Potato yield modeling based on meteorological factors using discriminant analysis and artificial neural networks

Abstract

ABSTRACT A reliable, pre-harvest, crop yield prediction based on meteorological factors is important to anticipate adverse effect of weather variables. Discriminant score-based regression models, MLP artificial neural network (ANN) models, and regression-ANN hybrid models were used to model potato (Solanum tuberosum L.) yield. Maximum and minimum temperatures, rainfall, and relative humidity, and their indices, were used to obtain discriminant scores for each year. These discriminant scores, along with a time variable, were used as inputs and potato yield as outputs for the development of models. A hybrid model consisting of linear and non-linear components performed better than individual models if combined linearity and nonlinearity are present in the data, else the ANN models were better than regression models. The best models can be used to obtain a reliable forecast of potato yield at 6–8 weeks before harvest using meteorological data.