The Implementation of a Bayesian Network for Watershed Management Decisions

Recently, the U.S. EPA issued the 303(d) list of impaired waters in Idaho State that contained the causes of impairment. This 303(d) list provides useful information that can be used to determine the Total Maximum Daily Loads (TMDLs). Implementation of TMDLs should result in pollutant reductions, which, in turn can lead to the restoration of these water bodies. Flow alteration is one of the potential sources of impairments in the Big Lost River in south-central Idaho, which have some negative impacts on the water quality and beneficial uses. Flow in the Big Lost River is altered, both in quantity and quality, and this reduces recreation activities, affects the fish assemblage, and changes the composition and relative abundance of aquatic species. The effect of riparian vegetation is another factor that needs to be predicted. In addition, three conservation schemes (construction of upstream reservoirs, downstream reservoirs, and canal linings) were proposed to restore flow in the downstream reaches of the river and compensate for water loss during the low flood seasons. However, there is no single predictive model that can be used to appropriately represent each of these issues as management decisions. In this paper, an expert system in the form of a Bayesian network, a graphical diagram of nodes and arcs, was implemented to examine all significant water management variables and relationships among these variables. Lining the irrigation canals was found to be the best scheme, followed by constructing an upstream reservoir. The TMDLs would benefit the water quality in the watershed but would not significantly increase the water quantity and solve the flow alteration problem. Consequently, this can be used to determine the sequence of decisions that can be taken in the future. Copyright Springer Science + Business Media, Inc. 2006