Application of Bayesian Network in Analysis and Management of Agricultural Water - Taking Kinmen for Example

By calculating the water demand and programming a fine irrigation project, the management and cultivating efficiency of traditional agriculture can be greatly improved. Taking rotational irrigation for example, the efficiency of irrigation can be maximized by adjusting water distribution routes, irrigation area allocation, and irrigation schedule planning. However, in actual operation, some problems are often encountered, such as how to persuade farmers and promote the designed irrigation project, and the negotiation of various stakeholders. Generally, due to the complexity of the irrigation design model, it is impossible to have an effective and immediate communication or presentation. Therefore, this study introduces the Bayesian network to presents the key points of the irrigation project after simplifying the relationship. In addition to being simpler for stakeholders to understand, it is also possible to adjust various parameters in time to obtain rough estimation results.

The research area of this study is a 100-hectare farmland, which is located in Kinmen County, Taiwan. For many years, local farmers have only relied on precipitation to cultivate sorghum, wheat and other crops. However, the precipitation in Kinmen is semiarid and unstable. In the past five years, the annual rainfall has been lower than the average in previous years, which directly led to a very bleak crop harvest. Therefore, we hope to establish an irrigation project in Kinmen, using recycled water as the water source to provide local farmers with a reliable water source.

The Bayesian network used in this study is a directed acyclic graphical (DAG) model based on conditional probability and Bayesian theorem to express the possible relationship between variables. In terms of operation, the different influencing factors in the research topic are converted into nodes, and the relationship between nodes is given by different conditional probabilities. This study uses GeNIe to establish a Bayesian network that can be used to estimate water profit and loss and other results. This Bayesian network can be divided into four sub-blocks, which are the relevant data of the irrigation area, the water demand, the water supply, and the final result calculation. Therefore, when the stakeholders are negotiating the irrigation project, they can discuss the different estimation results by adjusting each node of the first three sub-blocks.