Taming Uncertainties in Multi-Scale Pest and Disease Model and Decision Support Tools for Plant Security

NON-TECHNICAL SUMMARY: Agriculture faces severe challenges imposed by plant diseases, which left uncontrolled, can reduce crop quality and yields, and in many cases, make commercial crop production impossible. Most management and control efforts are aided by weather-driven models that indicate the risk that a disease infection or cycle can occur. Such models can help with management and containment of both established and newly introduced diseases. The western US includes mainly arid, topographically diverse regions that pose special challenges for plant disease epidemiologists, weather forecasters, and growers wishing to minimize plant disease losses. Another problem is that plant disease models inherently have errors, which should be measured and reported along with model predictions, to aid in model understanding by end-users, and to help the process of model improvement. The purpose of this project is to develop and deliver an improved system for predicting plant disease infection risk of numerous diseases over large geographic scales. We will improve the spatial resolution of these predictions, and document their errors and uncertainties as well. This system will be developed and tested for several existing diseases of crops, and designed to also be used for newly introduced disease organisms over large regions in the western US.

<P>

APPROACH: A multi-crop support system for plant disease modeling at regional scales and with capabilities for expansion will be built using a variety of technologies. By focusing on western regions, we will deal with problems imposed by the varied subclimates and terrain effects common to much of the West. The system will integrate numerous aspects of modeling and forecasting, including: a) Site-specific weather station data from thousands of publicly available stations will be gathered in real and near-real time. b) Dynamic NOAA weather forecast models will provide 0-4 day spatialized source inputs for analysis. c) Longer term, lower resolution spatialized NOAA and other forecasts will also be used in the system. d) Orographic-based enhanced weather forecast models will combine station data with NOAA forecast models to interpolate and downscale forecasts to higher spatial resolution needed for plant disease models. e) Leaf wetness estimation models, adapted to work in the west as well as in the more humid eastern regions, driven by the weather forecast models, will estimate a key moisture input component used in many plant disease models. f) We will use GIS spatial modeling to present model outputs at various scales. g) A new targeted climatologies modeling system, will be used to compare to the longer term dynamic forecast models for 5-12 day forecasts. h) Methods including climatologically aided interpolation (CAI) and downscaling will improve model accuracy and quality assurance. i) A modular and generic plant disease infection risk modeling framework will be used to host a large range of plant diseases. j) Delivery of pest models will be via the internet including web-based GIS. k) We will coordinate targeted validations and sensitivity analyses for a variety of crop pests and diseases. These technologies will be automated and integrated into a comprehensive web-hosted support system. Each component will have its inputs and outputs measured for errors and uncertainty contribution to the pest model and forecast output s at various temporal and spatial scales. Disease forecasts will be validated using targeted field studies in several regions and cropping systems, including grass seed stem rust, powdery mildew in hops, grapes, and tree fruits, gray mold in small fruits, and pear scab in pear. Data quality reports, with estimates of errors and uncertainties for all measureable steps in the modeling process, will be available to help interpret quality of model predictions and potential steps needing improvement. The system will be designed for large scale tracking and prediction of plant disease for plant biosecurity needs by using a generic modeling database and framework, which will allow new pathogens to be quickly added. The system will be tested and used with introduced diseases, such as blackberry rust, a newly-emerging problem in the Pacific Northwest. The system will be used by IPM researchers, decision makers, National Plant Diagnostic Network (NPDN) scientists, APHIS, and state departments of agriculture. Workshops and training sessions will be conducted to facilitate uptake of the new plant disease forecast tools.
<P>
PROGRESS: 2008/01 TO 2009/01 <BR>
