The predictability of mosquito abundance from daily to monthly timescales.

Journal Article

The prediction of mosquito abundance is of central interest in addressing mosquito population dynamics and in forecasting the associated emerging and re-emerging diseases. However, little work has focused on the systematic evaluation of how well adult mosquito abundance can be predicted as a function of observational resolutions, aggregation scales, and prediction lead time. We use a state space reconstruction (SSR) approach to compare the predictability of mosquito population dynamics at weekly, biweekly, and monthly scales. We focus on the analysis of Aedes vexans and Culiseta melanura populations monitored in Brunswick County (North Carolina, USA) and find that prediction over a 7-d lead time is improved when daily observations are used, compared to the commonly used once-per-week sample. Our results demonstrate that daily observations of mosquito abundance contribute to improving mosquito predictability in two ways: (1) daily observations better capture fluctuations over short timescales, which are missed when sampling at coarser resolutions, and (2) the aggregation of daily abundance observations reduces the impact of noise, thereby increasing the predictability of mosquito population dynamics as the aggregation scale is increased. We show that the evaluation of population dynamical models based on observed and predicted abundance can lead to a spuriously high apparent performance, due to the high autocorrelation in the observations used to update the model state at each successive time step. We show that the comparison of predicted and observed population change, expressed through per capita growth rates, leads to a more informative performance measure.

Full Text

Duke Authors

Cited Authors

  • Jian, Y; Silvestri, S; Brown, J; Hickman, R; Marani, M

Published Date

  • December 2016

Published In

Volume / Issue

  • 26 / 8

Start / End Page

  • 2609 - 2620

PubMed ID

  • 27865031

International Standard Serial Number (ISSN)

  • 1051-0761

Digital Object Identifier (DOI)

  • 10.1002/eap.1405

Language

  • eng