Predicting Potato Late Blight Disease Pressure using Artificial Neural Network

Abstract

Late blight causing Phytopthora infestans (Mont.) de Bary largely depends on weather parameters such as temperature and relative humidity for survival, spread and its ability to attach and infect new plants. The variations in weather across different agro-ecological zones can be used to explain the different levels of disease severity experienced across these regions. Potato late blight disease evaluation data from five locations was coupled with GISlinked weather data. A series of neural network models were developed and validated with 10-fold cross validation and the optimal model selected based on accuracy achieved on validation set. The selected model had 1 hidden layer with 14 nodes achieving an accuracy of 88% in the validation set. The final model was used to predict disease severity with 89% accuracy on new data. It was also found that the number of precipitation days and number of days with temperature and relative humidity favorable to disease development were amongst the top significant variables in the model hence a target for monitoring. This model can be used to estimate the expected late blight severity in a target region hence support the decisions on the appropriate varieties and management regimes to be used, reducing yield loss and excessive use of fungicides.