J Gen Plant Pathol DOI 10.1007/s10327-014-0532-4
BACTERIAL AND PHYTOPLASMA DISEASES
Risk factors for bacterial spot on peach in Okayama Prefecture, Japan Akira Kawaguchi
Received: 2 January 2014 / Accepted: 18 February 2014 Ó The Phytopathological Society of Japan and Springer Japan 2014
Abstract Bacterial spot, caused by Xanthomonas arboricola pv. pruni, is the most important disease that affects peach production in Okayama Prefecture, Japan. Currently, this disease is managed mainly with copper compounds applied at two stages, before flowering and after harvesting, or with antibiotics applied in May and June. Here we identified the disease risk factors that affect peach at harvest and developed a disease-forecasting model to help growers decide when to apply bactericides. The model was based on parameters for weather data collected for September and October of 2001 through 2012 and for April, May, and June of 2002 through 2013, combined with data on bacterial leaf spot incidence obtained from 28 to 30 fields per year in August from 2001 to 2012 and in May to July from 2002 to 2013. The model, developed using a logistic regression analysis, included the percentage of fields with a bacterial spot incidence (BSI) C1 % in midAugust of the previous season and the number of rainy days (C5 mm/day) during the current June as predictors, and explained 75.0 % of the variability. These results suggest that the previous season’s BSI and weather variables in the present season can be used to predict the risk of bacterial spot. Keywords Logistic regression Bacterial spot of peach Disease-forecasting model Nested case–control study Risk assessment
A. Kawaguchi (&) Research Institute for Agriculture, Okayama Prefectural Technology Center for Agriculture, Forestry and Fisheries, 1174-1 Koudaoki, Akaiwa City 709-0801, Japan e-mail:
[email protected]
Introduction Bacterial spot (also called bacterial shot hole) of peach (Prunus persica), which is caused by Xanthomonas arboricola pv. pruni (Xap), is an important bacterial disease that causes substantial economic losses worldwide (Stefani 2010). Symptoms include fruit spots, leaf spots, and twig cankers. Pitting, cracking, gumming, and water-soaking of tissue also affect fruit. Severe leaf spot infections can cause early defoliation, which reduces vigor and winter hardiness of trees. Severe defoliation can result in smaller fruit and in sunburn and cracking of the fruit. Xap also causes bacterial spot on plum and nectarine trees. Management of bacterial spot on peach is difficult owing to the limited management options that are available. In Japan, chemical control based on the application of copper compounds or antibiotics is the most common strategy in the field, but it has limitations: the copper compounds that can be used depend on tree phenology and are also persistent pollutants, and the use of antibiotics raises concerns about the development of resistant strains of Xap. Kawaguchi et al. (2014) reported that nonpathogenic strains AZ98101 and AZ98106 of X. campestris, which were isolated in Japan, effectively controlled bacterial spot on peach, but biocontrol is difficult, and the antagonistic agents produced by these bacteria for use in spray applications are still under development and cannot yet be sprayed in orchards. Xap overwinters in buds, in protected areas on the woody surface of the tree (e.g., in cracks in the bark), and in leaf scars that become infected during leaf drop in the previous season (Ritchie 1995). Leaf scar infections usually produce spring cankers, and Xap spreads to the newly emerging leaves from the cankers in dripping water and in splashing or wind-blown rain (Ritchie 1995). Xap can also
123
J Gen Plant Pathol
infect trees through natural openings such as lenticels or wounds, and high moisture conditions promote both leaf and fruit infections (Ritchie 1995). Although some meteorological and pathogenic risk factors are known to affect the emergence of bacterial spot on peach, the major risk factors that determine the incidence of this disease remain unclear, and there is no forecasting model to predict the severity of outbreaks (Stefani 2010). The advent of efficient dataloggers, the establishment of extensive networks of weather stations, and the development of new statistical tools have allowed the development of next-generation forecasting models. Many of these models have used nonparametric methods such as logistic regression (Hosmer and Lemeshow 1989) to predict the prevalence of diseases such as Sclerotinia stem rot in soybean (Mila et al. 2004), Fusarium head blight in wheat (De Wolf et al. 2003), gray leaf spot in maize (Paul and Munkvold 2004), Stewart’s wilt in corn (Esker et al. 2006), and white mold in dry beans (Harikrishnan and del Rı´o 2008). A logistic regression analysis selects the major risk factor or factors responsible for disease emergence and
supports the development of a forecasting model. Because peach growers need timely risk information, the primary objective of this study was to develop a disease-forecasting model based on logistic regression to estimate bacterial spot incidence (BSI) in Okayama Prefecture, Japan, using weather variables and data on the prevalence of fields with a threshold BSI on leaves as predictors.
