Location: Subtropical Plant Pathology Research2013 Annual Report
1a. Objectives (from AD-416):
1. Develop a generic epidemiological model that can be used to compare control scenarios and to optimize the probability of controlling and managing high-risk pathogens of agricultural significance. 2. Develop an economic extension to the epidemiological models to balance efforts between detection and control to maximize use of fiscal and manpower resources. 3. Develop user friendly model ‘front ends’ that can be used by researchers and regulatory agencies.
1b. Approach (from AD-416):
Data sets to parameterize the model will be collected from U.S. epidemics of high-risk pathogens such as citrus huanglongbing and citrus canker and compiled for model use. Parsimonious, stochastic computational models will be used to provide an efficient generic modeling platform in combination with Markov Chain Monte Carlo and ABC modeling methods to estimate model parameters. An economic module will be developed for the model to simulate multiple scenarios for parsing manpower and fiscal resources between detection and control efforts to maximize use of resources. A user friendly ‘front end’ interface will be developed in Adobe Flash or equivalent platform for use by research and regulatory personal that can be placed on the Web or deployed directly.
3. Progress Report:
This research is related to inhouse project objective 3: Develop or improve comprehensive integrated disease management strategies. During this cycle we have focused primarily on making the models more robust and flexible models to most efficiently analyze and predict the spread of emerging pathogens across a range of scales extending from within-plantation to the landscape and regional scales for both HLB and canker. We are continuing to test the models extensively to assess how host planting age affects the transmission of HLB. We consider the models to be and are developed as part of a flexible tool-box so that they can be readily adapted to new disease threats as well. Work has continued on parameter estimation and writing up the methodology. Removal or treatment of infected trees in a region in which the dispersal and transmission are being estimated for an emerging epidemic can have profound effects in under-estimating some parameter values if not allowed for. Accordingly we have also developed methods to allow for this during parameter estimation. This requires an understanding of how the control is likely to affect epidemiological parameters and where this is unknown how to compare alternative models. Once parameterized, the models allow us to predict where disease is likely to spread most rapidly. We have tested a risk-based strategy that uses epidemiological knowledge on where disease is likely to spread most rapidly to optimize control that involves removal of more susceptible hosts around key infected sites. Using a parameterized model of citrus canker on realistic landscapes, we have shown that the risk-based strategy would outperform a conventional fixed-radius approach when the dispersal and transmission parameters are accurately known. We are testing to what extent performance degrades under varying degrees of uncertainty. Additional work has also been completed in this quarter in collaboration with a graduate student on a model of host-vector systems, considering the spread of HLB within an individual tree in particular. The relative importance of pathogen transmission within the vascular system of the tree and between leaves via psyllid vectors has been studied. Currently, the models are being used to investigate the efficacy of potential methods for the control of HLB, including the use of insecticide, roguing, use of nutritional products and thermotherapy, amongst others. Following an extensive period of testing methods to parameterize stochastic models for HLB and citrus canker, and the development of flexible models for spread of epidemics through heterogeneous landscapes, we have focused on: (i) the practical use of the models to compare control scenarios at the landscape scale; (ii) to allow for uncertainties in the location of hosts at different scales in the landscape and in a policy makers knowledge of parameter estimates; and (iii) to incorporate meteorological dynamics in the models, particularly to allow for real weather in dispersal events. One of the principal challenges – also addressed in the development of user-friendly models - is to be able to run the models sufficiently fast to enable the user to compare a range of ‘what-if’ control scenarios with different levels of uncertainty, allowing rapid presentation of results within real time. Substantial progress has been made through the development and testing of efficient computational methods (to allow finely-resolved computation of multi-county and state-wide spread of disease. Two approaches have been used, the familiar individual-tree-based models and meta-population models. We have also analyzed the effectiveness of greatly increasing the speed of execution of stochastic, spatially-extended model using dedicated multi-core computers available to the Epidemiology and Modeling Group. The additional computing power has also allowed the incorporation of meteorological dispersal models into the epidemiological toolbox in objective (1) above. These models can be used to track intermediate and long-distance movement of vectors; providing measures of primary introduction of inoculum. Turning to detailed models within plantations, analysis of the effects of host planting age on the transmission of HLB has been completed. We have also completed the analysis of how to allow for removal or treatment of infected trees in a region in which the dispersal and transmission are being estimated for an emerging epidemic. We are continuing to test the risk-based method to show where disease is most likely to spread in order to optimize control that involves removal of more susceptible hosts around key infected sites. Recent work on citrus canker has examined weather-driven variability in the transmission rate of infection by extracting signals from Markov chain Monte Carlo estimates of parameters averaged over different time-scales of one to several months and correlating these with weather variables. The method suggests that it will be possible to use Florida data to estimate the spread of disease under different environmental conditions, for example those typical of Texas or California and to compare the risk of spread under different runs of months/years of favorable or unfavorable weather conditions. Building on a model for HLB within individual plants, we have compared the relative importance of pathogen transmission within the vascular system of the tree and transmission between leaves via psyllid vectors. The model is being used to compare the efficacy of roguing, application of insecticide and the use of nutritional products and thermotherapy for disease control and mitigation.