Bayesian Analysis for Inference of an Emerging Epidemic: Citrus Canker in Urban Landscapes

Neri, Franco M.; Cook, Alex R.; Gibson, Gavin J.; Gottwald, Tim R.; Gilligan, Christopher A.

We also used the analyses to infer what effects additional snapshots or different starting times for data collection would have had on epidemic predictions. For subsequent analyses, we introduce a classification of the models (Table 1 ) based upon the temporal window used for the estimation (with no reference to the specific form of the dispersal kernel) and the number of parameters used. The original three-parameter model, fitted to the entire dataset, will be denoted with M 0 . Cumulative windows (model M cum in Table 1 ) were used to identify the effect of recording more and more snapshots over time on the parameter estimates, by deriving estimates based upon snapshots for 0-3, 0-6, … 0-24 30-day intervals. Sliding windows, for example 0-6, 3-9, …12-18 30-day intervals (model M DT slid in Table 1 , with DT equal to the window width), were used to assess the effects of different starting times for data collection and fixed periods of observation on parameter estimates (hence, they represent scenarios for later detection and initiation of data collection).