Analysis of Fire Blight Shoot Infection Epidemics on Apple

Abstract

Fire blight incidence and spread of the shoot blight phase of the disease was studied in four apple cultivars in replicated blocks over 4 years (1994 to 1997). Cv. York was highly susceptible, followed by ‘Fuji’ and ‘Golden Delicious,’ which were moderately susceptible, and ‘Liberty,’ which was least susceptible. On York, the first appearance of shoot blight was within 48 h of its predicted appearance according to the Maryblyt model in 3 of the 4 years studied. Shoot blight epidemics in York in 1995 and 1996, and Fuji in 1995, were best described with a logistic model that showed apparent infection rates ranging from 0.05 to 0.20, indicating a low to moderately high rate of disease increase. The spatial positions (row and column) of all infected plants in each subplot were recorded on plot maps on each sampling date. The binomial and β-binomial distributions were fit to the data to test for spatial aggregation of disease incidence for each cultivar plot. Maximum likelihood estimation was possible for 92 (43.6%) of the 211 data sets subjected to this analysis. Of these, 35 data sets were better fit by the β-binomial distribution than the binomial distribution. The binary power law was used to characterize the relationship between the variance among quadrats within each plot to the variance expected for that plot given the observed level of disease incidence. The binary power law provided an excellent fit to the full data set and to nearly all of the subsets and, with b > 1, indicated that heterogeneity changed systematically with disease incidence. A covariance analysis was conducted to determine the effect of the factors ‘year,’ ‘cultivar,’ ‘orchard plot,’ and ‘observation date’ on the intercept and slope parameters of the binary power law. In general, plot followed by year had the greatest impact on parameter estimates and is an indication that location and seasonal factors impact heterogeneity of disease, although the specifics could not be ascertained from this study. Ordinary runs analysis was used to analyze the pattern of diseased trees within rows and detected significant nonrandom patterns of disease incidence in 63.5% of the orchard plots over the 4-year study. From these data sets, 68.7% had significantly fewer runs, particularly at disease incidences greater than 0.1. The fewer-than-expected runs at incidences greater than 0.10 provides strong evidence of localized spread.