Predicting rainfed wheat quality and quantity by artificial neural network using terrain and soil characteristics

Terrain attributes and soil properties are the major parameters to be considered for sustainable agricultural productivity and landscape management in dry farming. The objectives of this study were to predict the biomass, grain yield, and grain protein in wheat grown in hilly regions using an artificial neural network model and to identify the most important topographic attributes and soil properties affecting the variability of the wheat-yield components under rainfed conditions in the semiarid and hilly regions of western Iran. Wheat-yield data and surface-soil samples (0-30 cm depth) were collected from 1-m2 plots at 100 selected points. The sampling points were chosen in a stratified random manner on given geomorphic surfaces including summit, shoulder, backslope, footslope, and toeslope at the site. Primary and secondary terrain attributes were calculated using the digital elevation model. Selected soil physical and chemical properties were measured for the collected soil samples. Prediction of the artificial neural network models in the study area resulted in R 2 and root mean square error values of 0.95 and 0.022 for biomass, 0.93 and 0.021 for grain yield, and 0.89 and 0.063 for grain protein, respectively. The sediment-transport index was identified as the most important topographic attribute influencing the wheat biomass, while grain yield and grain protein were affected the most by the total nitrogen. Overall, our results indicated that the artificial neural network models could explain 89-95% of the total variability in wheat biomass, grain yield, and grain protein content. Also, the predictability of wheat yield and quality could be further improved by considering management practices followed during the growing season.