Abstract

This paper aims to assess the performance of a distributed hydrological model for simulating the transport of various heavy metals in rivers, to enhance and support environmental monitoring strategies for rivers in developing countries. In this context, we evaluated the performance of the Geophysical flow Circulation (GeoCIRC) model based on Object-Oriented Design (OOD) for the simulation of contamination from multiple heavy metals (Pb, Hg, Cr, and Zn) in Harrach River in Algeria. The results of the case study were in good agreement with the observations. Methodology for the assessment of data quality control and the improvement of monitoring procedures was proposed by using the hydrological model to simulate different scenarios. The GeoCIRC-model-based OOD allowed the prediction of the concentrations of heavy metals with minimal input data. Also, various heavy metals could be numerically treated simultaneously because the OOD increases the model's flexibility to allow the handling of many transportable materials. Therefore, the GeoCIRC model is a powerful tool for the monitoring of environmental contamination in rivers by various heavy metals.

INTRODUCTION

Industrial activities can cause severe damage to water eco-systems. The pollution of rivers by industrial activities can occur through wastewater disposal, where treatment facilities are inadequate or improperly operated. In urban areas, intensive industrial activities cause a significant increase in the quantities of hazardous materials being released into rivers, such as heavy metals. As a part of the project of technical cooperation between the Observatoire National de l'Environnement et de Developpement Durable (ONEDD) and the International Cooperation Agency of Japan (JICA), an evaluative study was carried out to determine the levels of heavy metal concentrations in Harrach River (an urban river in Algeria) and its tributaries, from 2004 to 2011. They found the concentrations of various heavy metals were extremely high in most of the sampling locations along the Harrach River basin. The study concluded that major industrial activity causes a significant source of pollution in Harrach River because of untreated wastewater that factories discharge directly into the river significantly increasing the concentrations of various heavy metals at downstream points (Yoshida et al. 2005). Heavy metals can result in adverse effects for the environment and organisms because of their potential to accumulate and persist in the environment, and their toxicity even at low concentrations (Duruibe et al. 2007), which makes the monitoring of these pollutants important for the management of river water quality. Several studies related to the contamination of rivers by heavy metals have been carried out with different focuses, i.e., distribution, contamination, and treatment methods. Zhao et al. (2017) assessed the spatial distribution of heavy metals in sediments and the state of heavy metal contamination using different indicators such as the enrichment factor and geo-accumulation index, and furthermore they analyzed and distinguished the possible sources of heavy metals. Roque et al. (2018) used starch-based xerogel from potato (Solanum tuberosum) peels for the depuration process of domestic sewage, textile effluents and acid mine drainage; the results showed that the treatment was effective at achieving the reduction of different heavy metal levels such as those of Zn, Cd. Tukura (2015) assessed the changes of various heavy metals in sediment and water in rainy and dry seasons.

The use of mathematical models has become widespread in many fields, particularly in environmental fields. Pak et al. (2015) used the Environmental Fluid Dynamics Code (EFDC) model for the simulation of discharge and total suspended solid concentrations upstream and downstream of the Baekje Weir installed in Geum River, Korea. The EFDC model was developed for application in estuaries, coastal regions, lakes and rivers to simulate many environmental factors, such as water level, velocity, salinity, movement of cohesive and non-cohesive sediment and toxic pollutant reactions (Pak et al. 2015). Zhang et al. (2017) proposed a predictive method by applying an integrated model consisting of the Radial Basis Function (RBF) neural network and improved case-based reasoning (CBR). Their proposed model effectively predicted the biochemical oxygen demand (BOD) in the treatment of wastewater. Gao et al. (2017) used the Hydrologic Engineering Center Hydrologic Modeling System (HEC-HMS) to simulate basin runoff and to examine the impact of urban agglomeration polders on flood events in the Qinhuai River basin in China.

Although mathematical models are used extensively in the field of environmental water quality assessment, water quality monitoring is generally conducted through large collection and analysis samples, in situ and/or in laboratories (Bartram & Ballance 1996). The modeling process takes place in three important steps. The first step includes collection, preparation of input data, and evaluation of parameters needed to set up a model, and prepare for execution of the model. The second step is the model testing, which involves calibration, validation, as well as, when possible, post-audit. In this step, the model will be evaluated to assess whether it can reasonably represent the purposes of the study. The final step includes the ultimate use of the model. The use of numerical models can facilitate the assessment procedure and provide a reasonable and fast descriptive framework for existing water quality problems with low costs (Falconer et al. 2005), whereas the simple water quality measurement methods need time and are labor-intensive as well as having high costs. The ease of use, accessibility and low cost of computer systems and software contribute significantly to the attractiveness of numerical models as useful monitoring tools. Thus, numerical models provide an opportunity to improve monitoring methods, both in their operational performance and cost (Palmer 2001).

Previous studies in the field of water quality modeling (e.g. Fanger et al. 1986; Kachiashvili et al. 2007; Ani et al. 2009) provided results which were helpful for understanding the dynamics of the transport of pollutants such as heavy metals in rivers. In the context of reducing the concentration of heavy metals, many studies have been conducted on the behavior and transport of these pollutants in rivers using hydrological models. Kashefipour & Roshanfekr (2012) and Falconer et al. (2005) studied the different processes involved in the transport and distribution of pollutants in rivers and estuaries using hydrological models, and they proposed a conceptual framework in order to fill the gaps which would hinder the application of these models in water quality management.

