Artificial intelligence-based single and hybrid models for prediction of water quality in rivers: A review
Introduction
Modeling of the water quality-related variables in surface waters is now considered a crucial issue concerning efficient and reliable management of water resources which can result in a significant reduction in costs due to the reliability of these models and the use of indirect techniques which are capable of determining the values of water quality-related variables in the future. The limitations in the ability of traditional techniques such as multiple linear regression (MLR) [[1], [2], [3], [4]], multiple non-linear regression (MNLR) [1], Mann-Kendall (MK) trend test [5], and auto-regressive integrated moving average (ARIMA) [1,6] models to accurately estimate the water quality in rivers due to the complexity and sophistication of the water quality time series led to the use of the black box methods such as the artificial neural network (ANN) [[7], [8], [9], [10], [11]], neuro-fuzzy (NF) [3,12], genetic programming (GP) [13], and support vector machine (SVM) [14]. Besides these single methods, many hybrid models have been widely used for water quality modeling in rivers as well.
During the past decades, many researchers have used different models to improve the accuracy of water quality predictions. All these models could be classified under two main categories: the conventional models and artificial intelligence (AI)-based models. Conventional statistical models, including MLR and ARIMA models, are linear in predicting future values and are forced to be linear functions of past observations. Due to their relatively simple concepts and ease of implementation, linear models have been the focal point of many investigations and the critical tool for time series modeling over the past few decades [15]. However, in most real problems, especially in the field of river water quality modeling, we encounter non-linear patterns that need non-linear models to deal with that. Therefore, to overcome the limitation of such linear models, in the past two decades, several non-linear models have been proposed in the literature. These models include ANN, fuzzy, GP, and SVM. Many studies have proved the capability and usefulness of these methods for modeling different river water quality measurements as responses. Furthermore, in the past decade, combining these models and also the use of data pre-processing techniques, as well as many other hybrid models have been presented in the literature. The neuro-fuzzy (NF) model, wavelet-neural networks (WANN) model, wavelet-support vector regression (WSVR), and wavelet-linear genetic programming (WLGP) model can be mentioned as examples of hybrid models which have been used to model river water quality.
In the following, the two categories of modeling techniques, namely, the conventional and the AI-based models, are presented. Within each category, the commonly used methods for water quality modeling in rivers are briefly illustrated. Subsequently, to lend some insight into the application and reliability of the model in the field of water quality modeling in rivers, several studies published between 2000 and 2016 are provided. Although in the literature some reviews are presented regarding the modeling of quality of water bodies such as rivers, lakes, and reservoirs, most of them focused on modeling a specific water quality variable as an output (e.g., suspended sediment) or the application of a single method (ANN model for example). In Table 1, a summary of papers containing reviews of water quality modeling in riverine systems is presented. Some of these reviews discussed the application of AI techniques for both water quality (e.g., suspended sediment) and water quantity (e.g., rainfall-runoff) modeling. A number of these papers reviewed the studies concerning water quality in two or more bodies of water (e.g., river, lake, or groundwater).
The lack of a comprehensive review of river water quality modeling can be inferred from Table 1. A thorough review should involve different types of water quality variables and include several modeling approaches as well. The current review covers numerous studies that have been carried out to model several river water quality-related variables through the use of 13 different conventional and AI modeling approaches. This review is the first attempt that evaluates the current studies on the modeling of almost all types of water quality variables in rivers by using various modeling approaches. However, modeling of water quality in rivers via software such as GIS, Mike, Qual2k, WASP is not included in this review paper.
This review focuses primarily on the application of AI-based approaches to model river water quality. For this purpose, the relevant papers were identified by a keyword search of the papers conducted on 20th February 2016, within the period between 2000 and 2016 using the keywords; water quality, river, modeling, forecasting, and prediction, accompanied by the names of one or more of the modeling approaches such as; neural networks, fuzzy logic, genetic algorithm, wavelet neural network, to name but a few. Then, from the presented results by search engines, the papers in English which were relevant to our purpose were selected. Finally, looking at the inside, 51 of the most relevant papers were selected. The selected papers are published in 25 separate international scientific indexing (ISI) journals. The list of these journals and their 2014 impact factors is summarized in Appendix B.
The selected papers for this review comprised the main bulk of the published materials, including the most renowned studies, which had been cited a total of 1716 times up to 20th February 2016. Fig. 1 shows the number of cumulative citations per year. Four of the papers published before 2010 can be highlighted through their citation counts that exceed 100. These are Cigizoglu [19], Alp and Cigizoglu [20], Singh et al. [9], and Faruk [6] with citations of 177, 112, 210, and 128, respectively. The four highly cited papers included in the study were published before 2010, and each one utilized the ANN technique to model water quality.
