Research paperEstimation of re-aeration coefficient using MLR for modelling water quality of rivers in urban environment
Graphical abstract
Introduction
Urban hydrology becomes the complex process with the significant variation in flow and deterioration in water quality due to increasing anthropogenic activities (Schirmer et al., 2013). Abrupt variation occurs in flow with the release of domestic, agricultural and industrial wastewater through drains/tributaries during the low flow period (Parmar and Keshari, 2012) and abstraction of freshwater for the water supply to the adjoining areas. Low flow in rivers and discharge of untreated wastewater have been the major reasons behind the vanishing of the river's ecosystem (Parmar, 2008, Kisi et al., 2013). With the discharge of untreated or partially treated wastewater from different sources that contains higher concentration of chemical and biological impurities also hampers the water quality at spatial and temporal scales (CPCB, 2006). Majorly biological waste is from urban distributed sources which joins the river and undergoes biological and chemical changes using dissolved oxygen. Concentration of dissolved oxygen indicates the health of the ecosystem (Hanbay et al., 2008) and biochemical oxygen demand (BOD) indicates the concentration of organic matter present in water (Dogan et al., 2009). However, due to spatial and temporal variations in hydrochemical and biological properties, continuous and regular monitoring programmes are required to have reliable information about the water quality (Singh et al., 2004).
To maintain the ecology of river, re-aeration is the most important process. Re-aeration is defined as the rate of transfer of oxygen from the atmosphere to water at the air-water interface (Grace and Imberger, 2006). Re-aeration process aids to minimize the oxygen deficit and leads to several changes in the metabolism of stream (Izagirre et al., 2008; Aristegi et al., 2009). The process is described by the rate of constants K2 as first order processes that is proportional to DO deficit. Re-aeration process mainly depends upon the degree of turbulence that can be measured directly, however, it is highly variable in both the space and time. This makes it difficult to measure re-aeration from stream turbulence (Tsivoglou et al., 1969, Thackston and Krenkel, 1969, Thackston and Dawson, 2001). Another method of re-aeration measurement is through the identification of oxygen balance of river, which is an indirect method and poorly understood (St. John et al., 1984, Moog and Jirka, 1995; Haider et al., 2003). Using the chemical, biological and hydraulic properties of water, several methods have been developed to measure the re-aeration coefficient (K2) of river such as tracer gases (Tsivoglou, 1967, Rathbun et al., 1978), empirical equations using hydraulic parameters (Melching and Flores, 1999), oxygen balance and diurnal variation in oxygen concentration (Hornberger and Kelly, 1975, Mcbride, 2002). Benson et al., 2014 used gas tracer technique to measure the re-aeration rate and conducted experiments at different creeks in the Sierra Nevada of California. Results obtained from the experiments were also compared with the established empirical formulas which show poor correlation between the data for the prediction of re-aeration coefficient. Numerous studies have been performed using predictive equations to measure the re-aeration coefficient (Moog and Jirka, 1995, Jha et al., 2001, Jha et al., 2004, Jain and Jha, 2005, Ugbebor et al., 2012). Applicability of predictive equations has been evaluated by Moog and Jirka, 1995 using mean multiplicative error and observed high errors for equation for low slopes. Most of these equations are based on the stream hydraulic parameters, such as depth, velocity, discharge and slope (Cox, 2003b, Raymond et al., 2012). However, none of the previously developed equations able to predict the accurate re-aeration coefficient and indicates that some more additional parameters should be considered in order to improve the accuracy (Wilson and MacLeod, 1974, Moog and Jirka, 1995, Melching and Flores, 1999, Haider et al., 2013, Benson et al., 2014).
