Potato late blight (Phytophthora infestans) disease forecasting using an auto-encoded long short-term memory recurrent neural networks in North-Western Algeria

Abstract

Potato late blight, caused by Phytophthora infestans (Mont.) de Bary, is the most destructive disease that threatens potato producers in North-Western Algeria. In these areas, it is not rare that potato growers miss crucial treatments or apply unnecessary ones during growing seasons. To address this situation, the introduction of a scientific-based decision support system should improve the disease management strategy and reduce growing costs to farmers, consumers, and protect the environment. For this purpose, we set up a supervised deep-learning classification process (long-short-term-memory/autoencoder) that classify and predict early late blight outbreak 30 days ahead of its occurrence based on a set of determinant climatic factors. The classification results of the long short term memory-autoencoders (LSTM’s-AE) regarding our multi-annual potato late blight disease occurrence database confirms that the unimodal BlightT_LSTM-AE model that accounts for temperature is the overall best model and temperature is the key parameter to a valid prediction of early late blight disease outbreak, in all targeted regions.