Application of the kohonen neural network in coastal water management: methodological development for the assessment and prediction of water quality
Introduction
The quality of the coastal environment is very important for the sustenance of both natural ecosystems and human activities (Karydis and Coccossis, 1990). Since the 1970s, the relationship between environmental quality and human activities has attracted the interest of both scientists and policy makers (OECD, 1978). In the Mediterranean coastal areas, the massive nutrient inflow to the sea, as a result of urbanization, has not only had a negative impact on coastal ecosystems, but also on tourism and other economic activities (Clark, 1996; Moriki and Karydis, 1994). In this context, coastal water quality managers have a keen interest in the development and application of analytical and operational tools which can facilitate the decision-making process. Statistical evaluations (Ignatiades et al., 1985; Austen and Warwick, 1989) based on univariate methods have low interpreting value for decision makers. Multivariate procedures have proved more suitable, given the complicated nature of pollution-induced ecological disturbances (Warwick, 1988), and they have been successfully applied in environmental quality assessment and management (Ignatiades et al., 1992; Karydis, 1996). Recently, artificial neural networks (ANN) have become the focus of interest across many scientific disciplines including ecology (Walley and Fontama, 1998; Foody, 1999; Brosse et al., 1999; Barciela et al., 1999; Lek and Guegan, 2000), analytical chemistry (McAlernon et al., 1999; Li et al., 1999), food analysis (Dolmatova et al., 1998) and water quality (Trautmann and Denceux, 1994; Mappa et al., 1996; Schleiter et al., 1999). Until now, the range of applications of ANN developed over recent years has not included its application in coastal water quality management.
ANN are computer techniques that attempt to simulate the functionality and decision-making processes of the human brain (Eberhart and Dobbins, 1990). They comprise a set of simple, interconnected units which work in parallel to categorize input data into output classes. Such networks may classify data more accurately than conventional statistical classifiers (Dolmatova et al., 1997; Kocjancic and Zupan, 1997). Different types of networks have been developed (Beale and Jackson, 1990; Haykin, 1994; Bishop, 1995). They all have in common the ability to learn from data. They may be broadly classified into networks which learn in a supervised or unsupervised way.
In supervised learning the network is given many different examples of a particular problem, including the desired solutions. Thus, the network is able to compare its outputs with the desired outputs to determine the extent of its errors. A training algorithm then modifies the internal parameters of the network so that the next time the same examples are presented to the network its predictions would be less erroneous. This process is repeated many times until the performance of the network is maximized. In unsupervised learning, the desired outputs are not given and the network learns to classify the examples by recognizing different patterns. Kohonen self organizing maps are a type of neural network designed for unsupervised pattern recognition tasks (Kohonen (1982), Kohonen (1989)). The goal of Kohonen neural network (KNN) is to map the spatial relationship between clusters of data points in hyperdimensional space. Once trained, they may be used to identify unknown data patterns.
In this paper, KNN has been applied to nutrient data obtained from the coastal waters of Almerı́a (Spain). The objective of this study is the development of a methodology, based on the output from the KNN, that makes it possible to assess and predict coastal water quality. To be of use in the management of coastal water quality, such a methodology had to be able to distinguish the mesotrophic conditions. KNN is evaluated as a possible tool in the decision-making process in coastal water quality management.
Section snippets
Sampling design
The coast of Almerı́a is one of the most important tourist areas in southern Spain. As in other Mediterranean countries, sewage discharges are the main contributory factors to nutrient enrichment of coastal waters (Aguilera, 1997). Taking this fundamental characteristic into account, sampling stations were established along the coast (Fig. 1): nine sampling stations (AD1, GV1, UR1, RM1, AG1, CC1, CB1, MJ1 and GR1) were located at the main sewage outfalls; nine sampling stations (AD2, GV2, UR2,
Interpretation of the activation maps
All data obtained for each sampling point were used on the Kohonen network. The mean nutrient concentrations of raw data are given in Table 1. Fig. 3 shows, for every sampling station, an activation map of 6×6 neurons. The convergence was mostly reached in 50,000 iterations. The activation maps obtained represent coastal water quality. Spatial placement of the activated neurons is indicative of different water qualities. For example, the activation maps for the two stations SB and CG indicate
Methodology of classification
Variables that enable the information in the activation maps to be extracted need to be based on the system of quadrats described in the methods section. Accordingly, the variables selected were: (1) quadrat containing the greatest activation, (2) quadrat containing the neuron with the highest degree of activation, (3) quadrat containing the centroid of the activation map. Table 2 shows the values obtained for these variables. Each variable comprises two values and each observation is defined
Conclusions
KNN can be an effective tool for the evaluation and prediction of the trophic status of coastal waters.