Usually in simulation studies, researchers or engineers consider heavy metals' transport processes in rivers separately for different elements and reactions. For example, Kashefipour & Roshanfekr (2012) provided a methodology predicting a varying reaction coefficient for dissolved concentrations of Pb and Zn using pH and EC, which affect the reaction coefficient in the advection–dispersion equation, for improved accuracy. Likewise, they introduced the best relationships for dissolved Pb and Cd reaction coefficients for the simulation of dissolved heavy metals. The simulation results were in good agreement with the corresponding measured values. Falconer & Lin (2003) predicted the distribution of heavy metal concentration along the Mersey basin, UK, using hydro-environmental models (Wu & Falconer 2000; Wu et al. 2001), with the three-dimensional advective–diffusion equation and dynamic partitioning coefficients being found to give improved agreement between predicted and measured heavy metals to improve the models' accuracy. It should be noted that most of the researchers considered the heavy metal transport process for each element in the rivers separately in each advection–dispersion equation, and the chemical reactions. On the other hand, the calibration of models requires large numbers of observations, which are rarely available, especially in developing countries, where contamination of rivers with heavy metals typically occurs, such as in Harrach River in Algeria. Harrach River has been severely polluted by many industries with heavy metals (Yoshida et al. 2005). The weakness in the enforcement of environmental laws and the absence of extensive pollution monitoring has led to the unavailability of sufficient observational data on pollutants in stream water and sediment to investigate environmental countermeasures.

According to Bouragba et al. (2017), the Geophysical flow Circulation (GeoCIRC) model based on Object-Oriented Design (OOD), originally developed by Nakayama & Shintani (2015), could be applied to estimate heavy metal concentrations when many unknown point sources exist. The model was successfully used to estimate lead (Pb) and mercury (Hg) contamination in Harrach River in Algeria based on only a few observations in stream water and sediment. If a hydrological model is applied to investigate multiple components within this framework, the inherent characteristics of the object-oriented approach allow an extension to include a new scheme in the model, which already exists in the module, and increase the model's flexibility without changing the fundamental structure (Kang et al. 2012). Also, the use of OOD allows the model to be easily extended to handle multiple transportable materials. Therefore, the incorporation of various heavy metals in the model can contribute to the assessment of the impact of point sources in different pollutants simultaneously even when only a few observations are available from water sampling campaigns.

This study aims to assess the performance of hydrological models for the prediction of the transport of multiple heavy metals in rivers in order to investigate strategies for countermeasures for contamination in developing countries, particularly where few observational data are available. In the present study, a methodology is presented to apply the GeoCIRC model to evaluate pollution by multiple heavy metals in Harrach River in Algeria, where few observational data are available because of difficulties with information disclosure. The concentration of four heavy metals originating from industrial activities (i.e., lead (Pb), mercury (Hg), chromium (Cr), and zinc (Zn)) in water and sediment in the river network were simulated. A series of simulations were also carried out in two improvement/evaluation steps: (1) computations including new point sources and (2) the intentional ignoring of pollution sources. The results showed the ability of the GeoCIRC model to simulate the transport of multiple heavy metals in a river simultaneously and present a comprehensive description of river contamination with only a minimal amount of observational data.

METHODS

The framework for the GeoCIRC model incorporating the transport of two heavy metals by Bouragba et al. (2017) was extended and applied in the present study.

Study site

Harrach River is located at the northern tip of Algeria (Figure 1), with a surface area of 1,236 km2. The climate is Mediterranean, with an accumulation of annual rainfall of 805 mm. The water resources in the river are rain, infiltration and surface runoff. Harrach River is fed by industrial wastewater outflow (Louati 2015), which may explain the pollution of the river with various chemical substances such as heavy metals. Since more than 300 factories are distributed in different ‘industrial areas’ around the Harrach region, as shown in Figure 2, industrial activity is one of the main sources of pollution with heavy metals in the Harrach watershed. Most of these industrial units do not have treatment equipment for heavy metals, which increases the level of the river contamination with these pollutants. The evaluation studies carried out by ONEDD/JICA on the effects of industrial activities on the water quality of Harrach River found that the river is heavily polluted with heavy metals such as Pb, Zn, Cr, As, and Hg (Yoshida et al. 2005).

Figure 1

Study area location.

Figure 1

Study area location.

Figure 2

Localization of industrial areas and observation points in the study area.

Figure 2

Localization of industrial areas and observation points in the study area.

Model formulation

The GeoCIRC model is a numerical model that can analyze the interaction between essential hydrological processes, and simulate the different flow regimes, i.e. water flow, river flow, infiltration flow and groundwater flow, based on physical processes. In this study, the GeoCIRC model has been extended and improved to simulate the transport of four heavy metals: Pb, Hg, Cr, and Zn.

River flow model

The longwave equations were applied to estimate water level and discharge. The predictor–corrector method proposed by Nakayama et al. (2000) was used, as in Equations (1a)–(1e). The validity and the applicability of the numerical scheme were verified from comparisons with the fully nonlinear and strongly dispersive internal wave equations (Nakayama & Kakinuma 2010; Kobayashi et al. 2017; Nakayama et al. 2018). One of the most significant characteristics of the GeoCIRC model is that all branch rivers are connected to the main river as lateral inflows, which enables the reproduction of progressive longwaves upstream without any discontinuity in the momentum (Equation (1c)): 
formula
(1a)
 
formula
(1b)
 
formula
(1c)
 
formula
 
formula
(1d)
 
formula
(1e)
where uin = velocity vector at grid i and time n, g = gravitational acceleration, hi = depth at grid i, Ai = cross-sectional area at grid i, Δt = time step, Zb = distance from datum level to bottom boundary, nm = Manning's roughness coefficient which is estimated as 0.04 in the present simulation, and qL = lateral inflow.

