Predictive model: current knowledge and MycoRed contribution.

Good agricultural and management practices are recommended during pre- and post-harvest stages of production and processing of crops to avoid mycotoxin contamination exceeding the legal limits. However, additional interventions increase costs, thereby reducing the profitability of the crop value chain. A sustainable approach can be based on the prediction of mycotoxin risk to optimise crop chain management and analytical efforts. A modelling perspective was followed as support for value chain improvement in MycoRed. Predictive models can be distinguished based on the approach followed for their development into mechanistic (or explanatory) and empirical (or descriptive) models. 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. Mechanistic models commonly do need few or none adaptation when used in conditions other than those considered for model development. On the contrary, a model calibration is requested when empirical models are applied to other conditions, such as other geographic areas. Therefore, the mechanistic approach was chosen in MycoRed because several countries, placed in different geographic areas, were involved in model development and validation. Several pathosystems and related mycotoxins were considered: Fusarium spp in wheat and deoxynivalenol, Fusarium verticillioides in maize and fumonisins, Aspergillus flavus in maize and aflatoxins, Aspergillus section Nigri in grapes and ochratoxin. Only one predictive model was available in literature for the former pathosystem; therefore, mechanistic models were developed for all the others during this study with the aim of predicting if the considered mycotoxin contamination at harvest is above the legal limit fixed by the European legislation. More than 500 wheat field samples, 500 maize field samples and 250 grape samples were collected, supported by cropping system data, and weather data from station placed in the neighbourhood of the fields, for model validation. Moreover, mycotoxin contamination data were obtained by chemical analyses. The results indicated a percentage of correct predictions between 60 and 90% of evaluations, depending on the pathosystem and the year. Some lack of knowledge was highlighted, especially regarding the role of cropping system and showed that filling these gaps would improve model performances. Finally, the joint use of different models was considered and it seems a promising approach.