Simulation of nitrate distribution under drip irrigation using artificial neural networks

Abstract

Accurate knowledge of nitrate distribution in the soil under fertigation through drip-irrigation systems is fundamentally important for system design and management. The determination of nitrate distribution through modeling represents a highly complex nonlinear problem that includes adsorption, transformation, convection, and dispersion. For this reason, an alternative methodology is proposed, which combines artificial neural networks (ANN) and laboratory experiments. Seventeen experiments with apparent discharge rates varying from 0.6 to 7.8 l/h, the apparent cylindrical applied volume from 6 to 15 l, and the input concentration from 100 to 700 mg/l were conducted to provide a database for establishing the ANN architecture. The model input parameters were initial soil water content, initial nitrate concentration in the soil, discharge rate, input concentration of fertilizer (NH4NO3), applied volume, and final soil water content. The model output was nitrate concentration in the soil after fertigation. A total of 298 vectors were used to train the ANN model, and 212 independent vectors were used to test the model. Results of the test show a good correspondence with a determination coefficient (r 2) of 0.83 between the model-estimated nitrate concentration in the soil and laboratory-measured nitrate concentration in the soil. These results show that the optimized ANN models are reasonably accurate and can provide an easy and efficient means of estimating nitrate distribution in the soil under fertigation through drip-irrigation systems.