Heavy metal transport model

The distribution of dissolved heavy metals in rivers can be modeled by the advection equation (Kashefipour & Roshanfekr 2012). The 1D advection equation was used to simulate the dissolved heavy metal concentration as follows: 
formula
(2)
where C = dissolved heavy metal concentration, Q = discharge, Aa = source/sink of dissolved heavy metal, and Ba = transformation flux from, or to, adsorbed particulate phase onto the sediment.
Sources/sinks of dissolved heavy metals can be estimated using the following relation (Kashefipour & Roshanfekr 2012): 
formula
(3)
where Ca = lateral inflow of heavy metal concentration, Δx = distance between two consecutive cross-sections which can be either constant or variable.
The transport of adsorbed particulate heavy metal is described by the following equation: 
formula
(4)
where SP = heavy metal concentration adsorbed in particulates, ASPa = source or sink of adsorbed particulate heavy metal, and BSPa = transformation flux from, or to, the adsorbed particulate phase onto the sediment.
The total concentration can be estimated by summing the dissolved heavy metal and the adsorbed particulate heavy metal. According to Falconer & Lin (2003), the distribution of heavy metals between the dissolved and adsorbed particulate phases can be described by the partition coefficient: 
formula
(5)
 
formula
(6)
where CT = total concentration, Kd = partition coefficient, SP = concentration of heavy metals adsorbed in particulates, and C = concentration of heavy metals dissolved in the water column.
From Equations (5) and (6), the dissolved heavy metal can be expressed as follows: 
formula
(6a)
In this study, the interaction of the heavy metal concentration between stream water and sediment was assumed to be in an equilibrium state, which is expressed in Equation (6a).

Data preparation

Watershed and river network characteristics

To resolve surface topography as well as the spatial distribution of wastewater factories, ArcGIS was used to generate the watershed and river network. Digital Elevation Model (DEM) data obtained from the Consortium for Special Information (CGIAR-CSI) were used to calculate the river network. The resultant rectangular raster grid has 246 columns and 201 rows. The delineated river network comprises five branches numbered from 0 to 4 (Figure 1). The watershed and the river network were simulated with a surface grid of 100 m cells.

Input data

In the first simulation scenario (Case 1), the input data includes the concentrations of four heavy metals, Pb, Hg, Cr, and Zn (Table 1). The model was run with the same input data which were used in a previous simulation (Bouragba et al. 2017) with the addition of Cr and Zn data. Moreover, new point source data which were not included in the previous simulation were added (Figure 3 and Table 1).

Table 1

Input data of the concentration of wastewater discharged from the factories

Factory nameArea locationDischarge locationFlow rate Q (m3/day)Pb (mg/L)Hg (μg/L)Cr (mg/L)Zn (mg/L)
ENMTPb river 0 0.20 8.36 0.01 
ENPCb II river 0 0.27 12a 
Raff Alger IV river 1 0.51 18a 
EMB1 III river 1 320 2.40a 21.21a 1.47 
BAG III river 1 100 0.45 3.54 0.60 
SOACHLORE river 1 930 5,720a 
Est KEHRIb IV river 1 50 0.23 7.28 0.96a 
Tan Semmache VI river 3 30 2.23a 11.04a 60.43a 
Tan KEHRIb VII river 3 120 0.27 0.54a 
AGENORE VI river 3 3,000 0.34 17a 0.15 1.08 
CATEL VI river 3 12 0.94a 0.63 
AVENTISb VI river 3 0.38 0.71 
ENAPb VI river 3 12.11a 
ENPEC VI river 3 150 37a 10.23a 
Hydrotraitment VI river 3 22a 0.68a 54a 
Factory nameArea locationDischarge locationFlow rate Q (m3/day)Pb (mg/L)Hg (μg/L)Cr (mg/L)Zn (mg/L)
ENMTPb river 0 0.20 8.36 0.01 
ENPCb II river 0 0.27 12a 
Raff Alger IV river 1 0.51 18a 
EMB1 III river 1 320 2.40a 21.21a 1.47 
BAG III river 1 100 0.45 3.54 0.60 
SOACHLORE river 1 930 5,720a 
Est KEHRIb IV river 1 50 0.23 7.28 0.96a 
Tan Semmache VI river 3 30 2.23a 11.04a 60.43a 
Tan KEHRIb VII river 3 120 0.27 0.54a 
AGENORE VI river 3 3,000 0.34 17a 0.15 1.08 
CATEL VI river 3 12 0.94a 0.63 
AVENTISb VI river 3 0.38 0.71 
ENAPb VI river 3 12.11a 
ENPEC VI river 3 150 37a 10.23a 
Hydrotraitment VI river 3 22a 0.68a 54a 

aConcentration values which are over concentration standards of the general regulations for wastewater qualities in Algeria (i.e., 0.5 mg/L, 0.5 mg/L, 0.01 mg/L, and 3 mg/L, for Pb, Cr, Hg, and Zn, respectively) (Yoshida et al. 2005).

bThe factories which were not included in input data in the previous simulation done by Bouragba et al. (2017).

Figure 3

Locations of the major factories as point sources of heavy metal pollution in the study area: (a) factories included in a previous simulation study (Bouragba et al. 2017), (b) factories included in the present simulation.

Figure 3

Locations of the major factories as point sources of heavy metal pollution in the study area: (a) factories included in a previous simulation study (Bouragba et al. 2017), (b) factories included in the present simulation.

Model setup

The hydrodynamic module of the GeoCIRC model was set up to simulate the distribution, velocity of water river and heavy metal concentrations in Harrach River. The transport of dissolved heavy metals such as Pb, Hg, Cr, and Zn was modeled as instances of transportable objects. The interaction of metal concentrations between stream water and sediment was estimated using a partition model under the assumption of an equilibrium state. The model was solved using Mathematica software by analyzing the transport process of heavy metal concentrations at each branch of the river network.

