Document Type : Research Article
Authors
1
Professor in Geomorphology, University of Tabriz , Tabriz ,Iran.
2
Professor in Geomorphology, University of Tabriz, Tabriz, Iran.
3
PhD in Geomorphology, University of Tabriz, Tabriz, Iran.
Abstract
One of the most important solutions for the principled and correct management of water and soil in basins is to use hydrological models for simulation of rainfall-runoff processes, which estimate and predict the components of the water balance, including runoff, evaporation, transpiration, and infiltration, especially in basins without data or with incomplete data. In this study, the water balance of Ojan Chai basin was simulated by the SWAT semi-distributed and SMAR models for 20 years (2002-2021). The SUFI-2 algorithm was used in the SWAT model and the genetic algorithm optimizer was used in the SMAR model for calibration (2005 to 2016) and validation (2017 to 2021).
The simulation results in the calibration phase with the objective functions of the Nash-Sutcliffe efficiency (NSE), coefficient of determination (R²), and RSR were equal to 0.80, 0.81, and 0.45, respectively, and for validation, 0.74, 0.75, and 0.51, respectively, which indicates the high efficiency of the model in simulating the water balance. For the SMAR method, the Nash-Sutcliffe coefficients (NSE) and R² were obtained as 0.60 and 0.56 in calibration and 0.625 and 0.67 in validation.
Both of these models perform better in simulating the baseflow and average flow rates than the maximum flow rates; however, the matching of the simulated flow rate with the observed flow rate in the SWAT model is superior to that of the SMAR model, which is due to the consideration of spatial changes in this type of model. Considering the very good performance results of the models, it is suggested that they be used in watershed protection measures and water and soil management.
Extended Abstract
Introduction
One of the most fundamental natural processes of basins is the hydrological process. Estimating and predicting hydrological balance components, including runoff, evapotranspiration, and infiltration, are important in basins. Protecting water and soil resources is an important strategy for managing basins. Hydrological models are one strategy used to simulate rainfall-runoff processes. Rainfall-runoff models can interpolate and extrapolate flow according to the input data to the model. Hydrological models can be classified into lumped, semi-distributed, and distributed types because of differences in basin simulation structure. Among the lumped models, we can mention the RRL software model series, which includes AWBM, SIMHYD, TANK, SMAR, and SACRAMENTO, and the SWAT model is the semi-distributed model with the highest usage. The Ojan Chay basin, located in Bostanabad County (East Azerbaijan Province), is one of the main sub-basins of the Aji Chay River. Due to its topographic features, semi-arid climate, and soil type, it is one of the important pastures of the Sahand Mountains and the location of various rainfed and irrigated crops. The purpose of this study is to evaluate and simulate rainfall-runoff over 20 years by using the lumped SMAR model and the semi-distributed SWAT model.
Material and Methods
The data required for this study includes topographic maps, soil maps, 30-meter DEM images, and land use maps. Rainfall data was extracted from Bostanabad, Bashsiz-Ojan, Saeedabad, Zarnak, Saray, Ghoshchi, Zinjanab, and Hashtrood weather stations. Synoptic stations in Bostanabad, Ahar, Sarab, Sahand, Mianeh, and Maragheh were used to extract minimum temperature, maximum temperature, wind speed, relative humidity, and evapotranspiration. Calibration and validation of the models were achieved by using discharge data from Bostanabad station during the period 2002-2021. SWAT and SMAR models were utilized in this study. In the SWAT model, the basin was first divided into sub-basins and hydrological response units (HRUs) by using DEM image, land use, slope, and soil maps, and data from weather stations in ArcMap environment. The SWAT-CUP software was utilized to simulate rainfall-runoff by extracting the model output. The SMAR model requires less input data than SWAT. In this model, monthly rainfall and evapotranspiration data from synoptic stations, as well as monthly discharge data from the Bostanabad hydrometric station, were entered. Genetic algorithm and related parameters were used to simulate rainfall-runoff. To evaluate the efficiency and accuracy of these models in the calibration and validation, the Nash-Sutcliffe efficiency (NSE), coefficient of determination (R²), and ratio of standard deviation of observations to root mean square error (RSR) were used.
Results and Discussion
Land use, slope, and soil were used to extract 13 sub-basins and 157 hydrological response units in the SWAT model. The model output was entered into SWAT-CUP software. The warm-up period was defined as the first 3 years of the statistical period 2002-2021. To determine the sensitivity of the parameters that affect runoff, the SUFI-2 algorithm was used. Sensitivity analysis was carried out for 26 parameters in the statistical period from 2005 to 2016 through monthly discharge data of the Bostanabad hydrometric station. According to the results, out of the 26 selected parameters, only 12 parameters were sensitive and effective and were included in the model with the range of their changes. The model was calibrated by using the new optimal values of the parameters for the statistical period 2005 to 2016 at monthly time scale to obtain the best values of the objective functions. By applying the optimized parameters in the calibration, Nash-Sutcliffe efficiency (NSE), coefficient of determination (R²), and ratio of standard deviation of observations to root mean square error (RSR) were obtained as 0.80, 0.81, and 0.45, respectively. The results in the calibration were in a very good range. For the validation, monthly discharge data during 2017 to 2021 were entered into the model. The Nash-Sutcliffe efficiency (NSE), coefficient of determination (R²), and RSR were 0.74, 0.75, and 0.51, respectively, which indicate the ability of the SWAT hydrological model to simulate the discharge of the Ojan Chay basin. In the SMAR model, monthly rainfall and evapotranspiration data from synoptic stations and monthly discharge data from Bostanabad hydrometric station were entered into the model according to the software format. Genetic algorithm and parameters related to water balance and flow routing were used to simulate rainfall-runoff. The model was calibrated in the statistical period 2005-2016 and validated in the statistical period 2017-2021, and the Nash-Sutcliffe efficiency (NSE) and coefficient of determination (R²) were considered to evaluate the performance of the SMAR model. The Nash-Sutcliffe efficiency (NSE) and coefficient of determination (R²) in calibration (0.60 and 0.56) and validation (0.625 and 0.67) indicate the acceptable performance of the SMAR model. Like the SWAT model, the SMAR model has simulated the base and average discharge values well, but it has not simulated the maximum discharges well.
Conclusion
In this study, the physical and semi-distributed SWAT model and the lumped SMAR model were used to simulate the runoff of the Ojan Chay drainage basin over a 20-year period. The performance of the hydrological models was evaluated by Nash-Sutcliffe efficiency (NSE), coefficient of determination (R²), and ratio of standard deviation of observations to root mean square error (RSR) during the calibration and validation periods. The results showed that both models performed well in average and base discharges. One of the weaknesses of both models was the inappropriate estimation of peak discharges and extreme events. The lack of required data (rainfall, minimum and maximum temperature, relative humidity, wind speed, etc.), the lack of proper data recording by the relevant organization, the need for a large amount of data in SWAT, etc. are all the reasons that reduce the accuracy of these models. Generally, it can be said that although both models have good evaluation in runoff simulation during the calibration and validation periods, the performance of the SWAT model shows its superiority over the SMAR model, due to the high values of the Nash-Sutcliffe efficiency and R² in both calibration and validation.
Main Subjects
)
Send comment about this article