Submitted to: Proceedings of the American Society of Agricultural and Biological Engineers International (ASABE)
Publication Type: Proceedings
Publication Acceptance Date: March 29, 2008
Publication Date: April 3, 2008
Citation: Shirmohammadi, A., Sexton, A., Montas, H., Sadeghi, A.M. 2008. Uncertainty consideration in watershed scale models. Proceedings of the American Society of Agricultural and Biological Engineers. Publication No. 701P0208.
Watershed scale hydrologic and water quality models have been used with increasing frequency to devise alternative pollution control strategies. With recent reenactment of the 1972 Clean Water Act’s TMDL (total maximum daily load) component, some of the watershed scale models are being recommended for TMDL assessments on watershed scale. However, it has been recognized that such models may have a large degree of uncertainty associated with their simulations, and that this uncertainty can significantly limit the utility of their output. This study uses two uncertainty methods in assessing the uncertainty in SWAT model’s output due to variability in input parameter values in a small watershed (Warner Creek Watershed) located in northern Maryland. Both Latin Hypercube Sampling (LHS) with constrained Monte Carlo Simulation (MCS) technique and Mean Value First Order Reliability Method (MFORM) were utilized. Additionally, results obtained with MFORM were used to evaluate the margin of safety (MOS) in the TMDL assessment for the selected watershed. Results showed that using average parameter values for the watershed without considering their variability may result in significant uncertainty in SWAT’s simulated streamflow, sediment, and nitrate-nitrogen. Results also indicated the capability of MFORM in capturing the uncertainty in SWAT’s simulations and identifying the most sensitive parameters. In addition, results of MFORM were successfully used to identify nutrient reduction rates that are necessary to meet watershed TMDL criteria with acceptable level of confidence. This study concluded that using a best possible distribution for the input parameters is much preferred over using an average value.