Development of Neural Network Model to Predict Deoxynivalenol (DON) in Barley Using Forecasted Weather Conditions

• Activity Number: 190
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Krishna Deepthi Bondalapati*+ and Jeff Stein and Kathleen Baker
• Companies: South Dakota State University and South Dakota State University and Western Michigan University
• Keywords: Fusarium Head Blight ; Scab ; Neural Network ; Mycotoxins ; DON ; Barley

Abstract

Fusarium head blight of barley, caused by the fungus Gibberella zeae (anamorph: Fusarium graminearum), is a devastating disease in U.S. Northern Great Plains. Losses occur through the blighting of florets, disruption of grain fill, and most importantly through the contamination of grain with mycotoxins, primarily deoxynivalenol (DON). A weather-based predictive model has been developed for estimating the risk of economically significant DON accumulation in barley grain, however this model utilizes environmental conditions that have already occurred and therefore can only facilitate reactive disease management. Such a system is useful, but does not fully support integrated crop disease management strategies. In this research, a single layer neural network (NN) model was developed using a combination of measured and forecasted weather data to predict the risk of economic DON levels 5-days in advance to the period of peak crop susceptibility. The developed NN model had a prediction accuracy of 91% with a sensitivity of 76% and specificity of 96%. The model resulted in 89% prediction accuracy when tested on 55 independent field data collected from locations in the region of interest.