Cross-validation of predictive models to predict Deoxynivalenol contamination in wheat at harvest.

Fusarium head blight (FHB) is a small grain cereal disease caused by a complex of fungal species belonging to Fusarium spp. Infection of wheat ears occurs during the flowering period and, under favourable environmental conditions, can cause considerable wheat yield and quality losses. Fusarium spp. received particular attention as producers of mycotoxins which may contaminate products along the cereal food-feed chain. The main mycotoxins found in small grain cereals are tricothecenes, with nivalenol (NIV), deoxynivalenol (DON) and its acetylated derivatives 3-and 15-acetyl deoxynivalenol, T-2 toxin and HT-2 toxin as the most relevant ones. The European Commission has established legal limits for Fusarium toxin contamination of food products and commodities to reduce human dietary exposure. The maximum level of DON in unprocessed wheat has been set at 1250 µg/kg and 1750 µg/kg for soft and durum wheat, respectively. Risk management procedures, aimed to limit mycotoxin contamination, must be applied through the entire cereal supply chain, to comply with the legal limits. As contamination occurs mainly during the cropping period, predicting DON contamination in wheat kernels at harvest can support decision making for tactic and strategic actions. For this reason, several mathematical models have been developed to predict FHB and/or DON in small grain cereals. Mathematical models can be distinguished based on the approach followed for their development into mechanistic (or explanatory) and empirical (or descriptive). The former are drawn considering the cause-effect relationship among variables, while the latter describe the relation between the driving factors of the phenomena and are developed by statistical analyses of data collected in field. To date, one mechanistic model for the prediction of DON in wheat has been published, aimed to predict Fusarium spp. infection pressure and risk for DON contamination at wheat harvest. Among the other models for DON in wheat published, a neural network was considered while the empirical approach was applied in several models. Among these, an empirical model was developed with the aim to predict DON contamination in wheat at harvest with output information destined to different groups of end-users. This study aimed to evaluate the predictive performance of an empirical and a mechanistic model for DON in wheat at harvest when used in other geographic areas. To this end, the mechanistic model developed in Italy and the empirical model developed in the Netherlands were used, as well as data collected in these two countries during 2001-2011. Results showed that predictions of both modelling approaches for independent wheat fields (sampled in Italy and The Netherlands) were good. Both models predicted correctly around 90% of the samples, to contain DON in kernels below the legal limit.