A neural network model for prediction of deoxynivalenol content in wheat grain based on weather data and preceding crop

Abstract

Deoxynivalenol (DON) is the most prevalent Fusarium toxin in Czech wheat samples and therefore forecasting this mycotoxin is a potentially useful tool to prevent it from entering into food chain. The data about DON content in wheat grain, weather conditions during the growing season and cultivation practices from two field experiments conducted in 2002–2005 were used for the development of neural network model designed for DON content prediction. The winning neural network is based on five input variables: a categorial variable – preceding crop, and continuous variables – average April temperature, sum of April precipitation, average temperature 5 days prior to anthesis, sum of precipitation 5 days prior to anthesis. The most important input parameters are the preceding crop and sum of precipitation 5 days prior to anthesis. The weather conditions in April, which are important for inoculum formation on crop debris are also of important contribution to the model. The weather conditions during May and 5 days after anthesis play only an insignificant role for the DON content in grain. The effect of soil cultivation was found inferior for model function as well. The correlation between observed and predicted data using the neural network model reached the coefficient <i>R</i>² = 0.87.