Materials and methods Disease and weather data collection Incidence of bacterial spot on leaves was documented in field surveys during the 2001–2013 growing seasons (May to August). In southern Okayama Prefecture, peach trees flowering in early to mid-April, the fruit-swell stage is from May to June, harvest is late July; by mid-August, harvest is over. In the field surveys, conducted under the Prevalence Reconnaissance Business by Prefecture (PRBP), 3–5 fields
Table 1 Variables used to predict bacterial spot incidence (BSI) of C10 % of peach trees in late July of the current season in Okayama Prefecture from 2002 to 2013 Factor
Predictor variable
Disease severity
Odds for the no. of fields with BSI (C1 %) in mid-August of previous season
Odds for the no. of fields with BSI (C1 %) in late May of current season
Odds for the no. of fields with BSI (C 1 %) in early May of current season
Odds for the no. of fields with BSI (C1 %) in late June of current season
Odds for the no. of fields with BSI (C1 %) in early June of current season Rain
No. of rainy days (C5 mm/day) in September of previous season
No. of rainy days (C5 mm/day) in October of previous season
No. of rainy days (C1 mm/day) in April of current season
No. of rainy days (C5 mm/day) in April of current season
No. of rainy days (C1 mm/day) in May of current season No. of rainy days (C1 mm/day) in June of current season
No. of rainy days (C5 mm/day) in May of current season No. of rainy days (C5 mm/day) in June of current season
No. of days with maximum temperature (C20 °C) in April of current season
No. of days with maximum temperature (C25 °C) in April of current season
No. of days with maximum temperature (C20 °C) in May of current season
No. of days with maximum temperature (C25 °C) in May of current season
No. of days with maximum temperature (C25 °C) in June of current season
No. of days with minimum temperature (\20 °C) in June of the current season
Typhoon frequency
No. of typhoons that hit Okayama from May to November of previous season
No. of typhoons that hit Okayama between May and June of current season
Wind
No. of days with maximum wind speed (C5 m/s) in September of previous season
No. of days with maximum wind speed (C10 m/s) in September of previous season
No. of days with maximum wind speed (C5 m/s) in October of previous season
No. of days with maximum wind speed (C10 m/s) in October of previous season
No. of days with maximum wind speed (C5 m/s) in April of current season
No. of days with maximum wind speed (C10 m/s) in April of current season
No. of days with maximum wind speed (C5 m/s) in May of current season
No. of days with maximum wind speed (C10 m/s) in May of current season
No. of days with maximum wind speed (C5 m/s) in June of current season
No. of days with maximum wind speed (C10 m/s) in June of current season
Temperature
123
J Gen Plant Pathol Fig. 1 Map of seven regions where the incidence of bacterial spot of peach was investigated (circles) and the three weather stations that provided climate data (triangles) in Okayama Prefecture, Japan. The symbols for the weather stations have the same color as the circles for the area that they were used to predict disease incidence
Region in Okayama and Akaiwa City Region in Kurashiki and Soja City Region in Kasaoka and Ibara City Weather station in Okayama City Weather station in Kurashiki City Weather station in Ibara City
20 km
per region were investigated in seven regions of southern Okayama Prefecture (Fig. 1), the primary peach-producing areas of Okayama Prefecture. To obtain a BSI for each field, 20 randomly selected current-year shoots (about 50 leaves per shoot) were randomly examined on each of 3–5 peach trees per field (60–100 shoots per field), and the proportion of the leaves with symptoms of bacterial spot was used to represent the BSI. The disease incidence data was obtained from the same 28–30 fields each year (n = 28–30 per year). In all, 345 data points (28–30 fields 9 12 years) were obtained from these fields during the 12 years (n = 345). One data point represents one BSI per field. It was not possible to survey fruit in all fields at each survey time because most of the peach growers were adding protective bagging on each fruit to protect them from sunshine and several diseases and insects. The disease severity was determined using the proportion of diseased leaves as follows: 0, \1 %; 1, 1 % B proportion \ 5 %; 2, 5 % B proportion \ 10 %; 3, 10 % B proportion. The BSI was binary-coded as either 0 (\1 % of the leaves) or 1 (C1 %) in leaves examined in mid-August of the previous season and in early or late May and early or late June of the current season, and as 0 (\10 %) or 1 (C10 %) in leaves examined in late July of the current season (Table 1). Though there was no information about economic injury level of bacterial spot on peach, I used two thresholds for damage to peach trees and yield loss: (1) \1 diseased leaf per 100 leaves, from my observations, at these levels, BSI does not contribute to significant yield loss; (2) 10 diseased leaves per 100 leaves; the point at which significant yield loss results from early