The selected papers are analyzed based on their year of publication, the region of their study, their modeling approaches, and the water quality variables modeled. In Appendix C, for each of the selected papers, the results of the analysis mentioned above are presented. It should be noted that two or more techniques were used by the majority of these papers to model water quality in rivers. Therefore, in Appendix C, for each study, the applied models are ordered based on their modeling accuracies. Fig. 2 shows the number of papers published annually from 2000 to 2016. From this figure, it is evident that during the past decade, there has been a growing interest in modeling water quality of rivers using AI-based models. With some fluctuations and drops in the number of papers published in 2005, 2006, and 2012, the overall trend in the number of publications per year is incremental. Most of the papers were published in the past few years. For example, 12 papers (about one-fourth of the entire papers) were published in 2014. The drop in the number of papers published in 2015 is partly due to the fact that this review was carried out on 20th February 2016.
The selected papers were categorized based on modeling approaches with superior performances. Table 2 indicates the number of papers for each modeling procedure. As can be observed in this table, the single ANN model and its combination with other methods are the widely used models among the various AI-based methods used for water quality modeling in rivers. Pre-processing the inputs for the ANN models using wavelet analysis is a common technique to improve the predictive ability of the ANN models. For each modeling approach, the number of papers published from 2000 to 2016 is shown in Fig. 3.
Section snippets
Brief introduction and bibliographic review on conventional models
Probably the most conventional and frequently used methods for modeling river water quality are based on regression techniques. MLR, MNLR, and ARIMA models are widely used for this purpose. However, they assume linearity and stationarity in the data and have a limited ability to capture non-stationarities and non-linearities involved in the hydro-environmental data [21]. These models are briefly illustrated in the following subsections, and the studies in which these techniques were used are
Artificial neural networks
In recent years, ANN has received more attention in water resources studies. It is a robust computational algorithm that is used to simulate complex non-linear relationships, especially in situations where the explicit form of the relation between the variables is unknown. ANNs are mathematical systems and are able to map a set of input data into a corresponding set of output data [37]. In an ANN predictive model, the historical data time series are used to predict its future values. The
Water quality data
According to the source and nature, the river water quality variables could be categorized into three main groups; chemical, physical, and biological. In the literature for river systems, different water quality variables have been modeled. For example, algae, nutrients, or nutrient-related chemical indicators, such as DO and BOD, and physical water quality variables, such as T, EC, TDS, salinity, pH, and SSL, are often of concern. In this paper, the selected papers for review cover the three
Summary and conclusions
Attempts were made in this paper to review the papers which used AI-based techniques to model water quality in rivers. To this end, 51 journal papers that were published from 2000 to 2016 and focus on water quality modeling in rivers were selected. In this review, almost all different types of single and hybrid AI models were taken into account. Analysis of the selected papers showed that in recent years there had been an increasing trend towards using these models in the field of water quality
Recommendations for future works
Based on the review of 51 papers on river water quality modeling, conducted in this paper, the following recommendations for future research are made:
- 1
Because many studies have proved that the combined or hybrid models such as WANN are more accurate comparing the single structure AI-models, it is recommended that more attention be given to the improvement of such models.
- 2
Because of the limited studies in the application of GA-NN, WGP, and WSVR models for modeling water quality in rivers, more
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (105)
- et al.
Suspended sediment load prediction of river systems: an artificial neural network approach
Agric. Water Manag.
(2011) Wavelet and ANN combination model for prediction of daily suspended sediment load in rivers
Sci. Total Environ.
(2011)- et al.
Daily suspended sediment concentration simulation using ANN and neuro-fuzzy models
Sci. Total Environ.
(2009) - et al.
Comparison of Mann–Kendall and innovative trend method for water quality parameters of the Kizilirmak River, Turkey
J. Hydrol.
(2014) - et al.
Artificial neural network modeling of the river water quality—a case study
Ecol. Model.
(2009) - et al.
Protocol for developing ANN models and its application to the assessment of the quality of the ANN model development process in drinking water quality modelling
Environ. Model. Software
(2014) - et al.
Support vector machines in water quality management
Anal. Chim. Acta
(2011) Time series forecasting using a hybrid ARIMA and neural network model
Neurocomputing
(2003)- et al.
Neural networks for the prediction and forecasting of water resources variables: a review of modelling issues and applications
Environ. Model. Software
(2000) - et al.
