Comparison of Performance between Genetic Algorithm and SCE-UA for Calibration of SCS-CN Surface Runoff Simulation

Abstract

Global optimization methods linked with simulation models are widely used for automated calibration and serve as useful tools for searching for cost-effective alternatives for environmental management. A genetic algorithm (GA) and shuffled complex evolution (SCE-UA) algorithm were linked with the Long-Term Hydrologic Impact Assessment (L-THIA) model, which employs the curve number (SCS-CN) method. The performance of the two optimization methods was compared by automatically calibrating L-THIA for monthly runoff from 10 watersheds in Indiana. The selected watershed areas ranged from 32.7 to 5844.1 km 2 . The SCS-CN values and total five-day rainfall for adjustment were optimized, and the objective function used was the Nash-Sutcliffe value (NS value). The GA method rapidly reached the optimal space until the 10th generating population (generation), and after the 10th generation solutions increased dispersion around the optimal space, called a cross hair pattern, because of mutation rate increase. The number of looping executions influenced the performance of model calibration for the SCE-UA and GA method. The GA method performed better for the case of fewer loop executions than the SCE-UA method. For most watersheds, calibration performance using GA was better than for SCE-UA until the 50th generation when the number of model loop executions was OPEN ACCESSWater 2014, 6 3434 around 5150 (one generation has 100 individuals). However, after the 50th generation of the GA method, the SCE-UA method performed better for calibrating monthly runoff compared to the GA method. Optimized SCS-CN values for primary land use types were nearly the same for the two methods, but those for minor land use types and total five-day rainfall for AMC adjustment were somewhat different because those parameters did not significantly influence calculation of the objective function. The GA method is recommended for cases when model simulation takes a long time and the model user does not have sufficient time for an optimization program to search for the best values of calibration parameters. For other cases, the SCE-UA program is recommended for automatic model calibration.

Relationship Between Calibration Time and Final Performance of Conceptual Rainfall-Runoff Models

“…From Arsenault et al (2014) one may learn that some algorithms initially converge quicker than the others, but such methods often perform worse than the best ones at the end of the search. A similar result was obtained by Jeon et al (2014), but for another goal, namely calibration of long-term hydrologic impact assessment model. On the contrary, although in Tolson and Shoemaker (2007) and Piotrowski et al (2017a) also differences in convergence speed between various algorithms applied for calibration of conceptual rainfall-runoff models are noted, finally vast majority of methods reach almost equal performance.…”

“…Would such results be similar to those obtained when different maximum numbers of function calls are assumed? For example, when setting the maximum number of function calls to 20,000, Jeon et al (2014) found that the variant of Genetic Algorithm they apply perform better than SCE-UA method (Duan et al 1992) until 5150 function evaluations is used, but SCE-UA is better afterwards. Does it means that when one sets the maximum number of function calls to 5000, Genetic Algorithm would still perform better?…”

Automatic Calibration Tool for Hydrologic Simulation Program-FORTRAN Using a Shuffled Complex Evolution Algorithm

“…Iskra and Droste [14] found that the random multiple search method (RSM) and the Shuffled Complex Evolution method (SCE-UA) could find a parameter set providing better model performance statistics than with PEST employing the GLM algorithm. Sahoo et al [4] calibrated the hydrologic components of HSPF using a generic algorithm (GA), but it has been suggested that the GA required greater computing resources and time for parameter calibration than SCE-UA, making running the model less efficient [18][19][20][21][22].…”

Radiometric Cross-Calibration of GF-4 PMS Sensor Based on Assimilation of Landsat-8 OLI Images

“…The idea of complex shape segmentation and mixing introduced in this algorithm improves search efficiency, computation speed, and global search ability. SCE-UA is not sensitive to the initial values of the optimization parameters, thus avoiding excessive reliance on prior knowledge during optimization process and solves the problems caused by missing initial parameter values [35]. Research shows that SCE-UA can converge to the global optimal solution efficiently due to its fast convergence speed and good stability [36].…”