In the upstream reach, there is a large catchment which is not industrialized or polluted. Therefore, we placed in river 1 a virtual inflow point as a boundary condition instead of river discharge from the upstream. The discharge of 105 m3/day was determined to be the same order of magnitude as the average river discharge (at the downstream end) multiplied by the ratio of the area of the upstream catchment to the area of the whole catchment, whereas the boundary inflow at the upstream end of river 1 was given in the concentrations of 1 mg/L, 5 × 10−3 mg/L, 0.4 mg/L, 0.01 mg/L in Pb, Hg, Cr and Zn, respectively, and were modeled as a virtual inflow at the river connection point of river 1, at the most-meandering point (Figure 2).

At some factories near rivers 0 and 3, the discharge data were not available (Table 1). However, these points are located in a highly industrialized area where there are many factories. Therefore, point source loads from these areas were assumed to be 104 m3/day and 103 m3/day at rivers 0 and 3, respectively, in order to represent the river discharge at the downstream end of river 0, following a previous study (Bouragba et al. 2017).

The concentration of metals in sediment was estimated with a partition coefficient. The model was tested with average partition coefficient values estimated from the measurement data (source: ONEDD) as KdPb = 430 L/kg, KdHg = 520 L/kg, KdCr = 950 L/kg, and KdZn = 4 × 103 L/kg for Pb, Hg, Cr, and Zn, respectively. However, according to the measurement data obtained in river 1, the concentration of Hg in sediment was larger and lower for Cr compared with the concentrations in other rivers (Table 2). Furthermore, the partition coefficient estimated from the observation results shows at least two unrealistic fluctuations. Therefore, the model was tested for river 1 with average partition coefficient values estimated from the measurement data from the area of river 1 as KdHg = 14 × 103 L/kg and KdCr = 380 L/kg for Hg and Cr, respectively, to ensure the results agree with the observation data.

Table 2

Heavy metal concentrations in stream water (mg/L) and sediment (mg/kg) at different points along the study area

LocationPb
Hg
Cr
Zn
Water (mg/L)Sediment (mg/kg)Water (mg/L)Sediment (mg/kg)Water (mg/L)Sediment (mg/kg)Water (mg/L)Sediment (mg/kg)
river 0 0.73a 200 24 × 10−4a 3.40a 0.20 640a 0.03 1,300a 
river 0 0.60a 83 18.2 × 10−4a 0.50 0.20 168 0.13 218 
river 0 0.20 287a 10−3a 0.20 0.20 374a 0.19 741a 
river 0 0.89a 170 12 × 10−4a 0.30 0.54a 190 1.20 1,100a 
river 1 0.20 144 10−3 0.80a 0.20 68 0.16 211 
river 1 0.57a 130 72 × 10−4a 105a 0.20 76 0.03 38 
river 3 0.89a 142 10−3a 0.20 0.54a 521a 0.20 741a 
EQS 0.50 218 5 × 10−4 0.71 0.50 370 410 
LocationPb
Hg
Cr
Zn
Water (mg/L)Sediment (mg/kg)Water (mg/L)Sediment (mg/kg)Water (mg/L)Sediment (mg/kg)Water (mg/L)Sediment (mg/kg)
river 0 0.73a 200 24 × 10−4a 3.40a 0.20 640a 0.03 1,300a 
river 0 0.60a 83 18.2 × 10−4a 0.50 0.20 168 0.13 218 
river 0 0.20 287a 10−3a 0.20 0.20 374a 0.19 741a 
river 0 0.89a 170 12 × 10−4a 0.30 0.54a 190 1.20 1,100a 
river 1 0.20 144 10−3 0.80a 0.20 68 0.16 211 
river 1 0.57a 130 72 × 10−4a 105a 0.20 76 0.03 38 
river 3 0.89a 142 10−3a 0.20 0.54a 521a 0.20 741a 
EQS 0.50 218 5 × 10−4 0.71 0.50 370 410 

aConcentration values which are over the concentrations specified in environmental quality standards (EQS) (Yoshida et al. 2005).

In order to assess the effect of the uncertainty of the input point sources in the simulation and to assess the model's performance in the simulation of multiple heavy metals with minimum input data, a series of simulations were conducted in this study. The model was run for seven different scenarios in two improvement/evaluation steps. In the first step, a new computation was conducted for each scenario (case 1 and case 2) by adding new point sources. In the second step, the model was run without including the pollution sources with high concentrations separately for each factory (cases 3, 4, 5, 6, 7), as shown in Table 3 and Figure 4. It is worth mentioning that the factories which were not included were chosen according to the magnitudes of the pollution they produced, i.e., AGENORE, SOACHLORE, Tan Semmache, and ENPEC have the largest mass fluxes of Zn, Hg, Cr, and Pb, respectively. EMB1 has the second largest mass flux of Zn and the third largest mass fluxes of Pb and Hg. Declining large mass flux sources in each heavy metal element corresponds to a kind of sensitivity analysis for each element in this river basin system.

Table 3

A summary of different simulation cases

CaseDescription
Case 1 Simulation of 4 elements (Pb, Hg, Cr, Zn) with the addition of 6 factories 
Case 2 Simulation of 4 elements (Pb, Hg, Cr, Zn) without adding 6 factories 
Case 3 Simulation of 4 elements (Pb, Hg, Cr, Zn) without AGENORE factory 
Case 4 Simulation of 4 elements (Pb, Hg, Cr, Zn) without EMB1 factory 
Case 5 Simulation of 4 elements (Pb, Hg, Cr, Zn) without SOACHLORE factory 
Case 6 Simulation of 4 elements (Pb, Hg, Cr, Zn) without Tan Semmache factory 
Case 7 Simulation of 4 elements (Pb, Hg, Cr, Zn) without ENPEC factory 
CaseDescription
Case 1 Simulation of 4 elements (Pb, Hg, Cr, Zn) with the addition of 6 factories 
Case 2 Simulation of 4 elements (Pb, Hg, Cr, Zn) without adding 6 factories 
Case 3 Simulation of 4 elements (Pb, Hg, Cr, Zn) without AGENORE factory 
Case 4 Simulation of 4 elements (Pb, Hg, Cr, Zn) without EMB1 factory 
Case 5 Simulation of 4 elements (Pb, Hg, Cr, Zn) without SOACHLORE factory 
Case 6 Simulation of 4 elements (Pb, Hg, Cr, Zn) without Tan Semmache factory 
Case 7 Simulation of 4 elements (Pb, Hg, Cr, Zn) without ENPEC factory 
Figure 4

Simple chart for the simulation cases in this study.

