Water quality modeling using artificial neural networks incorporating land use and sewage treatment plant factors
Loading...
Date
2022-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Teknologi Malaysia
Abstract
Skudai River has undergone a general decline in water quality in recent years due to agricultural practices, urbanisation, industrial and other human activities in the river catchment. It is classified as “slightly contaminated” by the Department of Environment (DOE), and as such, immediate actions are needed to prevent further deterioration and improve the water quality. Majority of existing research on water quality modelling focuses on water quality data and the impact of land use on water quality, while those on the effects of sewage treatment plants (STP) discharge on river water quality have also been conducted to a certain extent. However, limited research on water quality prediction is based on land use input, existing STP, and rainfall. This is due to the complicated relationships between these three factors and water quality parameters. River systems are highly complex, hierarchical and patchy. Accurate predictions of the time series concerning the changing water quality could support early warnings on water pollution and help with management decisions. Currently, artificial intelligence (AI) technologies can simulate this behaviour and complement the inherent deficiencies. Among the latest research in integrating AI into water quality modelling, artificial neural networks (ANNs) are the most popular techniques used. This study aimed to identify and determine key water quality parameters using principal component analysis (PCA) based on land use and pollution sources, to correlate and predict water quality index (WQI) based on in-situ parameters using ANNs, and to determine and predict the relationships between land use patterns, precipitation, STP, and WQI, also using ANNs. ANNs were employed in a total of 839 physical and chemical pollution data sets from the Skudai River from 2001 to 2019 as training (70%), test (15%) and validation data (15%) for the analysis in this study. River water sampling was also carried out to evaluate the modelling results (36 data sets). ArcMap 10.4 was used to prepare the map for the changes occurring in land use, observed from 2000 to 2019 The PCAs results indicated that the parameters causing water quality variations were mainly related to physical parameters (natural) and organic pollutants (anthropogenic). The study also showed that the cascade-forward net was the optimal ANNs-water quality index-1 (ANNWQI-1) model for WQI prediction with seven parameters: DO, pH, conductivity, temperature, TDS, salinity, and turbidity with an RSME of 7.15, and a coefficient of correlation (R) of 0.92. The analysis with Spearman correlation could explain that in-situ parameters correlated with the parameters used to calculate WQI values. The best ANNWQI-2 model was a feed-forward net with land use, STP service coverage, and precipitatin data as input data, resulting in RMSE of 6.98 and R of 0.80. An input data analysis with Spearman correlation could explain that land use data, STP and rainfall data correlated with the parameters used to calculate WQI values. The integrated model of ANNWQI-3 had RMSE and R of 6.01 and 0.92, respectively. ANNWQI-1 demonstrated that accurate WQI predictions could be made, with only seven in-situ water quality parameters, while ANNWQI-3 required more comprehensive input data to get almost the same R. More importantly, the input data was in-situ water quality parameters, and no laboratory analysis was needed. The study determined the effective input parameters using PCA for successful ANN modelling while illustrating the usefulness of ANNs for WQI prediction. Ultimately, the results will give decision-makers valuable information to identify the causes of water pollution and the critical source areas that are useful for protecting the environment in terms of sustainable water resources.
Description
Thesis (Ph.D(Civil Engineering))
Keywords
Neural networks (Computer science), Land use