Comparison of prediction model using spatial discriminant analysis for marine water quality index in mangrove estuarine zones

Highlights

• Comparison of SDA-ANN and SDA-MLR as a tool to predict Marine Water Quality Index (MWQI).

• No ultimate test to evaluate the model for mangrove estuarine MWQ

• SDA-ANN model provides very high MWQI prediction capacity (R² = 0.9044) than SDA-MLR (R² = 0.4863)

• SDA-ANN provides acceptable models for uncomplicated, quick computation and prediction of MWQI

Abstract

The prediction models of MWQI in mangrove and estuarine zones were constructed. The 2011–2015 data employed in this study entailed 13 parameters from six monitoring stations in West Malaysia. Spatial discriminant analysis (SDA) had recommended seven significant parameters to develop the MWQI which were DO, TSS, O&G, PO₄, Cd, Cr and Zn. These selected parameters were then used to develop prediction models for the MWQI using artificial neural network (ANN) and multiple linear regressions (MLR). The SDA-ANN model had higher R² value for training (0.9044) and validation (0.7113) results than SDA-MLR model and was chosen as the best model in mangrove estuarine zone. The SDA-ANN model had also demonstrated lower RMSE (5.224) than the SDA-MLR (12.7755). In summary, this work suggested that ANN was an effective tool to compute the MWQ in mangrove estuarine zone and a powerful alternative prediction model as compared to the other modelling methods.

Introduction

Mangroves are unique coastal ecosystems which are situated in the crossing point between land and ocean (Holguin et al., 2001). They play role as essential parts in tropical and subtropical areas. These ecosystems serve as buffer areas from tidal surges and act as stabilizing agent for the coastline (Duke et al., 2007). In addition, mangrove ecosystem provides reproducing grounds, asylum, nourishment and safe house for various essential marine species. Furthermore, due to their part in supporting marine life, mangroves are critical rearing and scrounging zones for feathered creatures and they protect these creatures from predators (Holguin et al., 2001). Based on their physical structures, mangrove forests become a buffer against hurricanes and seaside disintegration. They will discharge pollutants if their spaces are disturbed by catastrophic event or human activities.

In Malaysia, the study on mangroves is negligible and only limited data are available concerning the marine water quality (MWQ) of the mangroves estuarine zones. Marine water quality indices involve of sub-index scores assigned to each parameter by matching its measurement with a parameter-specific rating curve, optionally weighted, and combined into the final index. To assess the MWQ, marine water quality index (MWQI) has been established to provide category (Table 1) for surface marine water by computing the sub-index of standard parameters such as dissolved solid (DO), nitrate (NO₃), phosphate (PO₄), unionized ammonia (NH₃), faecal coliform, oil and grease (O&G) and total suspended solid (TSS). According to Malaysia Environmental Quality Report 2014, mangrove estuarine and river mouth water are classified as Class E (Malaysian Marine Water Quality Criteria and Standard). The MWQI scales ranges from 0 to 100 which classify the MWQ from “Excellent” to “Poor” (Table 1) (DOE, 2014). Even though MWQI development helps in understanding the water quality of one area, its development faces a great challenge to reflect the reality of a water body through a number. This challenge is due to complex environmental factors that renders the non-linear relationships among the measured parameters. In addition, MWQI based on DOE requirement does not include the heavy metals which had been reported to affect the MWQ (Maiti and Chowdhury, 2013). Hence, the heavy metals are also included in this study.

Predominantly mangrove environments are embellished with metals, resulting from urban and agricultural runoff, sewage treatment plants, industrial effluents, boating and recreational use of water bodies, chemical spills, leaching from domestic garbage dumps and mining operations (Peters et al., 1997). Most urban and industrial runoff contains a trace metals component in form of dissolved or particulate (Harbison, 1986). As stated by Thomas and Fernandez (1997) in Sandilyan and Kathiresan (2014), a number of Asian literatures has evidently identified that heavy metal pollution in Asian mangroves ecosystem is mostly because of the growth of industries in and around coastal areas in the recent past. Plus, most of the Asian mangroves’ estuaries have experienced significant direct input of contaminants as a consequence of their close proximity to urban and coastal developments. The major pollutants from anthropogenic inputs are heavy metals (MacFarlane and Burchett, 2002). As specified by several benchmark studies from the Asian and other region, the mangrove system is seriously polluted by heavy metals including Mercury (Hg), Arsenic (As), Lead (Pb), Cobalt (Co), Copper (Cu), Chromium (Cr), Cadmium (Cd), Manganese (Mn), Zinc (Zn), Nickel (Ni) and Iron (Fe) (An et al., 2010; Singh et al., 2010; Sundaray et al., 2012; Lim et al., 2012; Looi et al., 2013; Koh et al., 2015; Ali et al., 2016; Samsudin et al., 2017a, Samsudin et al., 2017b). Derisively, heavy metals are one of the main anthropogenic toxic compounds reported high in mangroves, ascending from various sources including agriculture runoff in Asia (Sandilyan and Kathiresan, 2014). Generally, water pollution by heavy metals resulting from anthropogenic impact has caused ecological complications around the world. This situation is triggered by an inadequacy of natural elimination processes for metals. As a consequence, there is metals alteration from one compartment within the aquatic environment to the ecosystem with harmful effects. In Malaysia, Shazili et al. (2006) listed the manufacturing sector is regarded as the major contributor to metal pollution in the environment. The manufacturing industries are mostly located on the west coast of peninsular Malaysia and thus, most of the studies on metal contamination have been focused there. On the other hand, only a small number of studies have provided data for the east coast of Malaysia recently (Shazili et al., 2006). To outline, the development of industrialization and other human activities in Malaysia is increasing. For that reason, heavy metals monitoring and assessment in mangrove estuary is still significant from time to time to avoid any difficulties and disastrous event in this zone. As we know, this ecosystem serves as a sink for trapping heavy metals under normal conditions. Once the mangroves are harmed either by anthropogenic activities or natural threats, they will become a source for releasing heavy metals. Mangrove loss will also diminish estuarine water quality, reduce biodiversity, annihilate fish nursery habitat and fish catches and adversely affect adjacent coastal habitats (Sandilyan and Kathiresan, 2014). This review revealed that the pollution sources in the world, especially in Malaysia, had originated from anthropogenic and minerals-related activities. The summarising of metals contamination in the estuary from this study will be useful for future comparative metals pollution studies and monitoring works on the assessment of land bases and marine pollution inputs especially in the mangrove estuary. Besides, updated metals profile from this research will aid the relevant authorities in reviewing present guidelines and enforce more stringent standards on pollutants emission into estuaries and coastal water system.

