| description abstract | The choice of a model for total maximum daily load (TMDL) development for impaired water bodies depends mainly on the modeling objectives and system complexity. System complexity and modeling objectives, in turn, determine the required model complexity. Nonlinked or stand-alone complex models or linked models are generally selected for complex systems that consist of large watersheds with urban and rural areas and with a network of flow-controlled water impoundments (e.g., lakes and reservoirs) and estuaries. As an example, we present herein three case studies to demonstrate the role that complex nonlinked and linked models play in the development of complex TMDL studies. Case 1 is the Cannon River watershed, which uses a nonlinked complex model known as Hydrological Simulation Program—FORTRAN (HSPF), which is the core watershed model of the Better Assessment Science Integrating Point and Non-Point Sources (BASINS). Case 2 is the Chesapeake Bay watershed, which uses linked models: air-shed, watershed loading, and estuary models. Note that Case 2 uses the Chesapeake Bay Phase 5.3 Community Watershed Model, which is a version of the HSPF model. Case 3 is the Sougahatchee Creek watershed, which uses three linked models: Loading Simulation Program in C++ (LSPC), Water Quality Simulation Program (WASP), and Environmental Fluid Dynamics Code (EFDC). The results obtained from the case study reports show an absence of quantitative model performance metrics for Case 3 and limited performance results for Cases 1 and 2. For Case 2—a linked model case—model performance is only available for the loading model but not for the other linked models. The use of complex nonlinked models or linked models is not a guarantee of a good modeling practice by itself unless the models are supported by performance metrics, such as Nash–Sutcliffe efficiency (NSE) values that are greater than or equal to 0.65. Only Case 2 reported simulated water quality constituent NSE values for the watershed loading model, and reported values were less than zero. The low predictive performances gleaned from the case study reports can be attributed to inadequate system representation, model structure uncertainty, and poor data quality. Unless complex stand-alone and linked models are supported by accurate system representation and good quality input data, their application for TMDL development may not be scientifically defensible. | |
