Tradeoffs among watershed model calibration targets for parameter estimation

Hydrologic models are commonly calibrated by optimizing a single objective function target to compare simulated and observed flows, although individual targets are influenced by specific flow modes. Nash‐Sutcliffe efficiency (NSE) emphasizes flood peaks in evaluating simulation fit, while modified Nash‐Sutcliffe efficiency (MNS) emphasizes lower flows, and the ratio of the simulated to observed standard deviations (RSD) prioritizes flow variability. We investigated tradeoffs of calibrating streamflow on three standard objective functions (NSE, MNS, and RSD), as well as a multiobjective function aggregating these three targets to simultaneously address a range of flow conditions, for calibration of the Soil and Water Assessment Tool (SWAT) daily streamflow simulations in two watersheds. A suite of objective functions was explored to select a minimally redundant set of metrics addressing a range of flow characteristics. After each pass of 2001 simulations, an iterative informal likelihood procedure was used to subset parameter ranges. The ranges from each best‐fit simulation set were used for model validation. Values for optimized parameters vary among calibrations using different objective functions, which underscores the importance of linking modeling objectives to calibration target selection. The simulation set approach yielded validated models of similar quality as seen with a single best‐fit parameter set, with the added benefit of uncertainty estimations. Our approach represents a novel compromise between equifinality‐based approaches and Pareto optimization. Combining the simulation set approach with the multiobjective function was demonstrated to be a practicable and flexible approach for model calibration, which can be readily modified to suit modeling goals, and is not model or location specific. Key Points Varied calibration targets result in substantially different flow simulations We present a flexible methodology that can be adapted for various modeling goals Our simulation set method was an improvement on optimized parameter values