There is no ultimate test to evaluate the model, and many predictive performance models have been formulated for mangrove estuarine MWQ (Wang et al., 2005; Aertsen et al., 2010). Most researches concentrated on developing the prediction model using single MWQ parameter; however, since the studied areas are influenced by physical, chemical and biological factors, the traditional prediction method based on individual linear relationship is not sufficient for this purpose (Amiri and Nakane, 2009; Zhang et al., 2017). However, we employ spatial discriminant analysis (SDA) to provide linear combinations of the parameters and thus construct statistical classification of samples into categorical-dependent values and select the most significant parameters to the dependent variables (Hajigholizadeh and Melesse, 2017). Surface water quality models can be beneficial implements to simulate and predict the levels, distributions, risks of chemical pollutants and the concentrations of pollutants in a given water body. Thus, the model can contribute to saving the labours’ cost and materials for a large number of chemical experiments to some degree. The modelling outcomes under different pollution circumstances are very significant components for environmental impact assessment and can aid for authorities and environmental management agencies to make correct resolutions. Furthermore, it is inaccessible for on-site experiments in some cases due to special environmental pollution issues. Hence, water quality models become an important tool to recognise water pollution and the final fate of pollutants in a waterenvironment.

Overall, this study deliberates two types of model which are used as a well-organized decision support systems and estuary management. Water quality models can imitate the biological and chemical processes that occur within a water body framework, on the basis of sources, info and responses. The water quality model’s development hinge on the several objectives and aims, which attribute to a number of different modelling methods. Modelling procedures are almost certainly categorized as being either deterministic (the output of the model is fully determined by the parameter values and the initial conditions), stochastic (possess some inherent randomness), hybrid ("mixture" of both Deterministic and Stochastic) or statistically based. Deterministic models attempt to portray all the physical and chemical progressions implicated in the terms of mathematical, with parameters gained either assessed by monitoring or empirically or from available data or experience. The differential equations are usually simplified in order to find solutions that are suitable for the purpose of the model.

Multiple linear regressions (MLR) is a multivariate statistical technique that was often applied in a particular study in order to predict relationships between input and output variables without detailing the causes of these relationships (Dominick et al., 2012; Samsudin et al., 2017a, Samsudin et al., 2017b). From the opposed point of view, MLR has also been carried out in order to measure the relationship between the independent and dependent variable (Guillén-Casla et al., 2011; Dominick et al., 2012; Samsudin et al., 2017a, Samsudin et al., 2017b). On the other side, Artificial Neural Network (ANN) model is an application of random or deterministic models which can allow a prediction of the complication of the variables (Juahir et al., 2004; Palani et al., 2008; Mutalib et al., 2013; Azid et al., 2014; Samsudin et al., 2017a, Samsudin et al., 2017b). Several studies have proven that the ANN model is the best tool to model a non-linear environmental relationship (Azid et al., 2014). ANN have the ability to learn non-linear relationships between the variables and complex patterns in datasets that are not well defined by simple mathematical formulae and they can be trained precisely when presented with a new dataset (Kurt et al., 2008; Azid et al., 2014; Samsudin et al., 2017a, Samsudin et al., 2017b). Model validation that developed using ANNs or MLR techniques are the suitable options for environmental modelling (Amiri and Nakane, 2009). According to Jeong et al. (2005) and Maier and Dandy (2001), algorithm in machine learning such as artificial neural network (ANN) is capable to deal with non-linear distribution data based on adaptive or heuristic methods. Thus, many analyses with sophisticated data had utilised the ANN and MLR techniques to establish specific modelling and understand the complex data matrix. Based on Zhang et al. (2017), the ANN model had proven its effectiveness as a practical tool for short term water quality forecasting. The purposes of this study to identify the most significance MWQ parameters based on region (West Coast and East Coast of Peninsular Malaysia), to develop a new MWQI model based on the significant parameters and to compare the best MWQI prediction model (MLR and ANN).