Abstract
An empirical model to forecast West Nile virus mosquito vector populations is developed using time series analysis techniques. Specifically, multivariate seasonal autoregressive integrated moving average (SARIMA) models were developed for Aedes vexans and the combined group of Culex pipiens and Culex restuans in Erie County, New York. Weekly mosquito collections data were obtained for the four mosquito seasons from 2002 to 2005 from the Erie County Department of Health, Vector and Pest Control Program. Climate variables were tested for significance with cross-correlation analysis. Minimum temperature (T min), maximum temperature (T max), average temperature (T ave), precipitation (P), relative humidity (R H), and evapotranspiration (E T) were acquired from the Northeast Regional Climate Center (NRCC) at Cornell University. Weekly averages or sums of climate variables were calculated from the daily data. Other climate indexes were calculated and were tested for significance with the mosquito population data, including cooling degree days base 60 degrees (C DD_60), cooling degree days base 63 (C DD_63), cooling degree days base 65 (C DD_65), a ponding index (I P), and an interactive C DD_65-precipitation variable (C DD_65 × P week_4). Ae. vexans were adequately modeled with a (2,1,1)(1,1,0)52 SARIMA model. The combined group of Culex pipiens-restuans were modeled with a (0,1,1)(1,1,0)52 SARIMA model. The most significant meteorological variables for forecasting Aedes vexans abundance was the interactive C DD_65 × P week_4 variable at a lag of two weeks, E T × E T at a lag of five weeks, and C DD_65 × C DD_65 at a lag of seven weeks. The most significant predictive variables for the grouped Culex pipiens-restuans were C DD_63 × C DD_63 at a lag of zero weeks, C DD_63 at a lag of eight weeks, and the cumulative maximum ponding index (I Pcum) at a lag of zero weeks.