Figure 4

Simple chart for the simulation cases in this study.

RESULTS AND DISCUSSION

In the present study, the proposed approach was verified with heavy metal concentration measurement data at different points along the study site (Figure 2 and Table 2). Since the input data for the point sources in the present study were mostly located in rivers 0, 1, and 3, the metal concentrations were obviously high in these rivers compared with the other rivers.

Assessment of data quality control and improvement of monitoring plan

Figures 5 and 6 show the simulation results for case 1, for the concentrations in stream water and sediment, respectively, for each metal in the downstream direction. The model predicted the transport of metals that originated from industrial wastewater along Harrach River. The root mean squared errors (RMSEs) of the concentration of metals dissolved in water were 9.7 × 10−2, 8.8 × 10−3, 5.2 × 10−2, and 3.5 × 10−2 mg/L for dissolved Pb, Hg, Cr, and Zn, respectively, and 74.85, 21.84, 148.72, and 179.13 mg/kg for Pb, Hg, Cr, and Zn, respectively, in sediment. These RMSEs are lower than the EQS values (Table 2) for Pb, Cr, and Zn in both stream water and sediment. However, the RMSEs of Hg in both stream water and sediment are higher than the EQS values (Table 2). The simulation results were possibly affected by errors in the input data due to unknown point sources or greater uncertainties in the chemical behavior of Hg compared with the other elements (Pb, Cr, and Zn).

Figure 5

Concentration pattern in stream water for each metal in the downstream direction (horizontal axis: distance from the upstream end (m); vertical axis: concentration; river 0, river 1, river 2, river 3, and river 4: river branches; ▴: point source; •: observation data; dashed line: simulation data (case 1) in stream water (mg/L)).

Figure 5

Concentration pattern in stream water for each metal in the downstream direction (horizontal axis: distance from the upstream end (m); vertical axis: concentration; river 0, river 1, river 2, river 3, and river 4: river branches; ▴: point source; •: observation data; dashed line: simulation data (case 1) in stream water (mg/L)).

Figure 6

Concentration pattern in sediment for each metal in the downstream direction (horizontal axis: distance from the upstream end (m); vertical axis: concentration; river 0, river 1, river 2, river 3, and river 4: river branches; ▴: point source; •: observation data; dashed line: simulation data (case 1) in sediment (mg/kg)).

Figure 6

Concentration pattern in sediment for each metal in the downstream direction (horizontal axis: distance from the upstream end (m); vertical axis: concentration; river 0, river 1, river 2, river 3, and river 4: river branches; ▴: point source; •: observation data; dashed line: simulation data (case 1) in sediment (mg/kg)).

In simulation case 1, the correlations between the simulation and the measurements for the metal concentration in stream water are high, i.e., 0.82, 0.95, and 0.98 for Pb, Cr, and Zn, respectively, although it is lower for Hg, i.e., 0.35 (Figure 7). The correlations for the metal concentration in sediment were more varied, i.e., 0.25, 0.79, 0.53, and 0.67 for Pb, Hg, Cr, and Zn, respectively, as shown in Figure 8. This may be because of the dependency of metal accumulation on sediment properties, as well the dependency of metal concentration on changes in physiochemical parameters. Pb has a higher tendency to be adsorbed to sediment organic matter than to complex with organic matter in water. Also, oxide and hydroxide of Pb have a higher solubility than those of Cr and Zn at constant pH. In addition, the ion valence and sediment particle influence metal adsorption to sediment particles, e.g. fine clay particles deposited in a mild slope river can adsorb heavy metals more than sand particles (Tukura 2015).

Figure 7

Correlation between calculation (case 1) and observation results for each metal concentration in stream water (mg/L).

Figure 7

Correlation between calculation (case 1) and observation results for each metal concentration in stream water (mg/L).

Figure 8

Correlation between the simulations (case 1) and observations for each metal concentration in sediment (mg/kg).

Figure 8

Correlation between the simulations (case 1) and observations for each metal concentration in sediment (mg/kg).

The contamination of Harrach River basin with multiple heavy metals (namely: Pb, Hg, Cr and Zn) could be numerically simulated with a few observation data. The GeoCIRC model with OOD was verified as a useful tool for simulation of multiple heavy metal contaminations because of its flexibility of handling many transportable materials (Bouragba et al. 2017). That is, the column object which constitutes the calculation domain is generally incorporated in order to govern the vertical transport of physical quantities, along with a connection object which controls the horizontal transport of physical quantities between column objects (Nakayama & Shintani 2015). Thus, various transportable quantities including the target chemical substances which are discharged in the watershed can be arbitrarily instantiated in column objects efficiently even when adding multiple metal elements. According to the simulation results, the chemical reaction under the equilibrium state seems reasonable for predicting metal transport because the chemical interaction between stream water and sediment is sufficiently fast compared with metal transport in the river under the observed conditions.

The averaged RMSEs for the different simulation cases were obtained using the original dataset for all of the measurement points in the study site in units of mg/L in stream water and mg/kg in sediment (Table 4).

