Skip to main content
ARS Home » Plains Area » Temple, Texas » Grassland Soil and Water Research Laboratory » Research » Publications at this Location » Publication #297066

Title: A framework for propagation of uncertainty contributed by parameterization, input data, model structure, and calibration/validation data in watershed modeling

Author
item YEN, HAW - Texas A&M Agrilife
item WANG, XIUYING - Texas A&M Agrilife
item FONTANE, DARRELL - Colorado State University
item Harmel, Daren
item ARABI, MAZDAK - Colorado State University

Submitted to: Journal of Environmental Modeling and Software
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/1/2014
Publication Date: 2/4/2014
Publication URL: https://handle.nal.usda.gov/10113/59501
Citation: Yen, H., Wang, X., Fontane, D.G., Harmel, R.D., Arabi, M. 2014. A framework for propagation of uncertainty contributed by parameterization, input data, model structure, and calibration/validation data in watershed modeling. Journal of Environmental Modeling and Software. 54:211-221.

Interpretive Summary: Recent improvements in computer science and development of automated-calibration techniques means that calibration of simulation models (which is the process of adjusting the model to better match real world conditions) is no longer a major challenge for watershed planning and management. Modelers now increasingly focus on challenges such as improved representation of watershed processes and on various sources of uncertainty that contribute to the uncertainty in model predictions. In the past modelers assumed that all prediction uncertainty was contributed by model parameters chosen to represent real-world processes; however, ignoring the additional sources of uncertainty may bias model predictions. In this study, a framework was developed to incorporate uncertainty contributed by parameterization, as well uncertainty from input data, model structure, and calibration/validation data jointly. With this framework, the influence from each uncertainty source can be identified and analyzed to target further investigation into watershed processes and/or their representation in models. Application of this framework showed that input data uncertainty generally has a greater impact on predictive uncertainty than other sources. In addition, the ranges of predictive uncertainty are significantly increased by including all four sources of uncertainty simultaneously. Moreover, calibrated parameter sets tended to more realistically represent watershed behavior when all uncertainty sources were included simultaneously. The proposed framework is an innovative tool to investigate and explore individually and jointly the significance of uncertainty sources, which enhances watershed modeling by improved characterization and assessment of predictive uncertainty.

Technical Abstract: The progressive improvement of computer science and development of auto-calibration techniques means that calibration of simulation models is no longer a major challenge for watershed planning and management. Modelers now increasingly focus on challenges such as improved representation of watershed processes and the various sources of uncertainty that contribute to prediction uncertainty. The common assumption that all prediction uncertainty is contributed by model parameterization is inadequate, and the failure to consider the additional sources of uncertainty may bias model predictions. In this study, a framework was developed by the application of Bayesian inferences and other approaches to incorporate uncertainty contributed by parameterization, input data, model structure, and calibration/validation data jointly. With this framework, the influence from each uncertainty source can be identified and analyzed to target further investigation into watershed processes and/or their representation in models. Application of this framework showed that input data uncertainty generally has a greater impact on predictive uncertainty than other sources. In addition, the ranges of predictive uncertainty are significantly increased by including all four sources of uncertainty simultaneously. Moreover, calibrated parameter sets as indicated by goodness-of-fit statistics results tend to more realistically represent watershed behavior when all uncertainty sources are included simultaneously. The proposed framework is an innovative tool to investigate and explore individually and jointly the significance of uncertainty sources, which enhances watershed modeling by improved characterization and assessment of predictive uncertainty.