Geo-referenced spatiotemporal analysis of the urban citrus canker epidemic in Florida

ABSTRACT Five areas in urban Miami were identified to study the spread of Xanthomonas axonopodis pv. citri to determine if the practice of removing exposed citrus trees within 38.1 m of trees affected by citrus canker was adequate to curtail further bacterial spread. To accomplish this, 18,769 trees in dooryards were surveyed, geo-referenced by differential global positioning systems (GPS), and assayed for disease severity, age of infection, citrus cultivar, location of infection in tree, and canopy size. For each tree, the date the tree became infected was estimated and used to separate trees into contiguous 30-day categories. For each area studied, distance measurements between focal trees and newly infected trees were calculated for various temporal periods of 30, 60, 90, and 120 days in duration, corresponding to intervals of inspection survey. A visual basic application was used to calculate the distances between each newly diseased tree and all prior focal trees. The nearest distance was used because it was considered the most conservative estimate possible. It is therefore likely to be an underestimate of spread but is a good estimate of the minimum possible distances of spread. For the first four 30-day periods among the five study areas, calculated maximum distances of spread ranged from 12 to 3,474 m, indicating a broad continuum of distance for bacterial spread was possible. Disease increased during the first two-thirds of the time studied and reached an asymptote due to dry conditions in the final one-third of the duration of the study. Cross correlation analysis indicated that disease was best visualized 107 days following rainstorms with wind. Analysis of regional spatial point patterns was performed temporally for each 30-day period via a modified Ripley's K-function. Spatiotemporal analyses between periods over areas larger than previously examined were accomplished via spatiotemporal semivariogram analysis. These methods in combination demonstrated rapid increases in range of spatial dependency and range of spatiotemporal dependency for all study sites. This corresponded to rapid spread of disease across the regions studied in response to rainstorms with wind followed by a "filling in" of disease on remaining noninfected susceptible trees through time by less intense rain events. A stochastic quadratization technique demonstrated that disease incidence and disease severity were not greatly affected by urban host density but were positively correlated to host susceptibility within local 0.25-km(2) quadrats.