Parameter uncertainty analysis in watershed total phosphorus modeling using the GLUE methodology

Deterministic watershed models are frequently used for agricultural non-point source (NPS) pollution simulations. However, parameter uncertainty should be analyzed before the modeling results are used to make decisions regarding watershed NPS pollution control programs. In this study, the Soil and Water Assessment Tool (SWAT) was used to simulate the total phosphorus (TP) loads caused by NPS pollution in the upper Daning River Watershed in China's Three Gorges Reservoir Area. The Generalized Likelihood Uncertainty Estimation (GLUE) methodology was used to analyze the parameter uncertainty in SWAT modeling. The impacts of three subjective options of GLUE, the parameter ranges, the level of confidence, and the threshold value of the likelihood measure, on the parameter uncertainty analysis results were analyzed. Specifically, we investigated if there was a combination of these factors that was most appropriate for expression of the uncertainty assessment results. The results indicated that the “observed data” may not always lie within the confidence intervals of GLUE, so the confidence interval was not sufficient to represent the uncertainty for the specific requirements of this study. Therefore we suggest there should be alternative measures to express the parameter uncertainty of GLUE.

► The initial parameter ranges and the 5–95% bands were the best in this study.
► CV and widths of the uncertainty bands decreased as the threshold value increased.
► The confidence interval had difficulties in reflecting a part of the uncertainty.
► Alternative expressions that could reflect that part of uncertainty may be needed.