Development of Production Functions for Important Cereals of Qazvin Plain under Water Shortage and Salinity Stress Using AquaCrop Model and Artificial Neural Network

In exploitation of low-quality water in arid and semi-arid regions, irrigation management is essential to increase water use efficiency. Determination of crop-water-salinity production function is an essential tool for proper irrigation management. In this study, the AquaCrop model was first evaluated by considering 4 soil and water salinity levels and 4 deficit irrigation levels for the major cereal crops including wheat, barley, and corn in Qazvin Plain. The results showed that the coefficients of determination for wheat, barley, and corn yield were 0.97, 0.86 and 0.91, respectively. Therefore, the model can evaluate the performance in salinity and deficit irrigation conditions with a good approximation. To determine the optimal production functions of each crop, the results of the plant model were compared with three models of linear and nonlinear regression, and artificial neural network. The neural network model was able to estimate the performance compared to the AquaCrop model with lower error and higher correlation (0.99). These values in the linear function for wheat, barley, and corn were 0.98, 0.95, and 0.78 and in the nonlinear function as 0.92, 0.86 and 0.81, respectively. Also, the error calculated in the neural network method for wheat, barley, and maize was 40.16, 62.09, and 57.08 kg, respectively, which were less than the linear model by 75 %, 70 %, and 95 %; and less than the exponential model by 90 %, 85 %, and 93%, respectively. The best trained network for determining the water-salt production function for barley and wheat 5 Nero and for corn 7 Nero was introduced in the single layer structure. Sensitivity analysis on wheat and barley showed that this model had low sensitivity to irrigation and salinity parameters and only corn plant showed a moderate range sensitivity to salinity parameter.