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.

system. There were 4730, 1113, 6056 and 6072 trees at sites B1, B2, D1 and D2, respectively. Each site was visited by teams of inspectors at successive intervals between October 1997 and October 1999. The locations of infected trees were identified and notional infection times were calculated by experienced personnel, from lesion size and other phenotypic characters. In order to account for errors in the assessment, the notional times were then grouped into 24 successive 30-day intervals (effectively used as censoring intervals for the true infection times). The data therefore provide spatial snapshots of the locations of susceptible and infected trees at successive 30-day intervals (see examples in Figures 1B,C) . The incidence of disease increased rapidly at all sites during the first 18 intervals with little infection thereafter coincident with the onset of dry conditions ( Figure 1D ). Further details of the collection of data are given in [28] . Disease was present in the area surrounding the census sites during the outbreak, with both susceptible and infected citrus trees between the sites ( Figure 1A ; see Figure S10 for a density map of citrus trees in the area). The data for an isolated small fifth site, also enumerated by the Agricultural Research Service (ARS) of the USDA, with a very small spread of infection around a single focus of three trees, are not analysed here because the small size of the outbreak precluded rigorous analysis. The effects of ingress of inoculum from infected trees outside the sites were incorporated into the rates for primary infection. Hence, for the purposes of the analyses, in this paper each site was treated as an independent subpopulation subject to external inoculum, and parameters were assumed to be independent among sites.