A novel solution to the variable selection problem in Window Pane approaches of plant pathogen - Climate models: Development, evaluation and application of a climatological model for brown rust of wheat

A model for predicting brown rust severity in France was developed using the systematic screening of climatic variables of the Window Pane approach and data from 400 field trials spanning 30 years. The model was built using novel methods to manage the variable selection problem posed by the very large number of predictor variables generated by Window Pane, namely the elastic-net, and a systematic cross-validation to determine the most frequently retained variables. The model predicts the final severity of brown rust with an RMSEP (root mean square error of prediction) of 22.4%. The model's ability to predict treatment decisions was tested and exhibited a good performance as shown by an area under the receiver operator curve of 0.85. We also evaluated the suitability of our model for use in France by confronting the range of the climate variables in our dataset against the climatological range of these same variables in France. The final model also gives important insights into the key factors behind variations in brown rust disease pressure.