Comparing data driven models versus numerical models in simulation of waterfront advance in furrow irrigation

Abstract

Accurate design, appropriate management, and knowledge of relationships between the parameters affecting on the performance of a surface irrigation system are the factors which play an effective role in increasing the efficiency of these systems. If parameters such as advance distance can be well estimated per specified flow rate, the volume of infiltrated water can be estimated, thereby preventing water loss and enhancing irrigation efficiency to a great extent. In the present study evaluated the accuracy of data-driven methods Random Forest (RF), Artificial Neural Networks (ANN), Adaptive Neuro Fuzzy Inference System (ANFIS), and M5 Model Tree and common numerical methods such as the Full hydrodynamic and Zero-inertia model (using SIRMOD software) and Zero-inertial model (using WinSRFR software) to predict the advance distance in furrow irrigation. To this end, seven series of data resulting from the evaluation of furrow irrigation system in various regions were collected. Each series included 12 input parameters of furrow length (L), furrow geometrical cross-section coefficients ( $$\sigma_{1} ,\sigma_{2}$$ ), furrow hydraulic cross-section coefficients ( $$\rho_{1} ,\rho_{2}$$ ), inflow rate (Q), Maning’s coefficient (n), field slope ( $$S_{0}$$ ), cut-off time $$(T_{\text{cutoff}} )$$ , final infiltration rate ( $$f_{0}$$ ), and the infiltration parameters of the Kostiakov equation (a and k). Comparison of the results showed that all the data-driven methods managed to estimate the advance distance of the wetting front in the furrow with higher accuracy than the numerical methods. From among these, the ANFIS model had the highest accuracy (RMSE = 1.842 m, MAE = 1.305 m) in estimating the advance distance in the furrow.