A framework for parameter estimation, sensitivity analysis, and uncertainty analysis for holistic hydrologic modeling using SWAT+

Authors: ABBAS, SALAM - Colorado State University; BAILEY, RYAN - Colorado State University; WHITE, JEREMY - Gns Science; Arnold, Jeffrey; White, Michael; CERKASOVA, NATALJA - Texas A&M Agrilife; GAO, JUNGANG - Texas A&M Agrilife

Submitted to: Hydrology and Earth System Sciences
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/21/2023
Publication Date: 1/2/2024
Citation: Abbas, S.A., Bailey, R.T., White, J.T., Arnold, J.G., White, M.J., Cerkasova, N., Gao, J. 2024. A framework for parameter estimation, sensitivity analysis, and uncertainty analysis for holistic hydrologic modeling using SWAT+. Hydrology and Earth System Sciences. 28:21-48. https://doi.org/10.5194/hess-28-21-2024.
DOI: https://doi.org/10.5194/hess-28-21-2024

Interpretive Summary: Parameter sensitivity analysis is important for understanding how different input data and parameters affect watershed model predictions. This study uses a new method for analyzing the SWAT+ model, which helps predict surface and subsurface water interactions. We focus on four watersheds in the United States with different characteristics, like tile drainage and groundwater pumping for irrigation. To optimize the model, PEST and PEST++ were used. Model accuracy was assessed using different statistics, like the Nash-Sutcliffe efficiency coefficient and mean absolute error of groundwater head. The Morris method was used to identify the most important factors affecting water movement in each watershed. Some of these factors include recharge delay, aquifer hydraulic conductivity, and soil available water capacity. Including the new gwflow module in SWAT+ more accurately represents all important factors in the water system, and results can be different when using just SWAT+ alone. An iterative ensemble smoother technique was employed to reduce the computational cost caused by the number of factors in the model to make the process more efficient.

Technical Abstract: Parameter sensitivity analysis plays a critical role in effectively determining main parameters, enhancing the efficiency of parameter optimization, and quantifying model uncertainty in watershed modeling. In this study, we present a sensitivity analysis approach for the integrated SWAT+ model, augmented to include physically based spatially distributed groundwater modeling with the new gwflow module. Main computed groundwater inflows and outflows include pumping, groundwater–surface water exchange, saturation excess flow, recharge, groundwater–lake exchange, tile drainage outflow, boundary exchange. We present the method for four watersheds located in different areas of the United States for 16 years (2000–2015), emphasizing areas of extensive tile drainage (Winnebago River, Minnesota, and Iowa), intensive surface–groundwater interaction (Nanticoke River, Delaware, and Maryland), groundwater pumping for irrigation (Cache River, Missouri, and Arkansas), and mountain snowmelt (Arkansas Headwaters, Colorado). The key parameters of SWAT+ and the gwflow module were optimized using the parameter estimation software programs PEST and PEST++. The Nash–Sutcliffe efficiency coefficient (NSE), determination coefficient (R2), Kling–Gupta efficiency coefficient (KGE), and percentage bias (PBIAS) are used to evaluate monthly streamflow and mean absolute error (MAE) of groundwater head. The Morris method is used to identify the key parameters influencing hydrological processes. Depending on the watershed, key identified parameters include recharge delay, aquifer hydraulic conductivity, aquifer specific yield, streambed hydraulic conductivity, streambed thickness, runoff curve number, soil evaporation compensation factor, plant uptake compensation factor, percolation coefficient, and soil available water capacity, Including parameters from the gwflow module allows for the identification of all governing parameters in the surface/subsurface system, with results varying significantly if the stand–alone SWAT+ models are used. Additionally, the iterative ensemble smoother (iES) is used as a technique for Uncertainty Quantification (UQ) and Parameter Estimation (PE) to reduce the computational cost caused by the number of parameters.