Predictive Analysis of Lake Water Quality Using an Evolutionary Algorithm

Abstract

Lakes are water bodies having considerable availability of water. Hence, it could be convincing and promising water resource in the area of acute shortage of water. Throughout the world, the water quality of lakes, natural or human-made, has been deteriorating because of urban, agricultural, industrial and other impacts. Widespread eutrophication of lakes leads to the overgrowth of plants and algae; the bacterial degradation of their biomass consumes more oxygen from water resulting in the state of hypoxia [Azhagesan in Water quality parameters and water quality standards for different uses. National Water Academy Report, 1]. Water quality monitoring of reservoirs is essential in the exploitation of aquatic resources and its conservation. Continuous monitoring of water quality of lakes and prediction of water quality will help to conserve the lakes. The authors offer an application of an evolutionary algorithm for predictive analysis of water quality of reservoirs/lakes and to discover a functional relationship between features in data (symbolic regression). We have used a nature-inspired technique of genetic programming (GP) as it can evolve the best individual (program). GP has also been used to discover a functional relationship between features in data (symbolic regression) and to group data into categories (classification). We have used monthly water quality data observed by Maharashtra Water Resources Department, Hydrological Data Users Group (HDUG) for the present study. Case study of Gangapur reservoir (an artificial lake) located in Nashik district in the state of Maharashtra, India, is used for demonstration of strengths of the evolutionary algorithm. We have developed cause-effect models for biochemical oxygen demand (BOD), chemical oxygen demand (COD) and faecal coliform bacteria (F-col). In situations of a scarcity of observed data, hybrid cause-effect models will be useful. We have used spectral analysis to eliminate the problem of time lag in case of hybrid models. Cross-validation is executed to ensure the overfitting of the models. We have performed performance evaluation by classification accuracy, mean absolute error and mean squared error. The evolved hybrid cause-effect model with spectral analysis is of immense use for small data sets.