Table 4

Calculated RMSE for each simulation case

Pb
Hg
Cr
Zn
In water (mg/L)In sediment (mg/kg)In water (mg/L)In sediment (mg/kg)In water (mg/L)In sediment (mg/kg)In water (mg/L)In sediment (mg/kg)
Case 1 9.7 × 10−2 74.85 8.8 × 10−3 21.84 5.2 × 10−2 148.72 3.5 × 10−2 179.13 
Case 2 9.8 × 10−2 74.89 8.8 × 10−3 21.86 5.2 × 10−2 148.79 3.5 × 10−2 179.32 
Case 3 9.8 × 10−2 74.82 9.1 × 10−3 21.94 5.2 × 10−2 148.52 3.9 × 10−2 217.32 
Case 4 9.7 × 10−2 74.94 9.3 × 10−3 13.43 5.2 × 10−2 148.66 3.6 × 10−2 200.23 
Case 5 9.8 × 10−2 74.80 9.5 × 10−3 21.84 5.2 × 10−2 148.54 3.5 × 10−2 197.10 
Case 6 9.8 × 10−2 74.89 9 × 10−3 22.37 13.6 × 10−2 153.73 3.5 × 10−2 197.36 
Case 7 11.1 × 10−2 77.70 1.4 × 10−3 22.25 5.2 × 10−2 148.54 3.5 × 10−2 197.54 
Pb
Hg
Cr
Zn
In water (mg/L)In sediment (mg/kg)In water (mg/L)In sediment (mg/kg)In water (mg/L)In sediment (mg/kg)In water (mg/L)In sediment (mg/kg)
Case 1 9.7 × 10−2 74.85 8.8 × 10−3 21.84 5.2 × 10−2 148.72 3.5 × 10−2 179.13 
Case 2 9.8 × 10−2 74.89 8.8 × 10−3 21.86 5.2 × 10−2 148.79 3.5 × 10−2 179.32 
Case 3 9.8 × 10−2 74.82 9.1 × 10−3 21.94 5.2 × 10−2 148.52 3.9 × 10−2 217.32 
Case 4 9.7 × 10−2 74.94 9.3 × 10−3 13.43 5.2 × 10−2 148.66 3.6 × 10−2 200.23 
Case 5 9.8 × 10−2 74.80 9.5 × 10−3 21.84 5.2 × 10−2 148.54 3.5 × 10−2 197.10 
Case 6 9.8 × 10−2 74.89 9 × 10−3 22.37 13.6 × 10−2 153.73 3.5 × 10−2 197.36 
Case 7 11.1 × 10−2 77.70 1.4 × 10−3 22.25 5.2 × 10−2 148.54 3.5 × 10−2 197.54 

In the simulation results for cases 1 and 2, it seems that the results of case 1 are slightly more accurate than for case 2, which means the inclusion of the additional six factories did not significantly affect the model performance. The RMSEs in cases 1 and 2 were (9.7 × 10−3, 9.8 × 10−3), (8.8 × 10−3, 8.8 × 10−3), (5.2 × 10−2, 5.2 × 10−2), and (3.5 × 10−2, 3.5 × 10−2) for Pb, Hg, Cr, and Zn in water (mg/L), respectively, and (74.8, 74.9), (21.8, 21.9), (148.7, 148.8), and (179.1, 179.3) for Pb, Hg, Cr, and Zn in sediment (mg/kg), respectively.

The results for simulation cases 3, 4, 5, 6, and 7 show that point sources with a high concentration and large discharge magnitude have a clear effect on the model's accuracy (Table 4).

For Pb, the RMSEs for the simulation results for cases 3, 4, 5, and 6 were slightly higher than for case 1 (Table 4). Case 7 had the highest RMSE, i.e., 11.1 × 10−2 mg/L in water and 77.70 mg/kg in sediment. This is because the ENPEC factory (the source with the highest concentration of Pb, Table 1) which discharged into river 3 was ignored in case 7.

For Hg, there was a large variation in the results, especially in cases 4 and 7. By omitting the EMB1 and ENPEC factory point sources, the RMSEs for these cases decreased for water and sediment, i.e., 1.4 × 10−3 mg/L in case 7 and 13.43 mg/kg in case 4, as shown in Table 4. In these cases, the EMB1 and ENPEC factories are located in rivers 1 and 3, respectively, and the RMSE for the Hg might be strongly affected by the large errors at F and G which are located in rivers 1 and 3, respectively. Inconsistencies between the partition coefficient in river 1 and the other rivers in the model may also be related to these errors. Therefore, from these results the errors in the input data for the factories in river 1, and the errors in the observed concentrations in stream water, and uncertainties in the chemical behavior of Hg, may be estimated.

Considering Cr, the simulation results which omitted the Tan Semmache factory (which had the highest concentration of Cr among all factories) in case 6 had the highest RMSEs, i.e., 13.6 × 10−2 mg/L in water and 153.73 mg/kg in sediment, respectively (see Table 4). The Cr discharged by the Tan Semmache factory into river 3 has a large impact on determining the Cr distribution in the study site.

For Zn, ignoring the EMB1 factory (which had the highest concentration of Zn among all factories) in simulation case 5 resulted in the highest RMSEs, i.e., 3.9 × 10−2 mg/L in water and 217.32 mg/kg in sediment, respectively (see Table 4). The discharge by the EMB1 factory into river 1 is thus the most important for determining the Zn distribution in the study site.

Omitting point sources with high concentrations can compromise the model's performance as mentioned above. The simulation results may also be affected by the quality of the input data and verification data, or by uncertainties in the chemical behavior such as for non-equilibrium reactions, accumulation in sediment, etc., for each element. Large variations in the results due to mixing two or more branch rivers can also affect the verification of the concentration of pollutants. In the present study, the relation between different factories and the concentrations of Pb, Cr, and Zn in water and sediment were shown. However, the RMSEs for the simulation results could only be verified with a dataset which contained irrelevant measurement points. To compensate for this, the downstream concentrations in river 0, the middle reach of river 1, and the upstream reach of river 3 where large variations in the downstream direction were found, as shown in Figures 5 and 6, should be closely monitored. Furthermore, although the simulations were shown to accurately capture the Pb, Cr, and Zn concentrations, the simulation of the Hg concentration requires further consideration.

