Integrated risk analysis and management of agricultural water supply and distribution systems by using the Fuzzy Dynamic Bayesian Network (FDBN)

Due to the presence of multiple hazards in agricultural water supply systems, it is essential to develop a framework to analyze risk in these systems. This framework is critical for the development of sustainable agricultural systems. The aim of this study is to develop a multi-hazard risk assessment framework by applying Bayesian Networks to agricultural water supply and distribution systems. The structure of this model is based on assessing risk for supply systems and agricultural water distribution and studies them at different component levels: supply, distribution, and their subcomponents. This structure consists of nodes and their causal relations, designed to calculate the risk of supply and distribution separately in agricultural water systems. This is achieved by the following inputs: river inflow, water delivered, groundwater withdrawal, and demand. The developed model is applied to the Roodasht Irrigation district located in the province of Isfahan, Iran, to demonstrate its capability. This Irrigation district is under the threat of drought, improper performance of the ditch- riders, and operational losses. The amount of risk was 12.8% on average in the system, and it was 11.5% and 14.0% on average for supply and distribution components, respectively. Modeling this structure was first done by using Static Hybrid Bayesian Networks. The results showed that this model has good accuracy, with RMSE values of 0.07 and 0.07 and R2 values of 0.71 and 0.70 for the training and test phases, respectively. In the next step, we considered temporal relations between risk nodes for the duration of the dataset. A new model was developed by using Dynamic Bayesian Networks, and the model accuracy was 87%. The next step was to incorporate uncertainty in parameters by using Fuzzy logic, and therefore, we developed a new model using Fuzzy Dynamic Bayesian Networks. Results showed that this new model achieved 90% accuracy after learning the parameters. By comparing the accuracy of different suggested models, the Fuzzy Dynamic Bayesian Network model can be trusted for evaluating risk in the system, and it can replace complicated mathematical formulas and simulations. In the second part of this study, the Fuzzy Dynamic Bayesian Network model was used to manage risk in the system. Different scenarios were defined for reducing risk in the following categories structural, non-structural, automated control, and combined methods. These scenarios were developed using our Fuzzy Dynamic Bayesian Network model and were applied to the Roodasht irrigation district. Results showed that scenarios could reduce the risk of the system in the High class by 13.36% on average and reduce the average risk of the system between 5% to 11% on average. In the next step, other criteria, including economic, social, environmental, and technical, were designed to better analyze the scenarios. After that, by designing a questionnaire and getting help from experts, those criteria were weighted using the AHP method. We needed multi-criteria decision-making methods to rank scenarios and choose the best ones. The following methods were used: WASPAS, TOPSIS, and MultiMoora. Also, the Copeland method was used to combine results from those decision-making methods. The results suggested short-term and long-term solutions for managers. Automated control methods were ranked the highest in managing risk for the Roodasht irrigation district. Next in the ranking was the combined scenario of structural and water delivery scheduling. Overall, the proposed model and research findings can help decision-makers and operators to better understand potential failures of the agricultural water system and its components due to hazards, and to develop effective plans for managing water based on predicted risks of different hazards.