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.

The posterior distributions for the dispersal kernel (a), transmission rate (b), and the ingress of external inoculum (e) are summarised in Figure 2 for one of the sites (B2) in Broward county. The results show the sensitivity of the posterior distributions of the parameters to the observation time window (cf. Table 1 ); similar results were obtained for all four sites. Initial inferences were done for cumulative windows (model M cum , Table 1 ), in which successively more monthly snapshots of the locations of infected and healthy trees were added. These results show how the availability of additional information during the epidemic affects the precision of the parameter estimates (Figure 2A) . The estimate for a is remarkably robust. There is a short, initial transient period (0-3 30-day periods) for which the parameter is not well estimated, by the end of which there are fewer than 21/1113 infected trees. Later estimates were remarkably close both in expectation and precision, with no further gain in precision after six months (Figure 2A) , when 69/1113 trees were recorded as infected.