Methods used for the development of neural networks for the prediction of water resource variables in river systems: current status and future directions
Environ. Model. Software
(2010)
Applications of hybrid wavelet–Artificial Intelligence models in hydrology: a review
J. Hydrol.
Estimation and forecasting of daily suspended sediment data by multi-layer perceptrons
Adv. Water Resour.
Suspended sediment load simulation by two artificial neural network methods using hydrometeorological data
Environ. Model. Software
A geomorphology-based ANFIS model for multi-station modeling of rainfall–runoff process
J. Hydrol.
Support vector machines and its applications in chemistry
Chemometr. Intell. Lab. Syst.
A modified Mann-Kendall trend test for autocorrelated data
J. Hydrol.
Development of a coupled wavelet transform and neural network method for flow forecasting of non-perennial rivers in semi-arid watersheds
J. Hydrol.
Modeling biological oxygen demand of the Melen River in Turkey using an artificial neural network technique
J. Environ. Manag.
Comparing various artificial neural network types for water temperature prediction in rivers
J. Hydrol.
Fuzzy logic modeling of the dissolved oxygen fluctuations in Golden Horn
Ecol. Model.
Fuzzy sets
Inf. Contr.
A Cascaded Fuzzy Inference System for Indian river water quality prediction
Adv. Eng. Software
A genetic programming approach to suspended sediment modelling
J. Hydrol.
Modeling discharge-suspended sediment relationship using least square support vector machine
J. Hydrol.
Forecasting of chlorophyll-a concentrations in South San Francisco Bay using five different models
Appl. Ocean Res.
Recent advances in wavelet analyses: Part 1. A review of concepts
J. Hydrol.
Recent advances in wavelet analyses: Part 2—amazon, Parana, Orinoco and Congo discharges time scale variability
J. Hydrol.
Joint multiple adaptive wavelet regression ensembles
Chemometr. Intell. Lab. Syst.
Information Quality Ratio as a novel metric for mother wavelet selection
Chemometr. Intell. Lab. Syst.
Estimation and forecasting of daily suspended sediment data using wavelet–neural networks
J. Hydrol.
A hybrid WA–CPSO-LSSVR model for dissolved oxygen content prediction in crab culture
Eng. Appl. Artif. Intell.
A wavelet–linear genetic programming model for sodium (Na+) concentration forecasting in rivers
J. Hydrol.
Modelling of chemical oxygen demand by using ANNs, ANFIS and k-means clustering techniques
J. Hydrol.
Adaptive neuro-fuzzy computing technique for suspended sediment estimation
Adv. Eng. Software
Assessment of arsenic concentration in stream water using neuro fuzzy networks with factor analysis
Sci. Total Environ.
Evolving neural network using real coded genetic algorithm for daily rainfall–runoff forecasting
Expert Syst. Appl.
A hybrid evolutionary data driven model for river water quality early warning
J. Environ. Manag.
Back-propagation neural networks for modeling complex systems
Artif. Intell. Eng.
Integrating data transformation techniques with Hopfield neural networks for solving travelling salesman problem
Expert Syst. Appl.
Linear and nonlinear modeling for simultaneous prediction of dissolved oxygen and biochemical oxygen demand of the surface water—a case study
Chemometr. Intell. Lab. Syst.
TOMCAT: a MATLAB toolbox for multivariate calibration techniques
Chemometr. Intell. Lab. Syst.
An ANN application for water quality forecasting
Mar. Pollut. Bull.
River suspended sediment load prediction: application of ANN and wavelet conjunction model
J. Hydrol. Eng.
A hybrid neural network and ARIMA model for water quality time series prediction
Eng. Appl. Artif. Intell.
Cascade correlation artificial neural networks for estimating missing monthly values of water quality parameters in rivers
Water Resour. Manag.
Application of artificial neural networks for water quality prediction
Neural Comput. Appl.
Modeling daily chlorophyll a dynamics in a German lowland river using artificial neural networks and multiple linear regression approaches
Limnology
Wavelet and neuro-fuzzy conjunction approach for suspended sediment prediction
Clean–Soil, Air, Water
Modeling of water quality parameters using data-driven models
J. Environ. Eng.
Comparison of multivariate regression and artificial neural networks for peak urban water-demand forecasting: evaluation of different ANN learning algorithms
J. Hydrol. Eng.
Cited by (136)
High-frequency data significantly enhances the prediction ability of point and interval estimation
2024, Science of the Total EnvironmentDeveloping a novel tool for assessing the groundwater incorporating water quality index and machine learning approach
2023, Groundwater for Sustainable Development