RESUMEN
Early warning systems for vector-borne diseases (VBDs) prediction are an ecological application where data from the interface of several environmental components can be used to predict future VBD transmission. In general, models for early warning systems only consider average environmental conditions ignoring variation in weather variables, despite the prediction from Schmalhausen's law about the importance of environmental variability for biological systems. We present results from a long-term mosquito surveillance program from Harris County, Texas, USA, where we use time series analysis techniques to study the abundance and West Nile virus (WNV) infection patterns in the local primary vector, Culex quinquefasciatus Say. We found that, as predicted by Schmalhausen's law, mosquito abundance was associated with the standard deviation and kurtosis of environmental variables. By contrast, WNV infection rates were associated with 8-month lagged temperature, suggesting environmental conditions during overwintering might be key for WNV amplification during summer outbreaks. Finally, model validation showed that seasonal autoregressive models successfully predicted mosquito WNV infection rates up to 2â¯months ahead, but did rather poorly at predicting mosquito abundance, a result that might reflect impacts of vector control for mosquito population reduction, geographic scale, and other artifacts generated by operational constraints of mosquito surveillance systems.