Assimilation versus optimization for SWAT calibration: accuracy, uncertainty, and computational burden analysis

Abstract

The accurate estimation of runoff by hydrological models depends on proper model calibration. Sequential Data Assimilation (DA), as an online method, is used to estimate complex hydrological models' states and parameters simultaneously. Although DA was applied for estimating the Soil and Water Assessment Tool (SWAT) model's state and/or parameter, the previous research did not pay attention to the model calibration by DA or the comparison between DA and popular SWAT calibration methods. This paper compares Ensemble Kalman Filter (EnKF), a well-known DA method, with Sequential Uncertainty Fitting (SUFI2), a popular SWAT calibration method, to calibrate the model. We test the impact of the selected objective function in the SUFI2 application. We evaluate the results based on the multiple deterministic and uncertainty-based Goodness of Fit (GOF) measures and compare all scenarios based on the simulation accuracy, computational burden, uncertainty assessment, and parameter ranges. Results show that under the SUFI2 application, some GOFs might be located in unsatisfactory ranges while the algorithm obtains (very) good results concerning the objective functions. On the other hand, EnKF simultaneously locates most GOFs in (very) good ratings. Moreover, we found that the selection of SUFI2's objective function and the specification of uncertainty's error in EnKF have significant effects on the results. HIGHLIGHTS We proposed EnKF as a calibration method for estimating the broad number of SWAT model parameters.; We compared EnKF with SUFI2, a popular SWAT calibration method, based on the computational burden, deterministic, and uncertainty-based GOF measures.; Compared to SUFI2, EnKF has less computational burden and leads to better results concerning different GOFs.;

Keywords

• data assimilation
• ensemble kalman filter
• sequential calibration
• sufi2
• swat