OUTPUTS: This project resulted in several new weather forecast methods for plant disease risk prediction, a fully implemented, online set of tools which integrate observed and virtual weather data, 6.5 day forecasts, numerous disease models, a data (QA/QC) system, statistical analysis tools for model and data verification, prototype methods for climatologically-aided forecasts and for targeted climatologically-based disease models, and new understanding of how to monitor and reduce the uncertainties inherent in weather driven models for pest management and plant biosecurity needs. The Fox Weather, LLC weather forecast system now uses 6-hr WRF mesoscale model 12km analysis and downscaling to hourly, 2km outputs using Fox's MtnRT model. These 6.5 forecasts are needed to run most plant disease models (11 now in production). An alternative, experimental PRISM-based weather analysis and forecast system was used to produce full forecast grids for NW Oregon during the 2008 field season and into early 2009. It takes advantage of the climatological "fingerprint" that physiographic features have on spatial weather patterns. The system starts with 0.8km PRISM mean temperature grids as a "first guess" for a given date/time, then modifies them using current station data and large-scale changes predicted by the GFS model, to produce forecast grids. In 2008, MAE of the 24-hour forecasts were generally unbiased, and ranged between 1.5 and 2.5C for all stations. In the area of virtual weather, IPPC developed an efficient local weather station-based, distance elevation weighted regression (DEWR) form of virtual weather data, which had lower error rates vs. all forecast-based estimation methods tested. For example, 75-day, MAE rates for this method were 0.9C, 1.6C, 0.9 mps, 2.0 mm, and 87% for hourly temperature, dew point, wind speed, 24 hr precipitation, and % correct hourly leaf wetness estimation, respectively, vs. 2.5C, 2.3C, 1.7 mps, 4.1 mm, and 79% for the latest Fox Weather forecast. In-field and in-orchard studies compared plant disease models driven by in-canopy, 15 min. inputs vs. standard height (1.5 m), 60 min. resolution inputs. Three different types of crop canopies (grape, hops and grass seed) comprised 16 site-years tested over 2007 and 2008. Errors associated with time resolution were small, and had little effect on disease model output and no effect on management decisions produced by the disease models. Errors associated canopy versus standard-height weather sensors were more substantial, consisting principally of a reduced temperature range (daily maximum minus daily minimum) at standard height. In most cases (12 of 16) sensor position did not alter disease management decisions (number of sprays). There was also no difference in management decisions or disease development in grapes for powdery mildew using canopy or virtual weather data. A relatively simple, crop-specific algorithm to correct standard-height temperature for canopy effects could mitigate the positional effects and essentially remove this source of error. <BR> PARTICIPANTS: Several partnerships and training opportunities have benefited from this project. The pest management decision support tools developed here have been used at numerous NPDN training workshops and grower and NRCS workshops conducted by IPPC in Oregon via the iSNAP and other extension related workshops. The members of the Western Weather Workgroup, funded by the Western Region IPM Center, continue to collaborate and work to improve concepts and ideas for building and delivering these tools. <BR> TARGET AUDIENCES: End-users of these technologies include growers, pest control advisors, extension personnel, researchers, students, USDA and State Dept. of Agriculture personnel dealing with invasive species, and others. <BR> <BR>
IMPACT: 2008/01 TO 2009/01<BR>
The plant disease risk forecast system operated by IPPC, and potentially many other IPM decision support systems, are progressively gaining improvements due to this research on the errors and uncertainties associated with weather data, models, and model delivery. This system is in operation at OSU IPPC (http://uspest.org/wea), providing data and models for plant biosecurity in the full US, and Fox Weather-developed forecasts for all of OR, WA, ID, and CA. Several crop production regions in the Pacific Northwest are already using and providing feedback on the first series of implemented products described above. We expect that several new products and decision support tools will result from our work on the virtual weather station concept. Already tests in the crops mentioned above show that valid management decisions can result from this development. This system is also taking form as plant biosecurity epidemiological investigation tools for invasive species, in that NPDN (National Plant Diagnostic Network) and USDA APHIS PPQ personnel are showing increasing interest in the tools for forecasting and planning monitoring programs. Results from field comparisons of within canopy and standard height weather observations, and frequency of data collection, may help in establishing standards and guidelines for modeling that result both in cost savings and fewer errors in management decisions. The Fox Weather/IPPC forecast system is in operation and is helping end-users with IPM decision making, and further efforts are underway to document these changes. The new knowledge gained from the experimental PRISM-based forecasts is expected to lead to further refinements and potentially, a higher resolution forecast system. We implemented online forecast and virtual data analysis programs which allow end-users to readily see error and uncertainty rates, allowing better understanding of the trade-offs in using these data to support decision making. This error analysis system has also already benefited methodologies for creating virtual and forecast data. The knowledge gained in the creation of virtual weather is leading to on-going proposals and ideas for implementation of virtual weather stations for end-users, in missing data estimation and data quality assurance, and in the development of forecast maps that may exceed any existing disease mapping products.