Development of Inductive Receiving Water Model for Application in TMDLs

Various receiving water models, such as HSPF, WASP5, and CE-QUAL are frequently used in the development of TMDLs, especially in the context of nutrient loadings and DO impacts. In most cases, such models can be extremely data intensive and difficult to use. This paper will discuss the development of two inductive models for DO response to nutrient loadings and its application in the development of a nutrient TMDL for the DO impaired Beargrass Creek in Louisville Kentucky. The associated models were developed using two separate AI modeling techniques: artificial neural networks and genetic programming/genetic algorithms. Data for use in constructing the two models was obtained from continuous water quality monitors that were strategically placed in the downstream reaches of the watershed as well as other discrete sampling for water quality constituents. Modeling of the system response was complicated by backwater effects from the Ohio River. The paper discusses the utility and advantages of use of inductive approach when adequate data sets are readily available.