Data-driven severity prediction of net blotch in spring barley

This study examined how the severity of barley net blotch, a common plant disease in spring barley, can be predicted using weather data. Barley net blotch can significantly reduce crop yields, so being able to forecast the disease early is important for farmers. The research used computational methods, meaning mathematical models and computer-based analysis, to make these predictions.

The study analysed 36 data sets covering 26 years of public weather records together with historical observations of barley net blotch severity. From the weather measurements, new calculated features were created using different mathematical operations. These features were then used to build prediction models. Several types of mathematical classifiers were tested to determine how accurately the disease severity could be predicted.

The results showed that weather data contain enough useful information to predict the occurrence of net blotch accurately already at the beginning of the growing season. However, the amount of useful information in the weather data changes as the season progresses. The study also shows that the best time to start the prediction process can be determined automatically by detecting the beginning of the growing season using outdoor temperature measurements.

The developed prediction method can be applied without separate field measurement campaigns, which makes it practical and cost-effective. The highest accuracy was achieved with ensemble models that combined several binary linear discriminant classifiers using the geometric mean. The prediction accuracy was nearly perfect with test data set. Such a reliable forecasting system helps farmers optimise pesticide use, lowering both environmental impact and production costs, and supporting sustainable agriculture.