Optimizing plant traits to increase yield quality and quantity in tobacco using artificial neural network

There are complex inter- and intra-relations between regressors (independent variables) and yield quantity (W) and quality (Q) in tobacco. For instance, nitrogen (N) increases W but decreases Q; starch harms Q but soluble sugars promote it. The balance between (optimization of) regressors is needed for simultaneous increase in W and Q components [higher potassium (K), medium nicotine and lower chloride (Cl) contents in cured leaf]. This study was aimed to optimize 10 regressors (content of N and soluble sugars in root, stem and leaf, leaf nicotine content at flowering and nitrate reductase activity (NRA) at 3 phenological stages) for increased W and Q components, using an artificial neural network (ANN). Two field experiments were conducted to get diversified regressors, Q and W, using 2 N sources and 4 application patterns in Tirtash and Oromieh. Treatments and 2 locations produced a wide range of variation in regressors, W and Q components which is prerequisite of ANN. The results indicated that configuration of 12 neurons in one hidden layer was the best for prediction. The obtained optimum values of regressors (1.64%, 2.12% and 1.04% N content, 4.32%, 13.04% and 9.54% soluble sugar content for leaf, stem and root, respectively; 2.31% nicotine content and NRA of 13.11, 4.74 and 4.70 µmol.NO2.g-1.h-1 for pre-flowering, flowering and post-flowering stages, respectively) increased W by 3% accompanied by 4.75% K, 1.87% nicotine and 1.5% Cl in cured leaf.