Multi-Objective Automatic Calibration of a Semi-Distributed Watershed Model using Pareto Ordering Optimization and Genetic Algorithm

This study explored the application of a multi-objective evolutionary algorithm (MOEA) and Pareto ordering in the multiple-objective automatic calibration of the Soil and Water Assessment Tool (SWAT). SWAT was calibrated in the Calapooia watershed, Oregon, USA, with two different pairs of objective functions in a cluster of 24 parallel computers. The non-dominated sorting genetic algorithm (NSGA-II), a fast MOEA, and SWAT were called from a parallel genetic algorithm library (PGAPACK) to determine the Pareto optimal set. One hundred fifty-five parameters were explicitly calibrated (9 for each 17 hydrologic response units [HRUs] and 2 for the whole watershed). With the root mean square error (RMSE) and mean absolute error (MAE) of the daily flows as objective functions, the Pareto front converged to a narrow range of solution set. A wider Pareto optimal front was formed when the RMSE of high and low flows were used as objective functions. The calibrated SWAT model simulated well the daily streamflow of the Calapooia River for a 3-year period. The daily NashSutcliffe efficiency was 0.85 at calibration and 0.80 at validation. Automatic multi-objective calibration of a complex process-based watershed model such as SWAT was successfully implemented using Pareto ordering optimization and an MOEA. Simultaneous automatic-calibration of flows and water quality parameters for the whole watershed and for different sub-basins, dynamic link with economic models, and integration of uncertainty and sensitivity methods are now explored.