In the present study, the model was able to accurately estimate the Pb, Cr, and Zn concentrations in Harrach River, whereas for Hg the predictions require further consideration. However, the model can be used to help calibrate and supplement the observation data for unknown parameters, such as the concentration and discharge from unrevealed point sources, the partition coefficient, etc. This is one advantage of using a simulation approach for multiple heavy metals in river basins with large sources of uncertainty. Also, from the perspective of environmental monitoring and management, data for multiple elements observed even over very short-term periods can be effectively utilized to calibrate hydrological models for future planning.

In the study published by Falconer & Lin (2003), they used salinity to model the partitioning coefficient of heavy metals in the Mersey estuary and found that their simulation results agreed well with the measurements. The RMSEs of both the Cd and Zn concentrations dissolved in water were obtained, 2.4 × 10−2 and 3.5 × 10−2 mg/L, respectively. These obtained RMSEs are lower than the corresponding EQS values (0.2 mg/L and 3 mg/L for Cd and Zn, respectively). Kashefipour & Roshanfekr (2012) provided a methodology for predicting the variable reaction coefficient for dissolved Pb and Cd using environmental factors that affect the reaction coefficient in the advection–dispersion equation, to improve the model's accuracy. The RMSEs between the simulations and observations were 67.1 × 10−3 and 3.5 × 10−3 mg/L for Pb and Cd, respectively. In either case with each element, the obtained RMSE values were lower than for the EQS (0.5 mg/L and 0.2 mg/L for Pb and Cd, respectively). Likewise, the model in the present study well simulated the concentration of heavy metals dissolved in water; where the RMSEs of dissolved heavy metals in case 1 were 9.7 × 10−2, 5.2 × 10−2, and 3.5 × 10−2 for Pb, Cr, and Zn, respectively, although the model did not include the dynamics of the chemical interactions of the heavy metals between the stream water and sediment. However, the chemical interaction between the stream water and sediment is faster than the metal transport in the river under the observation conditions, which suggests that the equilibrium assumption in the present simulation was suitable for predicting the concentration of heavy metals dissolved in water and accumulated in sediment. Thus, the present model could successfully simulate the concentration of heavy metals with greater accuracy than the EQS.

Compared with the heavy metal transport modeling systems adopted by other studies (Falconer & Lin 2003; Kashefipour & Roshanfekr 2012) which are not OOD, and either require large amounts of observation data and/or only consider a single element, the GeoCIRC model allows the implementation of multiple elements with minimal input data and using abstractions of many of the more complex details. In addition, the heavy metals are treated simultaneously but independently, because the OOD has sufficient flexibility to handle multiple transportable materials as instances. As a result, the application of the GeoCIRC model with systematic simulation cases in the present study can help to understand the riverine transport of multiple heavy metals which originate from industrial activities, and provides a comprehensive description of the river contamination with a minimum amount of observation data. It can thus reduce the number of samples which should be analyzed in situ and provide results in a short time. The model can enhance and support environmental monitoring strategies for the contamination of rivers with heavy metals caused by industrial activities in developing countries.

CONCLUSIONS

In this paper, detailed results were given for the numerical assessment of the transport of multiple heavy metals in the Harrach River basin using a hydrological model based on OOD. The model was successfully applied to estimate the Pb, Hg, Cr, and Zn concentrations in stream water and sediment in Harrach River in Algeria, whereas the estimation of Hg concentration was of lesser quality. Methodology for the assessment of data quality control and the improvement of monitoring practices was successfully proposed by using the hydrological model with systematic scenario tests. The GeoCIRC model was able to simulate multiple elements with a reasonable accuracy. The application of OOD improved the model's flexibility when many unknown point sources were present, and supported the inclusion of multiple heavy metals in the simulation, which in turn increases the model's effectiveness for the monitoring of heavy metal contamination in rivers.

ACKNOWLEDGEMENTS

The authors would like to thank the Observatoire National de l'Environnement et de Developpement Durable and the Office National Meteorologie, Algeria, for providing the data and supporting this research project. The sponsors had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.

