Prediction of Deoxynivalenol Content in Dutch Winter Wheat

Abstract

Predictive models for the deoxynivalenol (DON) content in wheat can be a useful tool for control authorities and the industry to avoid or limit potential food and/or feed safety problems. The objective of this study was to develop a predictive model for DON in mature Dutch winter wheat. From 2001 to 2007, the concentration of DON was measured in winter wheat samples taken just before harvest from 264 fields throughout The Netherlands. Agronomic and climatic variables were obtained for each field for a 48-day period, centered on the heading date. Multiple regression was used to determine the most important variables and to construct the predictive model. The first model (model 1) was based on 24-day pre- and postheading periods, while the second model (model 2) was based on eight time blocks of 6 days around the heading date. Although both models showed good statistical evaluations and predictive performance, model 1 showed the highest performance (R2 of 0.59 between observed and predicted values, fraction samples correctly below or above the 1,250 μg/kg threshold of 92%, and sensitivity of 63%). With both models, the predicted DON level increased with a higher average temperature, increased precipitation, and higher relative humidity, but decreased with increased number of hours with the temperature above 25°C. We observed a strong regional effect on the levels of DON, which could not be explained by differences in the recorded agronomic and climatic variables. It is suggested that future model improvement might be realized by indentifying and quantifying the mechanism underlying the region effect.