Document Type : Research Paper

Authors

• Alireza Ghaemi ¹
• Mahdi Azhdary Moghaddam ²
• Sarina Keikha ³

¹ Ph.D Student, Department of Civil Engineering, University of Sistan and Baluchestan, Zahedan, Iran

² Professor, Department of Civil Engineering, University of Sistan and Baluchestan, Zahedan, Iran

³ Msc Student, Department of Civil Engineering, University of Sistan and Baluchestan, Zahedan, Iran

Abstract

Introduction
Rivers are known as the vital resources of nature and the main foundations of sustainable development. Therefore, the quantity and quality of river water are considered valuable parameters. The increase in agricultural and industrial activities has reduced the quality of water resources in many places. The discharge of sewage, garbage and chemical fertilizers in the villages along the rivers is one of the most important sources of water quality pollution. The amount of urban and agricultural wastewater entering this surface has caused an increase in the amount of pollutants, so that in the period of 1993 to the end of 2017, the average amounts the three pollutants of total dissolved solid, chlorine and sodium in Varand Station are respectively 507.49, 2.16 and 2.47. Therefore, accurate estimation of water quality parameters is a basic requirement for water quality management, human health, public consumption and domestic use.
 
Materials and methods
Tajan River basin with an area of about 4147.22 square kilometers has an average river discharge and annual rainfall of 20 cubic meters per second and 539 mm respectively. The highest and lowest elevations of this River basin have been reported as 3728 and 26 meters, respectively. Various human activities such as agriculture and dam construction are carried out in this river. Therefore, evaluationg the water quality of this river basin is required. In this research, the combination of two Gene Expression Programming Models (GEP) and Artificial Neural Network (ANN) with a data preprocessing algorithm called Empirical Mode Decomposition (EMD) was used to estimate one of the important parameters of water quality called Total Dissolved Solids (TDS). For this purpose, in this research, some of qualitative parameters including sodium, calcium, magnesium, sulfate bicarbonate, sulfuric acid and chlorine, which were measured in the period of 1993 to the end of 2017 at Varand station, were used to estimate the concentration of total dissolved solids.
 
Results and discussion
At first, the results of the observation data during the sampling period indicated that the TDS values in about 80% of the samples were in the range of 300 to 600 mg.liter^-1, which reprsented the good quality of the water of this river. In order to compare the performance of independent and integrated approaches in estimating the quality parameters of the Tajan River in the training and testing stages, the evaluation benchmarks including Correlation Coefficient (R), Root Mean Square Error (RMSE), Mean Deviation of Error (MBE), Nash Coefficient (NSE), Objective Function (OBJ) and RSD ratio were applied. The results of this study demonestrated that the integrated model of Gene Expression Programming and Empirical Mode Decomposition (EMD-GEP) with the lowest error (RSD=0.23 and RMSE=24.41) was the most accurate model in TDS estimating compared to other models such as GEP (RSD=0.44 and RMSE=47.27). In addition, the integrated model of Artificial Neural Network and Empirical Mode Decomposition (EMD-ANN) with RMSE=36.64 and R=0.95 was stood at the second rank. Additionally, the outcomes of the Objective Function (OBJ) represented that EMD-GEP model could achieved the lowest OBJ value (15.92) than other techniques in the TDS modeling. While, the highest value of the OBJ=29.34 belonged to the GEP model.
 
Conclusion
ANN and GEP methods were applied in this research to estimate TDS concentarion in the Tajan River. After that, to increase the accuracy of the models, EMD technique was recruited to decompose the time series dataset. The results obtained from the integrated models were evaluated using some error statistical benchmarks such as correlation coefficient, root mean square error. The results showed that the EMD method could play an essential role in increasing the ANN and GEP performance so as to estimate this water quality parameter in Varand station. So that EMD-GEP and EMD-ANN could reduce the RMSE error by 48.35% and 14.02%, respectively, compared to the two independent models of GEP and ANN.

Keywords

• Artificial neural network
• Data pre-processing algorithms
• Gene expression programming
• Tajan river