Using random forest algorithm to improve Ceutorhynchus napi GYLL. (Coleoptera: Curculionidae) occurrence forecasting

Abstract

AbstractRandom Forest algorithm was used to predict on‐field presence probability of rape stem weevil in France as a function of climatic and landscape variables, based on a long‐term and multisite data set. A first version of the model included a set of 342 variables. A variable selection procedure was used to retain only the 15 most influential variables without significant drop in predicting performances. Most retained variables were temperature related and results showed that the sum of maximum daily temperature above 9°C during the week preceding observation was the predictor with the largest influence on rape stem weevil occurrence. This model reached a mean AUC of 0.77 and outperformed some other published models. As such, this model can help farmers to precisely time insecticide application. It has been integrated in a decision support system freely available in the Terres Inovia (French applied agricultural research and development institute dedicated to oilseed crops) website.