defoliation and the increased incidence of diseased fruits. For the response variable, I chose the proportion of fields with BSI C10 % in late July of the current season because this is the peak harvesting period for peaches in Okayama Prefecture. The disease incidence was represented as the odds of an infection, defined as: Odds ¼ Pi = ð1Pi Þ
ð1Þ
where Pi is the proportion of fields with BSI at disease severity level i. The weather data at the weather stations shown in Fig. 1, chosen because they were no farther than 20 km from the fields surveyed for disease incidence, were collected from April to June and from September to October of each year from the Automated Meteorological Data Acquisition System (AMeDAS) of the Japanese Meteorological Agency (http://www.jma.go.jp/jma/indexe.html). The weather parameters used in developing the regression model are summarized in Table 1. Rainfall of C5 mm/day and a maximum wind speed of C10 m/s favor disease development (Morimoto 2011). Typhoon incidence is also important because these storms combine strong winds with heavy rainfall and might therefore promote disease development. In addition, air temperatures of C20 °C are suitable for growth of Xap, and those C25 °C promote bacterial growth. Thus, I selected the following weather factors: number of rainy days (C1 mm/day or C5 mm/ day); number of days with a maximum or minimum temperature of \20 °C, C20 °C, or C25 °C; number of days with a maximum wind speed of C5 m/s or C10 m/s; and
123
J Gen Plant Pathol
number of typhoons that hit Okayama Prefecture during the growing season.
Ln½P= ð1PÞ ¼ a þ b1 x1 þ b2 x2 þ . . . þ bn xn
ð2Þ
where a is the y-intercept and bn is the coefficient associated with predictor variable xn. Equation (2) can be expressed as: P ¼ 1 = f1 þ exp½ða þ b1 x1 þ b2 x2 þ . . . þ bn xn Þg
ð3Þ
The logistic model was developed using binary-coded data as 0 or 1 for each level of BSI in each period in individual field and number of days of each weather condition and the number of typhoons from 3 weather stations. The logistic model was developed in two steps. First, all the selected weather variables were evaluated for the time periods shown in Table 1 to determine whether they were associated with BSI. This analysis was performed using the EZR (Kanda 2013) graphical user interface for R software (R Foundation for Statistical Computing, version 2.14.0). To run this analysis, data for each field were obtained from the weather station closest to that field. Only weather variables with a moderate to high and significant correlation (r [ 0.4, p \ 0.05) with BSI were included in the model. In the second step, the disease incidence data from the surveyed fields were entered into the logistic regression procedure provided by R using the EZR, along with the selected weather variables. The stepwise selection method was chosen over alternative procedures because, in the stepwise selection procedure, the variables must meet both entry and retention criteria, which provides a more stringent selection method. The stepwise selection of the explanatory variables was based on the value of Akaike’s information criterion (AIC) instead of using p values; AIC can select among models on the basis of an optimal combination of parsimony (limiting the model to the smallest number of parameters needed to explain the data) and goodness of fit (Akaike 1973). In this approach, lower values of AIC indicate a better model. AIC was defined as: AIC ¼ 2 ln L þ 2k
ð4Þ
where L is the maximum likelihood and k is the number of parameters (Akaike 1973). The AIC stepwise procedure, provided by R uses the EZR. The disease-forecasting model was validated by comparing the observed frequencies with the predicted probabilities of BSI of C10 % on peach trees from 2002 to 2013 in Okayama Prefecture in a linear regression to calculate the adjusted-coefficient of determination (R2).
