A probabilistic appraisal of rainfall-runoff modeling approaches within SWAT in mixed land use watersheds

Authors: TASDIGHI, ALI - Colorado State University; ARABI, MAZDAK - Colorado State University; Harmel, Daren

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/14/2018
Publication Date: 7/17/2018
Citation: Tasdighi, A., Arabi, M., Harmel, R.D. 2018. A probabilistic appraisal of rainfall-runoff modeling approaches within SWAT in mixed land use watersheds. Journal of Hydrology. https://doi.org/10.1016/j.jhydrol.2018.07.035.
DOI: https://doi.org/10.1016/j.jhydrol.2018.07.035

Interpretive Summary: This study investigates the performance validity of the two methods (CN and G&A) to determine runoff within the SWAT watershed model under different upstream land use conditions. An uncertainty assessment framework, which accounts for uncertainties from model parameters, inputs, structure, and measurements, was employed for exploring uncertainties in streamflow simulation using SWAT with different rainfall-runoff methods in a mixed-land use watershed. The model was calibrated only at the outlet of the watershed, while the performance of the methods was compared at different stream locations inside the watershed with various upstream land use conditions. At the watershed outlet, the methods performed similarly in simulating streamflows. Analogous results were obtained for the forested and agricultural subwatersheds. However, in subwatersheds with higher percentage of developed land, the G&A outperformed the CN method in simulating streamflow. In general, the G&A method better simulated streamflow especially during the high flow events. However, this came at the cost of slightly more prediction. Regarding the annual hydrologic budget, the CN method generated higher water yield contributing to streamflow especially at subwatersheds with highly developed land use. The results of this study have important implications regarding determination of the rainfall-runoff method within distributed hydrologic models that generates more reliable streamflow simulations in mixed-land use watersheds. For example in the present analysis, the CN and G&A methods in SWAT performed similarly at the outlet of a mixed-land use watershed; however, G&A captured the internal processes more realistically.

Technical Abstract: This study investigates the performance validity of the empirical Curve number (CN) and physically-based Green and Ampt (G&A) rainfall-runoff methods in SWAT under different upstream land use conditions using a probabilistic framework. A Bayesian total uncertainty assessment framework, which explicitly accounts for uncertainties from model parameters, inputs, structure, and measurements, was employed for exploring uncertainties in streamflow simulation using SWAT with different rainfall-runoff methods in a mixed-land use watershed. The models’ training was conducted only at the outlet of the watershed, while the performance of the methods was compared at different stream locations inside the watershed with various upstream land use conditions. At the watershed outlet, CN and G&A performed similarly in simulating streamflow with CN showing slightly better error statistics. Analogous results were obtained for the forested and agricultural subwatersheds. However, in subwatersheds with higher percentage of developed land, G&A outperformed the CN method in simulating streamflow based on various metrics. In general, the 95% confidence interval from the G&A simulations covered a higher percentage of observed streamflow especially during the high flow events. However, this came at the cost of slightly wider bands of uncertainty. Regarding the annual hydrologic regime, the CN method generated higher water yield contributing to streamflow especially at subwatersheds with highly developed land use. The deficiency of the CN method in capturing the high flow events in highly developed subwatersheds resulted in framework’s general tendency to sample parameter sets that generate higher streamflows in an attempt to capture peak flows. The results of this study have important implications regarding determination of the rainfallrunoff method within distributed hydrologic models that generates more reliable streamflow simulations in mixed-land use watersheds. For example in the present analysis, the CN and G&A methods in SWAT performed similarly at the outlet of a mixed-land use watershed; however, G&A captured the internal processes more realistically.