Optimization of traits to increasing barley grain yield using an artificial neural network

The grain yield (Y) of crops is determined by several Y components that reflect positive or negative effects. Conventionally, ordinary Y components are screened for the highest direct effect on Y. Increasing one component tends to be somewhat counterbalanced by a concomitant reduction in other component (s) due to competition for assimilates. Therefore, it has been suggested that components be manipulated in conjunction with other traits to break the competition-resulting barrier. The objective of this study is to optimize the effective components in conjunction with certain participant traits for increased barley Y using an artificial neural network (ANN) and a genetic algorithm (GA) as an alternative procedure. Two field experiments were carried out separately at the Agriculture Research Station located in Gonbade Kavous (37o16' N, 55o12' E and 37 asl), Iran. Ten genotypes were grown in each experiment, and the Y and certain traits/components were measured. Among the components/traits, those with a significant direct effect and/or correlation with Y were selected as effective for the ANN and GA analysis. The results indicate that the remobilization of stored pre-anthesis assimilates to grain (R1), crop height (R2), 1,000-grains weight (R3), grain number per ear (R4), vegetative growth duration (R5), grain-filling duration (R6), grain-filling rate (R7), and tiller number (R8) were effective. The R2 for the training and test phases was 0.99 and 0.94, respectively, which reveals the capability of the ANN to predicting Y. The optimum values obtained by GA were 14.2%, 104.34 cm, 36.9 g, 41.9, 100 d, 48 d, 1.22 mg seed-1 day-1, and 3.38 plant-1 for R1 through R8, respectively. The optimization increased the potential Y to 5791 kg ha-1, which was higher than that observed for the genotypes (3527 to 5163 kg ha-1).