Investigations of uncertainty in SWAT hydrologic simulations: a case study of a Canadian Shield catchment

Abstract

Uncertainty analysis is an integral part of hydrological modelling. We investigated uncertainty from model structure and parameters in the soil and water assessment tool (SWAT) hydrologic simulations for a Canadian Shield catchment (5.4 km2) in south central Ontario. We investigated influences of model structure on parameter identifiability and prediction uncertainty by comparing model performance of SWAT (version 2009.10.1 Beta3) and SWAT‐CS (a version representing hydrological processes in Canadian Shield catchments). Equifinality was found to exist for many parameters in process modules of snow water equivalent (SWE), streamflow and lake outflow. For SWAT, there are clearly identifiable parameters from process modules of snowmelt, overland flow, lateral flow and groundwater flow, while snowmelt parameters and lateral flow travel time are the only clearly identifiable parameters in SWAT‐CS. Model structure also notably influenced the optimum parameter values. Only 50–55% of observed SWE and 27–43% of observed streamflow were bracketed by the corresponding 95% confidence interval. The prediction uncertainty for SWE was mainly caused by the inaccuracy in timing of simulated snowmelt. This, along with use of a daily time step that is not able to capture the subdaily rainfall on snow pack, the limited capability of simulating groundwater by SWAT and the close‐to‐zero streamflow during dry seasons at the study catchment, could all contribute to the prediction uncertainty in streamflow. Comparatively, more reasonable model structure (SWAT‐CS) was shown to reduce prediction uncertainty. Copyright © 2015 John Wiley & Sons, Ltd.

Constraining SWAT Calibration with Remotely Sensed Evapotranspiration Data

“…The issue of equifinality can be especially acute for distributed and semi‐distributed hydrologic models such as SWAT. The equifinality issue can be addressed by either reducing the number of calibration parameters ( e.g ., Moreda et al ., ; Fu et al ., ; Her and Chaubey, ) or adding more observations (Qiao et al ., , ). A common approach in watershed modeling is the addition of additional streamflow observations at an interior gauge inside the basin ( e.g ., Lerat et al ., ; Qiao et al ., ).…”

A Modeling Study of Direct and Indirect N2O Emissions From a Representative Catchment in the U.S. Corn Belt

“…Uncertainties in hydrological modeling stem from model structure, parameters, and observational data for calibration and validation (Fu et al, ), and the present study is no exception. Compared to the direct N 2 O emissions from land surface, the understanding about the mechanism of indirect N 2 O emissions from streams is still limited, resulting in large uncertainties in the modeling of indirect emissions.…”

“…The maximum NSEs for calibration and validation of daily Q ([ ]) were 0.56 (0.51) and 0.68 (0.33), respectively. In this study, identifiable parameters were qualitatively judged from their dotty figures: the more concentrated the pattern of dots was near the upper envelope of the NSE values, the more identifiable the parameter was (Fu et al, ). Each parameter set was composed of 24 parameters, and the x axis and y axis of each panel of the dotty plot corresponded to the value of each parameter and the NSE value calculated using the parameter, respectively.…”

“…A total of 24 parameters in SWAT‐N 2 O were optimized against field data on Q and [ ] (Table ). The parameter selection and corresponding ranges were based on the studies of Zhang et al () and Fu et al (). Because SWAT‐N 2 O involves new parameters, the default SWAT calibration module (SWAT‐CUP) is no longer appropriate.…”

Stochastic response surface methods for supporting flood modelling under uncertainty

“…Previously, a large number of studies were conducted to address the forward uncertainty problems and diversified methodologies were developed (Balzter, 2000;Pappenberger et al, 2008;Peintinger et al, 2007;Beven and Hall, 2014;Fu et al, 2015;Jung and Merwade, 2015). Meanwhile, more and more concerns have been put on the inverse problems, especially for conditions where a robust predictive system is strongly sensitive to some parameters with little information being known before-hand.…”