Neural network modeling of chickpea grain yield on ameliorated soils

Abstract

The aim of the research is to improve the management practices for chickpea growing upon ameliorated soils in typical climate conditions of arid steppe of Ukraine. For precision programming of chickpea grain yield depending on four factors (water consumption, mineral feterlizers, plant density, and the depth of soil primary tillage) generalized regression method of artificial neural network GRNN (4-54-2-1) with 54 neurons in the first buried layer and two neurons in the second layer was applied; productivity neural network training is 0.22; control - 0.37; test - 0.36; training error - 0.29; control - 0.45; test - 0.47. Multiple correlation considering non-linear patterns of the impact of studying factors on chickpea grain yield was 0.96. Asymmetry parameters of actual and calculated yield were 0.37 and 0.23, respectively. The accuracy of the simulation was 92.08 %. Cross-validation of predictive models was done using statistical criteria for significance: mean error, mean absolute error, standard error deviation, average relative error, correlation coefficient. For processing the modifications of such software as Statistica Advanced and Automated Neural Networks for Windows v.10 Ru were used. Non-linear patterns of the impact of studying factors on the dynamics of forming the chickpea grain yield were determined: water consumption - 37.01 %; mineral fertilizer applying - 22.88 %; plant density - 22.29 %; depth of the primary soil tillage - 17.82 %. The results of neural network modeling presented in the work can be used for four-factor precise programming of chickpea grain yield on ameliorative soils in typical climate conditions of arid steppe.