123
Fields with bacterial spot (BSI) (%)
The logistic regression model was defined as:
70 60
50
40
30
20
10
0 Early May
Late May Early June
Late June Early July
Late July Mid August
Fig. 2 The percentage of fields with a BSI C1 % between May and August from 2002 to 2013 in Okayama Prefecture. The box represents 50 % of the data between the 25th and 75th percentiles (i.e., the lower and upper quartiles) and draws focus to the center of the distribution, the median, which is depicted by the line inside the box. The vertical lines extending from the box reach the minimum and maximum values
Fields with bacterial spot (BSI) (%)
Model development
80
80 70
≥1% diseased leaves
60
≥10% diseased leaves
50 40 30 20 10
nd
0
nd
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
Years
Fig. 3 Percentage of fields with low and high levels of BSI in late July from 2002 to 2013 in Okayama Prefecture
Nested case–control study I also conducted a nested case–control study, a variant of a case–control study that is performed within a cohort study. In this approach, cases of a disease that occur in a defined cohort (here, a specific month or part of a month) are identified, and for each case a specified number of matched controls is selected from among those in the cohort that have not developed the disease by the same time (Ernster 1994). The nested case–control design is particularly advantageous for studies of biological precursors of disease (Ernster 1994). To reveal any causal relationship between the numbers of fields with BSI from early May to early July
J Gen Plant Pathol
of the current season and the number of fields with BSI in late July of the current season, I defined a parameter called the odds ratio (OR):
Results Disease assessment Bacterial spot first appeared on peach in May, and the number of fields with BSI (C1 % diseased leaves per field) gradually increased to August (Fig. 2). Over the 12 years, the proportion of fields with BSI (C1 % diseased leaves per field) ranged from 5.4 to 75.9 %, and the proportion with significant incidence (C10 %) ranged from 0 to 20.7 % (Fig. 3). Incidence was highest in 2005 and lowest in 2009, and more than 20 % of the fields observed had BSI of C10 % in 2004 and 2005 (Fig. 3). Environmental data From 2002 to 2013 in Okayama City (in the surveyed region), the mean monthly temperature ranged from 14.6 to 28.8 °C between April and October (Fig. 4), and about 10 days had temperatures [35 °C in August. Total rainfall was more variable than temperature. On average, July was the wettest month, with about 170 mm of rain (Fig. 4). Logistic regression modeling The best model had AIC = 136.9 (Table 2). It included the odds for the number of fields with BSI of C1 % in
Temperature(C) (°C) Temperature
30
300
25
250 20 200 15 150 10
100 50
5
0
0
Temperature (°C)
where a is the number of fields with moderate BSI (C5 % diseased leaves per field; disease severity level 2) before late July of the current season (exposed group) but high disease severity (C10 %; level 3) in late July (disease case group); b is the number of fields with moderate BSI (C5 %; level 2) before late July of the current season (exposed group) and a similar disease severity (\10 %; level 2) in late July (control group); c is the number of fields with low BSI (\5 %; level 1) before late July of the current season (unexposed group) but high disease severity (C10 %; level 3) in late July (disease case group); and d is the number of fields with low BSI (\5 %; level 1) before late July of the current season (unexposed group) and more severe disease severity (\10 %; level 2) in late July (control group). OR is the ratio of the probability of developing a disease in the exposed group to that in an unexposed group (Sistrom and Garvan 2004). I calculated the 95 % confidence interval (CI) for OR and used Fisher’s exact test to identify significant parameters using R software.
