Wavelet Analysis as a Non-Stationary Approach to Validate a Simulated Mosquito Model

Abstract

IntroductionThe Arizona Department of Health Services, Pinal County Public Health Services District, and the University of Arizona collaborated to reform current surveillance practices for mosquitos and West Nile Virus (WNV) with strategies that are resilient to changes in climate. The Dynamic Mosquito Simulated Model developed by Morin and Comrie (2010) uses daily precipitation and temperatures to estimate mosquito abundance and was adapted and validated by Brown et al. (2015) for both Culex quinquefasciatus and Cx. tarsalis mosquito vectors. We sought to verify that the simulated mosquito model can be used as a reliable substitute for observed data, and that wavelet analysis can be applied to verify simulated model data as a substitute or supplement for observed data. To our knowledge, wavelet analysis has rarely been used with WNV data and none have used wavelet analysis to validate simulated mosquito abundance data. MethodsData were collected in Pinal County from 2012 to 2018 and restricted to 2015 to 2018 when trap locations were sampled greater than 5 nights per year for analysis. Daily weather data time series values were collected from the PRISM Climate Group at Oregon State University. The dataset was analyzed using the Wavelet_EETS wavelet analysis program developed by Dr. Cazelles and Dr. Chavez (2003) for time series, power spectrum, global spectrum, and coherency plots. ResultsTime series plots of the simulated and observed mosquito abundance show similar patterns. The simulated mosquito season begins prior to and ends after the observed season and except for the beginning of 2018 when the observed season began later in May, the observed season begins within two weeks of late-March and ends within two weeks of early-November. An extended simulated season compared to observed season was noted around November 2017 for both species, which corresponded with warmer than average temperatures compared to 2016. Comparison of wavelet power spectrums and wavelet coherency for the observed and simulated data show similar spectral patterns at 1-year periods. The plot of the phase differences between the two time series demonstrates an average lag period of 3-6 weeks between season changes of the observed and simulated data, with more variability in phase differences noted for Cx. tarsalis. DiscussionThe similar patterns in the time series, power spectrum, and global spectrum plots as well as strong association in coherency analyses demonstrate that the simulated mosquito abundance model can reliably be used as a substitute or supplement for observed data, validating it’s use as a meteorologically-based early warning system for the Pinal County region and surrogate for surveillance data in further analyses. Based on the extended simulated season and average instantaneous lag, the recommended extension for surveillance trapping is from mid-February to mid-December.