Modeling Zn Availability and Uptake by Citrus Plants Using Easily Measured Soil Characteristics

This study evaluated the relationship between Zn in citrus leaves and some easily available soil properties using artificial neural network (ANN), gene expression programming (GEP), and stepwise regression methods. The data of 40 soil samples collected at the South of Kerman region (Iran) were used. Different combinations of input parameters, including soil texture, pH, organic carbon (OC), total neutralizing value (TNV), electrical conductivity (EC), and soil available P (P_olsen), were considered. The coefficient of determination (R²), the normalized root means square error (NRMSE), mean absolute error (MAE), and Nash–Sutcliffe efficiency (NS) were used to evaluate the accuracy of the models. The results showed that the models with whole input parameters offered the lowest NRMSE and the highest R². The ANN model with four input variables showed that 77% of the Zn variation in leaves could be elucidated by the explicit soil parameters, including clay, silt, OC, and P_olsen. These results showed that the ANN model with six neurons in the hidden layer had the best performance in modeling Zn uptake. However, since the main goal of this research was to improve the models based on easily measurable variables, the GEP model with three input variables, including silt, OC, and P_olsen, was found to be beneficial in estimating 71% of Zn-leaves variability.