Quantifying stochastic error propagation in Bayesian parametric estimates using non-linear parameters of Anopheles gambiae s.l. habitats

Reliable models of the transmission intensity of malaria, based on vector mosquito aquatic habitat larval productivity, are urgently needed, especially in endemic areas of Sub-Saharan Africa (SSA). Such models are fundamental for estimating the scale of the problem, and, hence, the resources needed to combat malaria in urban environments. These models also provide benchmarks for assessing the progress of control and indicate the geographical regions that should be prioritized. In this research, individual urban aquatic habitats of Anopheles gambiae s.l., a major malaria vector in SSA, were examined in terms of their spatial covariations by modelling ecologically sampled predictor variables within a Bayesian framework. Field sampling was conducted in two urban environments in Kenya, from July 2005 to December 2006. QuickBird satellite data, encompassing visible and near-infrared (NIR) bands, were selected to synthesize images of An. gambiae s.l. aquatic habitats. Statistical Analysis Software (SAS) was used to explore univariate statistics, correlations and distributions, and to perform Poisson regression analyses. These preliminary tests showed good type I error control mechanisms and precise parameter estimates. The model coefficients were then used to define expectations for prior distributions in a Markov chain Monte Carlo (MCMC) analysis. By specifying coefficient estimates in a Bayesian framework, depth of habitat was found to be a significant predictor, positively associated with urban An. gambiae s.l. aquatic habitats. There was no significant autocorrelation present in either the residual error or the predictor variable depth of habitat.