350
Rainfall (mm)
ð5Þ
35
Rainfall(mm) (mm) Rainfall
Fig. 4 Mean total precipitation and air temperature (± SD) between April and October from 2002 to 2013 in Okayama City. The line represents temperature; the bars represent rainfall Table 2 Parameter estimates for the best-fit logistic regression model used to predict peach orchards with C10 % bacterial spot incidence (BSI) in Okayama Prefecture from 2002 to 2013 Variable
Parameter estimate
Standard error
z value
p value
Odds for the no. of fields with BSI (C1 %) in mid-August of previous season
3.654
0.752
4.859
\0.0001
No. of rainy days (C5 mm/day) in June of current season
0.227
0.095
2.385
0.0171
y-Intercept
-6.3707
1.009
-6.317 \0.0001
AIC (Akaike’s information criterion) = 136.9
25
Predicted frequency (%)
OR ¼ ða=bÞ = ðc=d Þ
400
y = 0.848x – 0.413, R 2 = 0.750, p = 0.0002 20 15
10
5 0 0
5
10
15
20
25
Observed frequency (%)
Fig. 5 Comparison of observed frequencies with predicted probabilities of bacterial spot incidence (BSI) of C10 % on peach trees from 2002 to 2013 in Okayama Prefecture
mid-August of the previous season and the number of rainy days (C5 mm/day) during June of the current season as predictor variables, and the proportion of fields with BSI of C10 % in late July of the current season as a response variable (Table 2). The resulting model was:
123
J Gen Plant Pathol Table 3 Nested case–control study of relationship between disease severity in late July of the current season and that in each period of current season Period
Disease severitya
Disease severity in late July Level 3 (disease case group)
Early May Late May Early June Late June Early July
Clevel 2
(Exposed group)
Blevel 1
(Unexposed group)
Clevel 2
(Exposed group)
Blevel 1
(Unexposed group)
Clevel 2
(Exposed group)
Blevel 1 Clevel 2
Totalb
BLevel 2 (control group)
0
0
0
25
324
349
1
1
2
24
323
347
8
1
9
(Unexposed group)
17
323
340
(Exposed group)
15
8
23
Blevel 1
(Unexposed group)
10
316
326
Clevel 2
(Exposed group)
19
18
37
Blevel 1
(Unexposed group)
6
306
312
Odds ratio (OR) (95 % confidence interval)
p value (fisher’s exact test)
–
–
13.5 (0.8–1285.9)
0.138
152.0 (18.0–1285.9)
\0.001
59.3 (20.4–171.7)
\0.001
53.8 (19.1–151.3)
\0.001
a
Disease severity levels: 0, proportion of diseased leaves\1 %; 1, 1 % B proportion of diseased leaves\5 %; 2, 5 % B proportion of diseased leaves \10 %; 3, 10 % B proportion of diseased leaves. Exposed means that group was exposed to a factor; disease severity over level 2 is the factor in this study
b
Number of fields with bacterial spot; sample size is 349 fields in 12 years
PJ ¼ 1= f1 þ exp½ð3:654D þ 0:227W 6:371Þg;
ð6Þ
where PJ is the proportion of fields with BSI of C10 % in late July of the current season, D is the odds for the number of fields with disease incidence of C1 % in mid-August of the previous season, and W is the number of rainy days (C5 mm/day) during June of the current season. When Eq. 6 was used to predict the bacterial spot incidence based on recorded values of D and W, the goodness of fit of the regression for the predicted versus observed incidence was strong and significant (adjusted-R2 = 0.750, p = 0.0002; Fig. 5). Nested case–control study A significant causal relationship existed between the number of fields with BSI in three periods (early June, late June, and early July of the current season) and the number of fields with BSI in late July of the current season (Table 3). The highest OR value (152.0) was calculated between the number of fields with BSI in early June of the current season and the number of fields with incidence in late July of the current season (Table 3).
