How to: understanding SWAT model uncertainty relative to measured results

Authors: ZEFF, HARRISON - Pennsylvania State University; Veith, Tameria - Tamie

Submitted to: Soil and Water Assessment Tool International Conference
Publication Type: Proceedings
Publication Acceptance Date: 8/5/2009
Publication Date: 12/18/2009
Citation: Zeff, H., Veith, T.L. 2009. How to: understanding SWAT model uncertainty relative to measured results. In Proceedings of the Soil and Water Assessment Tool International Conference, August 5-7, 2009, Boulder, Colorado. p. 171-179.

Interpretive Summary: An interpretive summary is not required.

Technical Abstract: Watershed models are being relied upon to contribute to most policy-making decisions of watershed management, and the demand for an accurate accounting of complete model uncertainty is rising. Generalized likelihood uncertainty estimation (GLUE) is a widely used method for quantifying uncertainty in hydrological models because of its ease of use and versatility in adaptation. In this paper, we examine the assumptions of GLUE and how they can be used to gain an understanding of the different parts of model uncertainty using the hydrologic model SWAT. Although GLUE, like other uncertainty methods, makes no attempt to account for the inherent uncertainty in model structures, structural uncertainty is addressed by evaluating GLUE uncertainty bounds for predictive capacity and the uncertainty range required to achieve this predictive capacity. It is found that while parameter uncertainty is overestimated using both formal (Bayesian) and informal likelihoods, formal likelihoods have the ability to reduce these overestimations with increased information and provide a more accurate understanding of the structural uncertainty inherent in SWAT.