Water quality assessment for River Mahanadi of Odisha, India using statistical techniques and Artificial Neural Networks

Abstract

The present study is conducted on Mahanadi River, which is considered as the lifeline for the state of Odisha. There is a persistent outcry of the people in upstream reaches of the catchment, blaming the industries for pollution of water, air and adjoining land, whereas the industries have taken all steps to remain as zero effluent industries. Also due to presence of the Hirakud dam and its water reservoir, the major multipurpose water resources project, near Sambalpur of Odisha state in India, there is a boost in agricultural production. The farmers are using various fertilizers and pesticides to increase crop production, which has made the portion of Odisha, close to the Hirakud dam, a food bowl of Odisha state. The management, utilization and protection of the water resources is based on right assessment of pollutants released from both point and non-point sources. It helps in improving the health condition of the people, and irrigation water use by controlling the disposal of waste load. To address the issue scientifically, the demand for development of water quality models has increased manifold. The objective of the current study is to develop a model to assess the temporal and spatial variability in the quality of water of Mahanadi river basin lying in the state of Odisha and to evaluate with respect to municipal and agricultural use. Pollution of the river water is mainly due to releases of agro-chemicals and releases from the local industries. To take a stock of the situation, six different locations (gauging stations) are selected at different locations of the river and the water quality data such as pH, Temperature, PO4, SO4, TDS, EC, Nitrate, K, F, Cl, Na+, Ca2+, Mg2+, Turbidity, T-Hardness, T-Alkalinity, TC, BOD, DO and BOD are collected from the year 2011–2019 for the purpose of analysis and modeling of water quality. To interpret the complex water quality data matrices and to identify the most influential water quality parameters, three different methods viz., Principal Component Analysis (PCA), Factor Analysis (FA), and Cluster Analysis (CA) were applied. Water Quality Index (WQI) for different seasons were calculated to evaluate the quality of water. The parameters such as Sodium Absorption Ratio (SAR), Permeability Index of soil, Percent Sodium, Magnesium ratio, Kelley's ratio, Residual sodium carbonate and Exchangeable sodium percentage were used to evaluate the quality of water from view point of irrigation.The PCA and ANN are applied jointly to predict the water quality parameters. Principal Component Analysis (PCA) was used to interpret the complex water quality data matrices and to choose the parameters which have major influence on water quality. Artificial Neural Networks (ANNs) were used to develop a model and to simulate as well as to predict the water quality parameters. It is observed that ANN model predicts the water quality with an appreciable accuracy. Scatter plots between actual and predicted WQI for training and validation phases showed high degree of co-efficient of determination (R2) with the values 90.14% and 98.2%. Hence it is observed that Back-propagation Neural Networks (BPNN) is able to predict the water quality one month ahead, thereby giving sufficient time to take preventive measures to keep river water free from pollution.