Development of Bayesian Network Model for Simulation and Assessment of Water Distribution and Delivery within Main Irrigation Canals

Irrigation and drainage networks are one of the most basic infrastructures of irrigated agriculture which are facing managerial and operational problems and necessarily needs to be improved. The performance of different operating methods in irrigation networks is influenced by various factors which can be analyzed by operational simulation and evaluation of water distribution and delivery. Mentioned processes are time and cost consuming, therefore, the primary purpose of this study is the development of a Hybrid Bayesian Network model to unify two mentioned processes, considering the different operational scenarios which can lead to time and cost saving. To achieve this goal, the hydrodynamic model of Roodasht irrigation network’s main canal was developed as a case study. Calibration and validation of the model were carried out based on the measured data in the crop year 2015-2016. The results of the calibration and validation of the HEC-RAS hydrodynamic model showed that the two processes were satisfactory. Besides, MAE, CRM, and RMSE indicators were 0.0027 cubic per second, -0.043 and 0.0028 cubic per second respectively in calibration and 0.0038 cubic per second, -0.061 and 0.004 cubic per second in the validation process. To develop an intelligent Bayesian model that can be used for different operational conditions, a wide range of available scenarios including standard (with canal inflow range variation of 4 to 4.5), water shortages (with canal inflow range variation of 3.36 to 4.49) and fluctuations (with canal inflow range variation of 0.44 to 6.6) were considered for model training. Accordingly, the inputs of the intelligent model for operational simulation and evaluation are “Canal Inflow” and “Scenario” and outputs are “Delivery Discharges of Off-takes” , “ Adequacy Indicators of Off-takes” , “Efficiency Indicators of Off-takes” , “Adequacy Indicators of Upstream, Midstream and Downstream of Canal”, “Efficiency Indicators of Upstream, Midstream and Downstream of Canal”, “Equity Indicators of Upstream, Midstream and Downstream of Canal”, “Adequacy, Efficiency and Equity Indicators of Canal” and “Desirability of Delivery and Distribution Indicator”. Eight different structure states were defined and verified to achieve the most appropriate network structure. The results of validation showed that the second structure as the superior model has a desirable accuracy. The average of MAPE and R2 indicators for all output nodes in the excellent model was 17.31%, and 0.96, respectively. Finally, it can be concluded that the Bayesian model presented in this study is a suitable model for operational simulation and assessment of water delivery and distribution in irrigation canals.

Development of Bayesian Network Model for Simulation and Assessment of Water Distribution and Delivery within Main Irrigation Canals Keywords:

operation of irrigation canals , operational simulation and evaluation intelligent , Bayesian Networks , Hybrid Bayesian Networks , intelligent models , surrogate models.