Beyond model metrics: The perils of calibrating hydrologic models

Authors: ACERO TRIANA, JUAN - University Of Illinois; CHU, MARIA - University Of Illinois; GUZMAN, JORGE - University Of Illinois; Moriasi, Daniel; Steiner, Jean

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/8/2019
Publication Date: 8/10/2019
Citation: Acero Triana, J.S., Chu, M.L., Guzman, J., Moriasi, D.N., Steiner, J.L. 2019. Beyond model metrics: The perils of calibrating hydrologic models. Journal of Hydrology. https://doi.org/10.1016/j.jhydrol.2019.124032.
DOI: https://doi.org/10.1016/j.jhydrol.2019.124032

Interpretive Summary: Hydrologic and water quality models have been used to quantify the impacts of climate variability, land use change, land management practices and conservation practices on soil and water resources. Since outputs from research modeling studies play a vital role in regulatory, planning, research, and decision-making activities, it is essential that these models be calibrated properly. The goal of this study was to develop a comprehensive calibration procedure for integrated surface-groundwater models. To illustrate this procedure, data from the Fort Cobb Reservoir Experimental Watershed (FCREW) located in Oklahoma, USA was used to calibrate the linked Soil and Water Assessment Tool (SWAT) and Modular Finite-difference Flow Model (MODLOW) model. The results indicated that the developed procedure, in which both the surface and groundwater model parameters are iteratively calibrated, provided a more realistic representation of the water balance at the FCREW using the linked SWAT-MODFLOW model. The calibration procedure developed in this study can serve as a general guideline for studies that encompass hydrologic modeling at both surface and subsurface domains.

Technical Abstract: The multi-metric assessment of model performance in a dominantly single-domain modeling approach (i.e., surface) may not be sufficient to overcome equifinality and avoid a distorted representation of the hydrologic system. Consequently, rating metrics can mathematically validate model results as satisfactory even when some of the simulated hydrologic processes are wrongly represented during the calibration process. This study exemplifies the perils of calibrating a semi-distributed model in a single domain using the widely used model metrics such as the Nash-Sutcliffe coefficient (NSE), PBIAS, and RMSE. The coupled Soil and Water Assessment Tool (SWAT) and the Modular Finite-difference Flow Model (MODLOW) were used to represent the surface water - groundwater interactions in the Fort Cobb Reservoir Experimental Watershed (FCREW) located in Oklahoma, USA. The NSE was employed as the initial objective function to calibrate the SWAT model. An iterative approach was then employed to calibrate SWAT and MODFLOW to reduce equifinality and ensure the appropriate representativeness of the coupled systems. Our results showed that a validated SWAT model, or any other model parameterized using a single-domain approach, does not guarantee a realistic representation of groundwater recharge patterns nor the groundwater levels. This issue calls into question the real capability of models that are parameterized using a single domain to represent the hydrologic processes in an open system. Accordingly, a full calibration of two or more domains is strongly recommended in which the consistency and the margin of errors of model simulations must be iteratively verified. This approach is expected to reduce equifinality based on the acceptable goodness-of-fit metrics and will result in a more accurate representation of the hydrologic processes. The calibration procedure developed in this paper can serve as a general guideline for studies that encompass hydrologic modeling at both surface and subsurface domains.