The information contained in the activation maps was successfully extracted using the quadrat system. This system preserves the salient features of the topographic map (Kohonen, 1990). The information drawn from the activation maps enables a classification of the trophic status of the coastal waters into four classes (potentially eutrophic, highly mesotrophic, slightly mesotrophic and
References (36)
- et al.
Comparison of univariate and multivariate aspects of estuarine meiobenthic community structure
Estuarine Coastal Shelf Sci.
(1989) - et al.
Modelling primary production in a coastal embayment affected by upwelling using dynamic ecosystems models and artificial neural networks
Ecol. Modelling
(1999) - et al.
The use of artificial neural networks to assess fish abundance and spatial occupancy in the littoral zone of a mesotrophic lake
Ecol. Modelling
(1999) - et al.
Quantitative analysis of paper coatings using artificial neural networks
Chem. Intell. Laboratory Syst.
(1997) Applications of the self-organising feature map neural network in community data analysis
Ecol. Modelling
(1999)- et al.
A possible method for evaluating oligotrophy and eutrophication based on nutrient concentration scales
Mar. Pollut. Bull.
(1992) - et al.
A new method based on counter propagation network algorithm for chemical pattern recognition
Anal. Chim. Acta
(1999) - et al.
Using artificial neural network for solving chemical problems. II. Kohonen self-organizaing feature maps and Hopfield networks
Chem. Intell. Laboratory Sys.
(1994) - et al.
Modelling water quality, bioindication and population dynamics in lotic ecosystems using neural networks
Ecol. Modelling
(1999) - et al.
Neural network predictors of average score per taxon and number of families at unpolluted river sites in Great Britain
Water Res.
(1998)
The level of taxonomic discrimination required to detect pollution effects on marine benthic communities
Mar. Poll. Bull.
Quality assurance of pre-determination steps for dissolved nutrients in marine samples
Neural ComputingAn Introduction
Neural Networks for Pattern Recognition
Coastal Zone Management Handbook
Identification of modified starches using infrared spectroscopy and artificial neural network processing
Appl. Spectrosc.
Neural Networks PC ToolsA Practical Guide
Cited by (97)
Retention analysis of droplets over compliant substrates
2023, Colloids and Surfaces A: Physicochemical and Engineering AspectsSelf-organizing map improves understanding on the hydrochemical processes in aquifer systems
2022, Science of the Total EnvironmentCharacterizing land use effect on shallow groundwater contamination by using self-organizing map and buffer zone
2021, Science of the Total EnvironmentReal-time probabilistic forecasting of river water quality under data missing situation: Deep learning plus post-processing techniques
2020, Journal of HydrologyCitation Excerpt :Thirdly, deep learning is commonly used to discover intricate structures in relation to large data sets (Schmidhuber 2015). In the past decades, ANNs were successfully utilized for water quality and environmental prediction, classification and pattern recognition (Aguilera et al., 2001; Peleato et al., 2018). For instance, the Random Forest (RF), the Quantile Regression Neural Networks (QRNN), the Back Propagation Neural Networks (BPNN), the Radial Basis Function (RBF), the Self-Organizing Map (SOM), the Support Vector Machine (SVM), the Non-linear Auto-Regressive with eXogenous inputs neural network (NARX), the Adaptive Neural Fuzzy Inference System (ANFIS), the Convolutional Neural Networks (CNN) and the Long-Short Term Memory (LSTM) were widely introduced for water quality (Pearce et al., 2013; Jiang et al., 2016; Zhang et al., 2018; Gerhard and Gunsch, 2019; Helbich et al., 2019) and hydro-meteorological forecasting (Cannon 2011; Chang and Tsai, 2016; Zhou et al., 2019a,b).
Using self-organizing map for coastal water quality classification: Towards a better understanding of patterns and processes
2018, Science of the Total EnvironmentCitation Excerpt :Water quality degradation may threaten the long-term sustainability of the fishery. Numerous water quality investigations have been conducted by some research institutes in China for decades (Cai et al., 2007; Cai and Bai, 2011; Li et al., 2014; Lin et al., 2013; Liu et al., 2003; Wang et al., 2015; Yan et al., 2012), and the assessment and classification of water quality can facilitate the decision-making process for management (Aguilera et al., 2001b). However, most of the water characterizations or water pollution assessments have focused on one bay (or estuary) at once, and investigations of the integrated area are rare.
A cloud model-based approach for water quality assessment
2016, Environmental Research