Document Type : Research Paper

Authors

• Ashkan Banikhedmat ¹
• hosein salehi ²
• saeed golian ³
• farshad koohian afzal ⁴
• nazanin ezati boorestan ⁵

¹ PhD Candidate of Water Resources Engineering and Management, Faculty of Civil Engineering, Shahrood University of Technology and Planning expert of Water and Wastewater Company of Gilan Province, Iran

² Master of Water Resources Engineering and Management, Faculty of Civil Engineering, Shahrood University of Technology

³ Associate Professor, Department of Surveying, Faculty of Civil Engineering, Shahrood University of Technology and Senior Researcher, Faculty of Geography, University of Maynooth, Ireland

⁴ Ph.D. in Hydrology and Water Resources, Water Research Institute, Tehran, Iran

⁵ PhD student, Geotechnical Engineering, Faculty of Civil Engineering, Mohaghegh Ardabili University, Iran

Abstract

Introduction
One of the methods for estimating the amount of runoff resulting from precipitation is the use of hydrological models. The SWAT model is one of the widely used tools for simulating the quantity and quality of water at the watershed level. This model is a conceptual model that is capable of simulating large watersheds with different management scenarios. One of the major challenges of this model and many other hydrological models is the calibration of effective and sensitive parameters for estimating the amount of runoff. In general, calibration methods can be divided into two groups: manual and automatic. Manual calibration of a model requires the modeler to have a good understanding of the model's physics. On the other hand, due to the time-consuming nature, existing complexities and the development of new optimization algorithms, nowadays automatic calibration has gained more attention. Automatic calibration is based on three components: the objective function, the optimization algorithm, and the station information. The use of a single objective function in model calibration may lead to an increase in error in other aspects of the simulation. Scientific experience in single-objective calibration has shown that no single objective function, even with high efficiency, can accurately represent all the characteristics and properties of a watershed. Therefore, the use of an appropriate optimization algorithm to improve calibration results includes the use of multiple objective functions to identify a set of efficient solutions.
Materials and methods
The study area is located in the western part of Iran, in Kermanshah Province, with an area of 5467 square kilometers. The minimum and maximum elevations in the area are 1275 and 3360 meters, respectively. The average precipitation in the watershed is about 505 mm, with the highest rainfall occurring in the months of November and Decemeber, and the lowest rainfall in the months of Julay and August. The main rivers in this watershed are Mark, Gharehsoo, and Razavar. In this study, the SWAT rainfall-runoff model was calibrated using the NSGA-II algorithm under three calibration scenarios. For model calibration, the first scenario used the NSE objective function, which focuses on maximum flows. In the second scenario, to focus on minimum flows, the logarithmic transformation of the simulated and observed streamflow series was used, and the NSE efficiency coefficient was adopted as the objective function, represented as LogNSE. The third scenario was a combination of the first and second scenarios, where the non-concordant objective functions NSE and LogNSE were used simultaneously.
Results and discussion
The results of this study showed that based on the NSE evaluation index values (0.83, 0.74 and 0.83 for the first to third scenarios) and the model overestimation and examination of the flow graph in the first scenario, which showed a tendency towards higher flows, this scenario would be more efficient in estimating maximum flows. Additionally, considering the LogNSE evaluation index (0.69, 0.74 and 0.72 for the first to third scenarios), the second scenario with the LogNSE single objective performed better in minimum flows. However, the model constructed using two non-concordant objective functions aimed to achieve a balance and showed satisfactory performance in simultaneously estimating maximum and minimum flows.
Conclusion
In general, it can be concluded that if the objective of the study is to investigate maximum and minimum flows, such as flood or drought studies, single-objective algorithms will perform better. However, if the objective is to control the water balance and achieve satisfactory performance of a model in both maximum and minimum flows, a two-objective scenario with a non-concordant approach can yield better results compared to single-objective algorithms.

Keywords

• Genetic Algorithm
• Simulation of Rainfall-Runoff
• NSGA-II. Validation