaded areas) and locations of all weather stations with data for the years under analysis are included. Cook county and DuPage county are the area of analysis, while data from the five surrounding counties provided additional support for the spatial interpolation of variables. Mosquito trap locations are shown as gray areas to be illustrative of the general pattern of the traps, rather than to show specific locations, which varied somewhat from year to year. Of the 1263 hexagons (not shown) used for the spatial analysis, 409 (32%) had a trap location in them in at least one of the years.

Figure 2

Mosquito infection by week, 2004 to 2008, in Cook and DuPage counties in Illinois. The values reflect the minimum infection rate calculations averaged from the 2000 m hexagons used in the regression models.

Figure 3

Correlation between weekly MIR and precipitation. Weekly values are averaged on 1, 3 and 5 weeks at different lags during each year from 2004 to 2007 in Cook and DuPage counties, Illinois. There are 18 weeks with 15 possible temporal lags for each year.

Figure 4

Weekly Differences in Degree Weeks. Cumulative Degree Week values defined at Tbase = 22°C and portrayed as the difference from average for the years from 2002 to 2008.

Figure 5

Estimated and observed mosquito MIR from 2002 to 2008. Lines shown are from observations (solid line) and simulations (dashed line) for Model 1 and Model 4. Only MIR between 2004 and 2007 were simulated by model 4, and MIR in 2008 was only simulated by model 1.

Figure 6

Mosquito infection measured from observed and spatial model estimated data. The mosquito infection was measured for weeks 32 to 34, 2005, and the infection rate predicted by the random forest model using all variables (right) for the same time period.