Discussion I used a logistic regression to develop a model that predicted the probability of a BSI C10 % in peach tree fields in late July, the peak harvest period for peach fruit in Okayama Prefecture. The logistic model included the odds
123
for the number of fields with this BSI in mid-August of the previous season and the number of rainy days (C5 mm/ day) during June of the current season. The model accurately predicted bacterial spot incidence in Okayama (adjusted-R2 = 0.750 in a linear regression model for the predicted versus observed incidence). Using the logistic model, we can predict BSI 3–4 weeks in advance. In Eq. 6, PJ is the expected percentage of fields with BSI of C10 % in late July of the current season for all of Okayama Prefecture, D is the observed value for all of Okayama Prefecture, and W is the value for each region and has three different values. However, W values from each weather station in Okayama City, Kurashiki City, and Ibara City during 2002–2013 were almost the same (data not shown) because there were many rainy days in many regions in Okayama Prefecture and June is rainy season. Thus, when we predict PJ using Eq. 6, we should assign one W value from the weather station in Okayama City as typical data in southern Okayama Prefecture. When the model predicts a significant incidence in late July, peach growers can spray their orchard with bactericides such as oxytetracycline, validamycin, and oxolinic acid. For late-ripening peaches, typically harvested in August, there is likely to be an even longer advance warning that will allow growers to take measures to control bacterial spot. However, the predictive model must be improved to provide 2–3 months’ warning to better implement measures to control bacterial spot on peach. In the logistic model, two predictor variables (1 disease severity variable and 1 weather variable) were selected from 30 predictor variables (5 disease severity variables
J Gen Plant Pathol
and 25 weather variables) as the most important risk factors. The significant association between BSI in the current season and the two factors (BSI in the previous season and the number of rainy days) generally agrees with previous observations of bacterial spot of peach (Ritchie 1995). The estimate for the proportion of the fields with BSI in midAugust of the previous season was larger than that of the number of rainy days during June of the current season, indicating that disease incidence in the previous season affects risk more strongly than the number of rainy days does; a higher incidence of bacterial spot in August of the previous season would produce more leaf scar infections that could serve as a source of inoculum during the current season. Thus, effective management of bacterial spot on peach might be achieved by applying copper compounds from September to October to reduce the sources of inoculum for the next growing season. Pest management on peach during the autumn by applying bactericides is very important to permit cultivation of peach trees in open fields in Japan. However, some growers have not practiced any autumn pest management in recent years to reduce their labor and chemical costs. Since these results demonstrate the importance of autumn pest management, growers should continue to spray peach trees with copper in autumn. The model also predicts that rainfall during June of the current season will promote disease development. The number of rainy days (C5 mm/day) during June of the current season was the most important of the 25 weather factors examined for Okayama Prefecture. Since it is not possible to control rain in open field culture, growers should spray appropriate amounts of bactericides before the June rainy season. In May, streptomycin can be applied because it is an effective antibiotic compound, but it cannot be used within 60 days before harvest because of the Japanese pesticide usage standard. In a recent report, Nekoduka et al. (2009) performed a nested case–control study of the disease severity of Alternaria blotch of apple and revealed a highly probable causal relationship between disease severity during the early and late phases of an epidemic under commercial apple orchard management. In the present study, the highest value of OR was calculated between the number of fields with bacterial spot in early June of the current season and the number of fields with bacterial spot in late July of the current season, indicating the existence of a strong causal relationship between bacterial growth in early June and late July. Among nine fields in the exposed group in early June, eight were the same as those in the disease case group in late July (Table 3), indicating that most of the fields with a BSI of C5 % in early June had a high disease severity (C10 %) in late July. Thus, according to the results here, growers should spray enough bactericides in May to control
