Calibration and uncertainty analysis of the SWAT model using Genetic Algorithms and Bayesian Model Averaging

• Published in: Journal of hydrology (Amsterdam) Vol. 374; no. 3; pp. 307 - 317
• Main Authors: Zhang, Xuesong, Srinivasan, Raghavan, Bosch, David
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier B.V 15.08.2009; [Amsterdam; New York]: Elsevier; Elsevier
• Subjects: ALGORITHMS; BASIC BIOLOGICAL SCIENCES; Basin; Bayesian model averaging; CALIBRATION; China; Earth sciences; Earth, ocean, space; ENVIRONMENTAL SCIENCES; Exact sciences and technology; Freshwater; GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; genetic algorithms; GENETICS; Georgia; hydrologic models; Hydrology; Hydrology. Hydrogeology; mathematics and statistics; model validation; Modeling; Optimization; RIVERS; SIMULATION; Soil and Water Assessment Tool model; SOILS; statistical analysis; stream flow; SWAT; Uncertainty; uncertainty analysis; VALIDATION; WATER; watershed hydrology; watersheds; YELLOW RIVER; China; Georgia; Huang Ho; Far East; Asia; United States; China, People's Rep., Huang He R; SWAT; Uncertainty; Modeling; Basin; Optimization; bayesian analysis; rivers; algorithms; models; digital simulation; case studies; North America; optimization; prediction; soils; calibration; uncertainties; streamflow
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2009.06.023

In this paper, the Genetic Algorithms (GA) and Bayesian Model Averaging (BMA) were used to simultaneously conduct calibration and uncertainty analysis for the Soil and Water Assessment Tool (SWAT). In this combined method, several SWAT models with different structures are first selected; next GA is used to calibrate each model using observed streamflow data; finally, BMA is applied to combine the ensemble predictions and provide uncertainty interval estimation. This method was tested in two contrasting basins, the Little River Experimental Basin in Georgia, USA, and the Yellow River Headwater Basin in China. The results obtained in the two case studies show that this combined method can provide deterministic predictions better than or comparable to the best calibrated model using GA. The 66.7% and 90% uncertainty intervals estimated by this method were analyzed. The differences between the percentage of coverage of observations and the corresponding expected coverage percentage are within 10% for both calibration and validation periods in these two test basins. This combined methodology provides a practical and flexible tool to attain reliable deterministic simulation and uncertainty analysis of SWAT.