Utilization of Weather Data in Predicting Bread Loaf Volume by Neural Network Method

Abstract

The unique ability of wheat (Triticum aestivum L.) flour to produce cohesive dough has helped to make it the most widely used cereal crop for bread and other baked food products. Measurement of end‐use qualities, such as loaf volume, is ideally carried out through assessing loaves of bread; however, this is resource intensive. Predictive testing methods are more often utilized to identify wheat genotypes with potentially acceptable loaf volume, although more accurate predictive methods would be beneficial. Our objectives were to study the influence of weather observations on bread loaf volume and flour and dough quality data, to use a neural network (NN) model to predict loaf volume with select input data, and to compare the best multiple regression models identified with their NN counterparts. Weather data collected at 20 days after heading (DAH) showed the highest correlations with bread loaf volume when compared with prior intervals. A NN model containing maximum, minimum, and nighttime temperatures produced the highest coefficient of determination for predicting loaf volume. Our results showed that the NN model explained up to 20% more loaf volume variation than a similar regression model. Weather parameters representing conditions at 20 DAH played a significant role in loaf volume prediction.