Deterministic models are more realistic for accurate prediction, but require large amount of input data to simulate the water quality (Vieira et al., 2013). However, tradition method of data analysis are rarely being used now a days due to the involvement of large number of parameters causing variation in the water quality (Xiang et al., 2006). With the use of artificial computing techniques, challenges of hydrodynamic and water quality models can be minimized, which has wide application in water resource related problems. Multivariate Linear Regression (MLR) is the most commonly used and accepted machine learning method. It has a wide application in water-related research to identify the direct relation between the predictor and response variable as it generates minimum data set of indicators (Doran and Parkin, 1996). Using MLR, all the variables can be incorporated in to single model for the estimation of degree of interaction between the variables. Mustapha and Aris (2012) used MLR model to predict the pollution load in the Jakara-Getsi river system. Chenini and Khemiri (2009) also used the MLR model to investigate the relation between parameters using the hydro-chemical data to quantify the composition of groundwater samples in Maknassy Basin, central Tunisia. Chen and Liu (2015) developed a model using MLR to simulate the water quality of reservoir. Arora and Keshari (2016) developed a re-aeration model using MLR and established predictive equation. Water quality index was designed by Nasir et al. (2011) to predict the water quality of Klang River, Malaysia using MLR model. Kumar and Padhy (2014), have compared the various multivariate tools and their application on classification, monitoring and assessment of surface water quality parameters. All these studies indicate that, MLR model can be used as a suitable tool for water quality modelling, yet the design of adequate formulation for assessment of the re-aeration coefficient remains a problem.
Therefore, present study was carried out with the objective to develop set of models using MLR to estimate the re-aeration coefficient (K2) that could be applicable to a wide range of study areas and identification of best fit model on the basis of their applicability and performance analysis using RMSE, R2 and R. The model obtained can be coupled with the water quality model to design the pollution control strategies for the urban water management.
Section snippets
Study area
Yamuna River is the longest tributary of Ganga River that joins the Ganga River in Allahabad, India after travelling 1376 km from its origin at Yamunotri, India. Yamuna River travels around 348 km before entering in to Delhi at Palla village. Total length of Yamuna in Delhi is 48 km and out of which first 22 km in between the Wazirabad Barrage and Okhla barrage that receives the wastewater from 16 drains contain untreated/partially treated wastewater (CPCB, 2006) as shown in Fig. 1. Till the
Input parameters
For the prediction of water quality and estimation of representable re-aeration coefficient for the study area and for the area of similar characteristic, selection of appropriate hydraulic and water quality parameters are of prominent importance. Identification of combination of parameters responsible for changes in K2 is a complex process. In order to identify the correlation between the parameters, data set of 240 samples have been used for the analysis. Hydraulic parameters, such as flow (m3
Conclusions
MLR models were developed to predict the re-aeration coefficient for the estimation of assimilative capacity of river. Performance of model were evaluated using RMSE, R and R2 to obtain the best fit model. The model equation generated, are based on the linear relation between the predictor and response variables. The correlation coefficient obtained from the best fit model was 0.894 and results shows that MLR could use as powerful tool for the modelling of re-aeration coefficient and could be
Acknowledgements
We thank Indian Institute of Technology Delhi, India and GD Goenka University, Haryana, India for providing the infrastructure.
Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
References (57)
A review of dissolved oxygen modelling techniques for lowland rivers
Sci. Total Environ.
(2003)- et al.
Modeling biological oxygen demand of the Melen River in Turkey using an artificial neural network technique
J. Environ. Manag.
(2009) - et al.
Prediction of wastewater treatment plant performance based on wavelet packet decomposition and neural networks
Expert Syst. Appl.
(2008) - et al.
Current research in urban hydrogeology – a review
Adv. Water Resour.
(2013) - et al.
Multivariate statistical techniques for the evaluation of spatial and temporal variations in water quality of Gomti River (India)-a case study
Water Res.
(2004) - et al.
A critical appraisal of empirical equations and models for the prediction of the coefficient of reaeration of deoxygenated water
Water Res.
(1974) - et al.
Comparison of several methods to calculate reaeration in streams, and their effects on estimation of metabolism
Hydrobiologia
(2009) Standard Methods for the Examination of Wastewater and Water
(1985)- et al.
Standard Method for Examination Water and Waste Water
(2005) - et al.
Evaluation and development of re-aeration equation using multivariate linear regression
J. Eng. Res.
(2016)