Can machine learning classification methods improve the prediction of leaf wetness in North-Western Europe compared to established empirical methods?

Leaf wetness is an important input parameter into disease prediction models. The use of machine learning algorithms for the classification of leaf wetness measurements from 30 meteorological stations in North Western Europe during the period of January 2014 to October 2018 was assessed in this study. The accuracy of the empirical models utilised within in this study was enhanced by increasing the relative humidity threshold from 90% to 92%. Increasing the relative humidity threshold led to an average increase in the classification accuracy of 1.12%. The use of machine learning classification algorithms consistently provided more accurate results for the prediction of leaf wetness when compared to the empirical models that were studied with an average increase in the classification accuracy of 4.85%. The sub-division of the data into regional subsets had a greater effect on the accuracy of the models than the temporal sub-division of the data. Machine learning classification techniques performed well compared with previously established empirical models for the prediction of leaf wetness. Further improvements in the algorithms are possible, making the techniques studied here a viable research tool.