Jacob BG, Morris JA, Caamano EX, Griffith DA, Novak RJ. Geomapping generalized eigenvalue frequency distributions for predicting prolific Aedes albopictus and Culex quinquefasciatus habitats based on spatiotemporal field-sampled count data.
Acta Trop 2011;
117:61-8. [PMID:
20969828 DOI:
10.1016/j.actatropica.2010.10.002]
[Citation(s) in RCA: 6] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 05/24/2010] [Revised: 09/01/2010] [Accepted: 10/08/2010] [Indexed: 10/18/2022]
Abstract
Marked spatiotemporal variabilities in mosquito infection of arboviruses require adaptive strategies for determining optimal field-sampling timeframes, pool screening, and data analyses. In particular, the error distribution and aggregation patterns of adult arboviral mosquitoes can vary significantly by species, which can statistically bias analyses of spatiotemporal-sampled predictor variables generating misinterpretation of prolific habitat surveillance locations. Currently, there is a lack of reliable and consistent measures of risk exposure based on field-sampled georeferenced explanatory covariates which can compromise quantitative predictions generated from arboviral mosquito surveillance models for implementing larval control strategies targeting productive habitats. In this research we used spatial statistics and QuickBird visible and near-infra-red data for determining trapping sites that were related to Culex quinquefasciatus and Aedes albopictus species abundance and distribution in Birmingham, Alabama. Initially, a Land Use Land Cover (LULC) model was constructed from multiple spatiotemporal-sampled georeferenced predictors and the QuickBird data. A Poisson regression model with a non-homogenous, gamma-distributed mean then decomposed the data into positive and negative spatial filter eigenvectors. An autoregressive process in the error term then was used to derive the sample distribution of the Moran's I statistic for determining latent autocorrelation components in the model. Spatial filter algorithms established means, variances, distributional functions, and pairwise correlations for the predictor variables. In doing so, the eigenfunction spatial filter quantified the residual autocorrelation error in the mean response term of the model as a linear combination of various distinct Cx. quinquefasciatus and Ae. albopictus habitat map patterns. The analyses revealed 18-27% redundant information in the data. Prolific habitats of Cx. quinquefasciatus and Ae. albopictus can be accurately spatially targeted based on georeferenced field-sampled count data using QuickBird data, LULC explanatory covariates, robust negative binomial regression estimates and space-time eigenfunctions.
Collapse