bacterial populations in June to keep the incidence of bacterial spot in early June to fewer than 5 % of the leaves. This control countermeasure suggested by the nested case– control study corresponds with the suggestion based on the logistic model and provides additional support for this management recommendation. The logistic model explained 75.0 % (adjustedR2 = 0.750 by linear regression model for the predicted versus observed incidence) of the variability in the observed disease incidence. However, the model did not predict the decreased percentage of fields with BSI in late July of 2006 and 2007 (Fig. 3). According to the model, a high number of fields with disease in the previous season (which is the case in 2005; Fig. 3) should have promoted disease development in the current season. This failure to predict the 2006 and 2007 results can be explained as follows. First, the Okayama Prefectural Government warned of an outbreak for bacterial spot of peach on 27 May 2005. In 2006, it also announced ‘‘that (based on the disease incidence in the previous year) growers should spray aggressively to manage bacterial spot in the coming year.’’ The growers were therefore encouraged to manage bacterial spot by spraying with bactericides, and appropriate spraying was carried out during the growing season and in the autumn in 2005 and 2006. This response decreased the proportion of fields with BSI in 2006 and 2007; the reduction might not have occurred without the government warnings because growers would have been less likely to spray. It is difficult to repeatedly survey disease incidence in the same fields over many years. However, plant disease incidence data can be obtained from the regular surveys that are conducted under the PRBP in each prefecture. Various kinds of disease incidence data are available for periods longer than 10 years at pest control stations in each prefecture. Therefore, we could utilize disease incidence data that have already been recorded by regular reconnaissance surveys under the PRBP program. Combining the disease incidence data from PRBP and the weather data from AMeDAS will support the development of more powerful statistical forecasting models to predict disease incidence and whether a given disease will become a problem in an area. Stefani (2010) noted that there is a need to develop a reliable model for forecasting BSI, since the disease is strongly influenced by climatic conditions. Unfortunately, no forecasting model is available thus far. So far as I know, the present study is the first report of the development of a logistic model capable of predicting BSI on peach. This model is currently useful only in Okayama Prefecture, because different factors may be important in other regions, but the method used to develop the model should nonetheless allow development of region-specific versions of
123
J Gen Plant Pathol
the model using local weather and disease severity data. This model will be used by Okayama Prefectural government and help growers make better-informed decisions on whether to apply bactericides to control bacterial spot and on the optimal timing for an application. Acknowledgments This work was supported by the Science and Technology Promotion Program for Agriculture, Forestry, Fisheries and Food Industry from the Ministry of Agriculture, Forestry and Fisheries, Japan (23037). I am grateful for the comments provided by the journal’s anonymous reviewers.
References Akaike H (1973) Information theory and an extension of the maximum likelihood principle. In: Petrov BN, Caski F (eds) Proceedings of the 2nd international symposium on information theory. Akadimiai Kiado, Budapest, pp 267–281 De Wolf ED, Madden LV, Lipps PE (2003) Risk assessment models for wheat Fusarium head blight epidemics based on withinseason weather data. Phytopathology 93:428–435 Ernster VL (1994) Nested case–control studies. Prev Med 23:587–590 Esker PD, Harri J, Dixon PM, Nutter FW (2006) Comparison of models for forecasting of Stewart’s disease of corn in Iowa. Plant Dis 90:1353–1357 Harikrishnan R, del Rı´o LE (2008) A logistic regression model for predicting risk of white mold incidence on dry bean in North Dakota. Plant Dis 92:42–46
123
Hosmer DW, Lemeshow S (1989) Applied logistic regression. Wiley, New York, pp 1–392 Kanda Y (2013) Investigation of the freely available easy-to-use software ‘EZR’ for medical statistics. Bone Marrow Transplant 48:452–458 Kawaguchi A, Inoue K, Inoue Y (2014) Biological control of bacterial spot on peach by nonpathogenic Xanthomonas campestris strains AZ98101 and AZ98106. J Gen Plant Pathol 80:158–163 Mila AL, Carriquiry AL, Yang XB (2004) Logistic regression modeling of prevalence of soybean Sclerotinia stem rot in the north-central region of the United States. Phytopathology 94:102–110 Morimoto R (2011) Occurence and control of bacterial shot hole on peach (in Japanese). Plant Prot 65:210–214 Nekoduka S, Hada H, Iwadate Y, Ishiguro K (2009) A nested case– control study on disease severity of Alternaria blotch of apple at early and late phases of epidemics (in Japanese). Jpn J Phytpathol 75:314–322 Paul PA, Munkvold GP (2004) A model-based approach to preplanting risk assessment for gray leaf spot of maize. Phytopathology 94:1350–1357 Ritchie DF (1995) Bacterial spot. In: Ogawa JM, Zehr EI, Bird GW, Ritchie DF, Uriu K, Uyemoto JK (eds) Compendium of stone fruit diseases. APS Press, St. Paul, pp 50–52 Sistrom CL, Garvan CW (2004) Proportions, odds, and risk. Radiology 230:12–19 Stefani E (2010) Economic significance and control of bacterial spot/ canker of stone fruits caused by Xanthomonas arboricola pv. pruni. J Plant Pathol 92(Suppl 1):99–103