Northeast Region
IPM Grants
|
|
| States: | New York, West Virginia |
| Project Directors: | Herb Aldwinckle, Alan R. Biggs, |
| Institutions: | Cornell University, West Virginia University |
| Project Type: | Research and Extension |
| Award*: | $159,585 research and $17,928 extension |
| Term: | 36 months |
|
Setting: |
apple |
*Award shown is total amount to be used over the course of the project term.
Fire blight, caused by the bacterium Erwinia amylovora, is one of the most
destructive and difficult-to-control diseases of apple. Over the past 15 years,
consumer and market demands have forced major changes in horticultural practices
that have resulted in an overall increase in orchard risk for infection. These
horticultural changes have not only increased the potential incidence of infection
but also the level of damage likely to occur. Fire blight can be controlled
to manageable levels in most years with the antibiotic streptomycin. Streptomycin
is highly effective at controlling disease when applied at the appropriate timings.
In the Northeast, this is achieved through the use of the forecaster MARYBLYT,
a computer program for forecasting fire blight that predicts the four distinct
types of infection events (i.e., blossom, shoot, canker, and trauma blight)
as well as the appearance of symptoms that follow.
Blossom blight is the most destructive phase of the disease; providing inoculum
for the shoot, root, and trauma blight phases. Consequently, management practices
focus on controlling this phase. MB monitors four risk factors to identify possible
infection events: 1) blossom development; 2) epiphytic inoculum potential (EIP,
a measure of the pathogen population); 3) moisture in the form of rainfall or
dew; and 4) average daily temperature. We are currently revising MB to calculate
a system of ‘risk points’ as function of the MB risk factors. This will incorporate
greater flexibility in the model and allow management thresholds to be selected
based on the accumulation of ‘risk points’ that can be tailored to factors such
as variety, inoculum pressure, and the users comfort for assuming risk. Ideally,
however, management thresholds should be tied to a crop or economic loss function.
Receiver Operator Characteristic (ROC) curve analysis can be used to achieve
this goal. ROC analysis is a graphical approach for comparing and selection
of “competing” forecasters or variations of the same forecaster. The ROC curve
is a summary of the performance of a forecaster in terms of its ‘sensitivity’
and ‘specificity’ over the complete range of its output values. The ‘sensitivity’
is the ability of forecaster to predict disease when disease occurs; the ‘specificity’
is the ability of a forecaster to predict the absence of disease when disease
does not occur. ROC curve analysis not only facilitates model selection, but
the ROC curve itself provides a means for risk management.
The ROC curve can be described in a functional form (i.e., an equation) for
any forecaster. Metz (1978), showed how the ROC curve can be used to select
thresholds based on the economics of disease management and an estimate of disease
prevalence. Specifically, the costs of correctly managing disease, the expected
losses based on mismanagement, and a means to estimate disease prevalence are
required. The objectives of this research are to: 1) Develop a spreadsheet to
calculate the costs of disease management and the projected losses over the
orchard recovery or reestablishment period under different risk scenarios; and
2) Develop a prevalence model for estimating inoculum pressure. Fulfillment
of both objectives will provide the necessary information to optimize threshold
selection in MARYBLYT or any other forecaster for which an ROC curve has been
developed. In addition, completion of the first objective will provide a tool
to assist growers in deciding whether trees in blocks affected by fire blight
should be pruned or replaced.
Based on theory developed by Metz (26), the ROC curve can be used to select thresholds that minimize the average costs associated with each of the four possible outcomes when management decisions are based on forecasts. This requires as input: a) the costs of correctly managing disease and the expected losses based on mismanagement; and b) an estimate of disease prevalence.
The objectives of this research are to: 1) Develop a spreadsheet to calculate the costs of disease management and the projected losses over the orchard reestablishment period under different risk scenarios; and 2) Develop a prevalence model for estimating inoculum pressure. Fulfillment of both objectives will provide the necessary information to optimize threshold selection in MARYBLYT or any other forecaster for which an ROC curve has been developed. In addition, completion of the first objective will provide growers with a tool to assist them in deciding whether trees in blocks affected by fire blight should be pruned or replaced.
With the increased level of interest in high-density orchards the economics of fire blight management are quickly becoming a driving factor in apple production (8,50). In an era where the Northeastern apple industry is struggling to compete, epidemics such as those that occurred in 1991 and 2000 are economically crippling to growers adapting to the changes of the industry. Add to this the fact that over the last 20 years spray expense per bearing acre is the fastest growing component of total farm costs next to labor in the Northeast (50). Growers must be able to make informed choices when managing fire blight. This means: (1) having a clear understanding of which blocks are at the greatest risk for damage when environmental conditions favor the development of fire blight; and (2) understanding how to select and order susceptible blocks for treatment so that the risk of economic loss is minimized; that is which blocks get sprayed first, which get sprayed last.
ROC analysis can be used to achieve this goal. The ROC curve is a summary of the performance of a forecaster in terms of its ‘sensitivity’ and ‘specificity’ over the complete range of its output values (see ROC theory below). The ‘sensitivity’ is the ability of forecaster to predict disease when disease occurs; the ‘specificity’ is the ability of a forecaster to predict the absence of disease when disease does not occur. ROC curve analysis not only facilitates model selection (as used above), but the ROC curve itself provides a means for risk management. The ROC curve can be described in a functional form (i.e., an equation) for any forecaster; this forms the basis for the selection of optimal economic thresholds for disease management. Specifically, it is possible to integrate a cost function that can summarize the costs of correctly and incorrectly managing disease into the function characterizing the ROC curve. Entomologists have argued for over 20 years that management thresholds should be tied to a crop damage function or an economic indicator (33). It is interesting to note that neither MB, nor any other forecaster, considers the potential economic losses as criteria for deciding whether a spray is warranted or not. This research represents a fundamental step forward showing how the economics of disease management can be integrated with disease forecasting and represents a realization of “precision agriculture.” The MARYBLYT disease forecasting system not only provides us with a “model system”, but results from this research has the potential to significantly alter how one of the most destructive diseases of apple is managed.
| |
|
|
|