Evaluation of Location-Specific Predictions by a Detailed Simulation Model of Aedes aegypti Populations

BACKGROUND: Skeeter Buster is a stochastic, spatially explicit simulation model of Aedes aegypti populations, designed to predict the outcome of vector population control methods. In this study, we apply the model to two specific locations, the cities of Iquitos, Peru, and Buenos Aires, Argentina. These two sites differ in the amount of field data that is available for location-specific customization. By comparing output from Skeeter Buster to field observations in these two cases we evaluate population dynamics predictions by Skeeter Buster with varying degrees of customization. METHODOLOGY/PRINCIPAL FINDINGS: Skeeter Buster was customized to the Iquitos location by simulating the layout of houses and the associated distribution of water-holding containers, based on extensive surveys of Ae. aegypti populations and larval habitats that have been conducted in Iquitos for over 10 years. The model is calibrated by adjusting the food input into various types of containers to match their observed pupal productivity in the field. We contrast the output of this customized model to the data collected from the natural population, comparing pupal numbers and spatial distribution of pupae in the population. Our results show that Skeeter Buster replicates specific population dynamics and spatial structure of Ae. aegypti in Iquitos. We then show how Skeeter Buster can be customized for Buenos Aires, where we only had Ae. aegypti abundance data that was averaged across all locations. In the Argentina case Skeeter Buster provides a satisfactory simulation of temporal population dynamics across seasons. CONCLUSIONS: This model can provide a faithful description of Ae. aegypti populations, through a process of location-specific customization that is contingent on the amount of data available from field collections. We discuss limitations presented by some specific components of the model such as the description of food dynamics and challenges that these limitations bring to model evaluation.