REFERENCES

REFERENCES
Ani
E.-C.
,
Wallis
S.
,
Kraslawski
A.
&
Agachi
P. S.
2009
Development, calibration and evaluation of two mathematical models for pollutant transport in a small river
.
Environmental Modelling & Software
24
(
10
),
1139
1152
.
doi: 10.1016/j.envsoft
.
Bartram
J.
&
Ballance
R.
1996
Water Quality Monitoring – A Practical Guide to the Design and Implementation of Freshwater Quality Studies and Monitoring Programs
, 1st edn.
Chapman Hall
,
London
,
UK
,
ch. 2
.
Bouragba
S.
,
Komai
K.
&
Nakayama
K.
2017
Heavy metals transport simulation by physically based distributed modeling approach (case study: Harrach River, Algeria)
.
Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering)
73
(
4
),
1159
1164
.
Duruibe
J. O.
,
Ogwuegbu
M. O. C.
&
Egwurugwu
J. N.
2007
Heavy metal pollution and human biotoxic effects
.
International Journal of Physical Sciences
2
(
5
),
112
118
.
Falconer
R. A.
&
Lin
B.
2003
Hydro-environmental modelling of riverine basins using dynamic rate and partitioning coefficients
.
International Journal of River Basin Management
1
(
1
),
81
89
.
doi: 10.1080/15715124.2003.9635194
.
Falconer
R. A.
,
Lin
B.
&
Kashefipour
S. M.
2005
Modelling water quality processes in estuaries
. In:
Computational Fluid Dynamics: Applications in Environmental Hydraulics
(P. D. Bates, S. N. Lane & R. I. Ferguson, eds),
John Wiley & Sons, Ltd
,
Chichester, UK
, pp.
305
328
.
doi: 10.1002/0470015195.ch12
.
Fanger
H.-U.
,
Kuhn
H.
,
Michaelis
W.
,
Müller
A.
&
Riethmüller
R.
1986
Investigation of material transport and load in tidal rivers
.
Water Science and Technology
18
(
4–5
),
101
110
.
https://doi.org/10.2166/wst.0185
.
Gao
Y.
,
Yuan
Y.
,
Wang
H.
,
Schmidt
A. R.
,
Wang
K.
&
Ye
L.
2017
Examining the effects of urban agglomeration polders on flood events in Qinhuai River basin, China with HEC-HMS model
.
Water Science and Technology
75
(
9
),
2130
2138
.
https://doi.org/10.2166/wst.023
.
Kachiashvili
K.
,
Gordeziani
D.
,
Lazarov
R.
&
Melikdzhanian
D.
2007
Modeling and simulation of pollutants transport in rivers
.
Applied Mathematical Modelling
31
(
7
),
1371
1396
.
doi: 10.1016/j.apm
.
Kang
K.-M.
,
Rim
C.-S.
&
Yoon
S.-E.
2012
Object-oriented prototype for tightly coupled GIS based hydrologic modeling
.
Journal of Korea Water Resources Association
45
(
6
),
597
606
.
doi: 10.3741/JKWRA
.
Kashefipour
S. M.
&
Roshanfekr
A.
2012
Numerical modelling of heavy metals transport processes in riverine basins
. In:
Numerical Modelling
(
Miidla
P.
, ed.),
InTech
,
Rijeka, Croatia
, pp.
91
114
.
doi: 10.5772/35931
.
Kobayashi
K.
,
Nakayama
K.
&
Sakaguchi
S.
2017
Investigation of applicability of a long wave equation for inundation analysis
.
Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering)
73
(
4, part 2
),
618
623
.
Louati
B.
2015
Study of Industrial Depollution of Harrach Watershed
.
Master's thesis
,
Faculty of Earth Sciences, Baji Mokhtar University
,
Annaba
,
Algeria
(in French with English abstract)
.
Nakayama
K.
&
Kakinuma
T.
2010
Internal waves in a two-layer system using fully nonlinear internal-wave equations
.
International Journal for Numerical Methods in Fluids
62
(
5
),
574
590
.
Nakayama
K.
&
Shintani
T.
2015
Development of object-oriented geophysical flow circulation model, GeoCIRC, with focus on infiltration process
.
Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering)
71
(
4
),
283
288
(in Japanese with English abstract)
.
Nakayama
K.
,
Horikawa
T.
&
Ejima
K.
2000
Numerical study on a shallow water model with lignosa from large-eddy simulation
.
Proceedings of Hydraulic Engineering
44
(
2
),
801
806
(in Japanese with English abstract)
.
Nakayama
K.
,
Kakinuma
T.
&
Tsuji
H.
2018
Oblique reflection of large internal solitary waves in a two-layer fluid
.
European Journal of Mechanics – B/Fluids
74
(
3–4
),
81
91
.
Pak
G.
,
Mallari
K. J. B.
,
Baek
J.
,
Kim
D.
,
Kim
H.
,
Jung
M.
,
Kim
Y.
&
Yoon
J.
2015
Modelling of suspended sediment in a weir reach using EFDC model
.
Water Science and Technology
73
(
7
),
1583
1590
.
https://doi.org/10.2166/wst.574
.
Palmer
M. D.
2001
Water Quality Modeling
.
World Bank Publications
,
Washington, DC, USA
,
ch. 1
, pp.
1
4
.
Roque
F.
,
Diaz
K.
,
Ancco
M.
,
Delgado
D.
&
Tejada
K.
2018
Biodepuration of domestic sewage, textile effluents and acid mine drainage using starch-based xerogel from recycled potato peels
.
Water Science and Technology
77
(
5
),
1250
1261
.
https://doi.org/10.2166/wst.008
.
Tukura, B. W.
2015
Heavy metals pollution of water and sediment in Mada River, Nigeria
.
Journal of Scientific Research & Reports
6
(
2
),
157
164
.
Wu
Y.
,
Falconer
R. A.
&
Lin
B.
2001
Hydro-environmental modelling of heavy metal fluxes in an estuary
. In:
Proceedings of 29th IAHR Congress, Beijing, China, September 2001
,
Tsinghua University Press
,
Beijing, China
, pp.
732
739
.
Yoshida
M.
,
Moali
M.
,
Houas
O.
,
Lakhdari
M.
,
Nechaoui
L.
,
Guerrida
D.
,
Chatal
A.
,
Oussalem
S.
,
Makour
F.
,
Khelifi
F.
&
Laleg
A.
2005
Environmental pollution in Oued El Harrach area, Alger – a preliminary report on mercury and heavy metals contaminations. In: Report of Workshop and Seminar Conjoint between Algeria and Japan in Quality Standards and Environmental Protection Efforts, Alger, MATET-ONEDD-JICA
, pp.
19
37
.
Zhang
J.
,
Du
C.
&
Feng
X.
2017
Research on a soft measurement model of sewage treatment based on a case-based reasoning approach
.
Water Science and Technology
76
(
12
),
3181
3189
.
https://doi.org/10.2166/wst.417
.
Zhao
G.
,
Ye
S.
,
Yuan
H.
,
Ding
X.
&
Wang
J.
2017
Surface sediment proprieties and heavy metal pollution assessment in the Pearl River Estuary, China
.
Environmental Science Pollution
24
(
3
),
2966
2979
.
doi.10.1007/s11356-016-8003-4
.