Farzaneh Naseri; mahmood azari; Mohamad Taghi Dastoorani
Abstract
Introduction: Determining the amount of watershed sedimentation and its spatial distribution by using field measurements in practice faces a serious challenge. In recent decades, hydrological models have been widely used by hydrologists and water resource managers as a tool for analysing water resource ...
Read More
Introduction: Determining the amount of watershed sedimentation and its spatial distribution by using field measurements in practice faces a serious challenge. In recent decades, hydrological models have been widely used by hydrologists and water resource managers as a tool for analysing water resource management systems. The SWAT model is one of the semi-physical and semi-distributed hydrological models that have been widely used in recent years. Despite the wide use of the SWAT, simulation of the sediment has been associated with a large error in comparison to flow. These errors may come from using empirical methods such as the sediment rating curve for estimating sediment based on measured data. Therefore, in this research, the capabilities of the genetic algorithm (GA) were used to optimize the relationship between discharge and sediment and further optimal equation used for calibration and validation of the model.
Materials and Methods: The studied area is Fariman dam watershed with an area of 278.8 km2 which is located at latitude of 35 ˚ 33' to 35˚ 41' and longitude of 59 ˚ 34' to 59 ˚ 44' in Razavi Khorasan province. In this study, SWAT model was used to simulate runoff and sediment yield of Fariman dam watershed. In order to run the model, meteorological and hydrometric data including daily rainfall and maximum and minimum temperatures and sediment yield and discharge data, soil and land use maps of the watershed were achieved from relevant resources. The capabilities of the genetic algorithm were used to optimize the discharge -sediment relationship and were compared with sediment rating curve. For this purpose, optimization problem was defined for the genetic algorithm in MATLAB software as a search space of continuous values of the discharge –sediment coefficients. After that, sediment yield was extracted based on discharge data and calculated monthly sediment for SWAT calibration and validation. Sensitivity analysis, calibration and validation of the model were performed using the SUFI-2 algorithm using SWAT-CUP software. For this purpose using high sensitive parameters, the model was calibrated and validated for the period of 1991 to 2000.
Results and Discussion: Optimal coefficients extracted by GA indicate a better performance of the genetic algorithm in estimating the sediment yield. The comparative results of the sediment estimation models, revealed better performance of the genetic algorithm with RMSE = 70.9, NSE =0.46 and R2= 0.72 than the sediment rating curve. According to senetivity analysis of SWAT model, twelve parameters for stream flow and seven parameters for sediment yield were found to be sensitive. The most sensitive parameters for stream flow were SCS runoff curve number (CN2), effective hydraulic conductivity in tributary channel (CH_K1) and base flow alpha factor for bank storage (ALPHA_BNK) and the most sensitive parameters for sediment yield were peak rate adjustment factor for sediment routing, USLE equation soil erodibility factor (USLE_K), sediment concentration in lateral flow and groundwater flow (LAT_SED) and exponent parameter for calculating sediment reentrained in channel sediment routing (SPEXP). The SWAT calibration and validation results showed that the Nash-Sutcliffe efficiency index for monthly sediment and discharge for calibration period was 0.75 and 0.73, respectively and in the validation period was 0.85 and 0.76, respectively. Calibration and validation of the SWAT model was done with genetic algorithm model as an optimal method for deriving sediment data from measured daily discharge. The Nash-Sutcliffe efficiency coefficient for monthly discharge was 0.75 and 0.85 in the calibration and validation periods. Nash-Sutcliffe efficiency coefficients for monthly sediment yield were 0.86 and 0.81 for the same periods. SWAT evaluation results indicate that the model simulation is acceptable for predicting sediment yield and river flow. The performance of SWAT model in predicting of sediment in low flow is poor, which can be due to the effect of the parameters and model simplifications in the simulation of the sediment load.
Conclusions: In this research, simulation of runoff and sediment flow for Fariman dam watershed was performed using SWAT model. For this purpose, the capabilities of the genetic algorithm were used to optimize the relationship between discharge and sediment yields; then the results were used to calibrate and validate the SWAT model. The results indicate that genetics algorithm can be used for optimizing coefficient of sediment discharge equation and the result is better than sediment rating curve. Simulation of watershed hydrology using SWAT shows that the capability of the model in prediction of sediment yield and water flow is good. Using genetic algorithm to optimize the relationship between discharge and sediment has an important role in extracting daily sediment yield and simulation accuracy of the model. Also, the use of evolutionary algorithms can have a significant role in extracting the discharge -sediment relations, which usually is accompanied with a large error in experimental models such as a sediment rating curve.
K. Roshangar; M.T. Aalami; F. Vojoudi Mehrabani
Abstract
Introduction: Exact prediction of transported sediment rate by rivers in water resources projects is of utmost importance. Basically erosion and sediment transport process is one of the most complexes hydrodynamic. Although different studies have been developed on the application of intelligent models ...
Read More
Introduction: Exact prediction of transported sediment rate by rivers in water resources projects is of utmost importance. Basically erosion and sediment transport process is one of the most complexes hydrodynamic. Although different studies have been developed on the application of intelligent models based on neural, they are not widely used because of lacking explicitness and complexity governing on choosing and architecting of proper network. In this study, a Genetic expression programming model (as an important branches of evolutionary algorithems) for predicting of sediment load is selected and investigated as an intelligent approach along with other known classical and imperical methods such as Larsen´s equation, Engelund-Hansen´s equation and Bagnold´s equation.
Materials and Methods: In this study, in order to improve explicit prediction of sediment load of Gotoorchay, located in Aras catchment, Northwestern Iran latitude: 38°24´33.3˝ and longitude: 44°46´13.2˝), genetic programming (GP) and Genetic Algorithm (GA) were applied. Moreover, the semi-empirical models for predicting of total sediment load and rating curve have been used. Finally all the methods were compared and the best ones were introduced. Two statistical measures were used to compare the performance of the different models, namely root mean square error (RMSE) and determination coefficient (DC). RMSE and DC indicate the discrepancy between the observed and computed values.
Results and Discussions: The statistical characteristics results obtained from the analysis of genetic programming method for both selected model groups indicated that the model 4 including the only discharge of the river, relative to other studied models had the highest DC and the least RMSE in the testing stage (DC= 0.907, RMSE= 0.067). Although there were several parameters applied in other models, these models were complicated and had weak results of prediction. Our results showed that the model 9, with the most DC and the least RMSE (DC=0.694, RMSE= 0.081), had the relative advantage to the other none dimensional models. Finally it is clear that the model 6 had more predicting capability rather than the model 9, so among all the models, model 6 was the best referring model for estimation of sediment load of the Ghotoorchay river.
Conclusion: It was observed that the model including only the discharge of the Ghotoorchay river the best model for estimation of sediment load and it was applied for comparing all the other sediment predicting models such as some classic methods that includes Larsen´s equation, Engelund-Hansen´s equation and Bagnold´s equation and optimized rating curve. Among all methods, it was concluded that the genetic programming was superior to other methods in predicting sediment load of the mentioned river. Therefore, genetic programming that is a branch of evolutionary algorithms, with high prediction capability was offered as a powerful tool for optimizing and explicit predicting of total sediment load of the Ghotoorchay River.
