Water Research 164 (2019) 114911

Contents lists available at ScienceDirect

Water Research
journal homepage: www.elsevier.com/locate/watres

Behaviour of metals in an urban river and the pollution of estuarine

environment
Buddhi Wijesiri a, b, An Liu a, b, c, *, Beibei He b, Bo Yang a, c, Xu Zhao d, Godwin Ayoko b,
Ashantha Goonetilleke a, b
a
College of Chemistry and Environmental Engineering, Shenzhen University, Shenzhen, 518060, China
b
Science and Engineering Faculty, Queensland University of Technology (QUT), GPO Box 2434, Brisbane, Qld, 4001, Australia
c
Shenzhen Key Laboratory of Environmental Chemistry and Ecological Remediation, Shenzhen, 518060, China
d
Key Laboratory of Drinking Water Science and Technology, Research Centre for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, 100085,
China

a r t i c l e i n f o a b s t r a c t

Article history: The high pollutant loads discharged from cities pose risks to urban waterways, and in turn the estuarine
Received 11 January 2019 environments, making it challenging to improve urban liveability. Past studies on the behaviour of
Received in revised form pollutants in rivers have largely investigated their transport along the waterway, primarily focusing on
17 July 2019
the movement of water and sediment. However, the current approaches in pollutant transport modelling
Accepted 22 July 2019
provide limited insights into how pollutant transfer between water and sediment phases influences their
Available online 24 July 2019
transport from the upstream towards the estuarine environment. This research study firstly identified
typical patterns of metal loads along an urban river in a highly populated city in China. The outcomes
Keywords:
Estuarine environment
were then used to conceptualise metal transfer between water and sediment phases. It was noted that
Metals physico-chemical characteristics of water and sediments play a key role in metal transfer between the
Pollutant transfer two phases, and the dominant transfer path (sediment to water/water to sediment) is different between
Pollutant transport different metals, independent of their origin (crustal, anthropogenic or marine-related). Several sce-
Urban liveability narios were derived from the conceptualisation of metal behaviour. These in turn were then used to
Urban river pollution develop real-world scenarios of metal transport in rivers based on the field data. The conceptualisation of
metal behaviour confirmed that each metal is likely to have a dominant phase of transport (sediment/
water), which is influenced by the dominant transfer path of that metal between water and sediments.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction (estuary) of a large river. In fact, urban rivers as the primary source
of water ranging from consumptive uses to recreational activities,
The transition of human population from rural areas to cities play an important role in the liveability of cities (WB_DRC, 2013).
has been increasing over the past decades, and currently more However, the cities commonly discharge stormwater and treated
than half of global population is urban. According to the United wastewater that carry non-biodegradable pollutants such as
Nations, as of 2016, 512 cities around the world are inhabited by metals into the river, posing risks to both, human and ecosystem
over one million people, and this number will increase to 662 health (Islam et al., 2015; Landrigan et al., 2018; Li et al., 2014; Xu
cities by 2030. Further, 31 of these cities (the most in China e 6) et al., 2014).
had more than 10 million inhabitants in 2016 and this figure is It is also likely that river waters transport large amounts of
projected to increase to 41 cities by 2030 (still the most from pollutants to the estuarine environment. Further, there is pollutants
China e 8) (UNDESA, 2016,2018). Most of these highly populated transfer between water and sediments. As such, sediments will also
cities are geographically located in the proximity of the bay area transport pollutants to the estuarine environment (Ciszewski and
Grygar, 2016; Rügner et al., 2019; Xiao et al., 2013; Zhou and
Broodbank, 2014). Moreover, the degradation of the estuarine
environment associated with urban rivers can be substantial
* Corresponding author. College of Chemistry and Environmental Engineering,
because of the increase in pollutant loads over time (as the river
Shenzhen University, Shenzhen, 518060, China.
E-mail address: liuan@szu.edu.cn (A. Liu). continues to transport pollutants) and sea water intrusion into the

https://doi.org/10.1016/j.watres.2019.114911
0043-1354/© 2019 Elsevier Ltd. All rights reserved.
2 B. Wijesiri et al. / Water Research 164 (2019) 114911

river due to tidal effects (Brady et al., 2014a,b; Kennish, 2017). The 2. Materials and methods
degree of pollution of the estuarine environment is dependent on
the capacity of the river to transport pollutants through the water 2.1. Study area
column as well as sediments.
Although the flow dynamics of the water column and sedi- Water and sediment samples were collected from Shenzhen
ments have been well documented (Merritt et al., 2003; River located in Shenzhen, South China, where a large number of
Nakagawa, 2017; Van Rijn, 1993), the transfer of pollutants be- production facilities of major technology companies are located.
tween the water and sediment phases has received only limited The river, which is also the natural border between mainland of
attention. The modelling approaches used to replicate pollutant China and Hong Kong, spans over 37 km with a catchment area of
transport along rivers (Ani et al., 2009; Benedini and Tsakiris, 297 km2 (172 km2 in Shenzhen and 125 km2 in Hong Kong), and
2013; Markus et al., 2016; McBride and Rutherford, 1984; flows into Shenzhen Bay. The sampling was undertaken at 11 lo-
Peria
n
~ ez et al., 2013) are largely based on the principles of stream cations from the upstream to Shenzhen bay. These locations are
hydraulics. For example, analytical and numerical modelling of surrounded primarily by residential, commercial, industrial, agri-
pollutant transport by Ani et al. (2009) relied on advection- cultural and natural land uses as shown in Fig. 1.
dispersion equations to predict pollutant mass transport gov-
erned by water flow characteristics (e.g. flow velocity) and chan- 2.2. Sampling and laboratory analysis
nel characteristics (e.g. wetted area of the channel). However, they
did not note the possibility of pollutants transfer with sediments. At each location, three water samples and three sediment
Markus et al. (2016) modelled the transport of engineered metallic samples were collected. Sample collection and handling procedures
nanoparticles along the water column of a fast flowing river. are discussed in detail in the Supplementary Information. The basic
Although they accounted for the transfer of nanoparticles be- water quality parameters (pH, salinity, dissolved oxygen and
tween river water and sediments, little attention was given to the turbidity) were tested in-situ, and the data is provided in Table S1 in
transport of nanoparticles via the sediment mass or transfer of the Supplementary Information. The salinity of water was found to
nanoparticles from sediments to water. These studies have typi- increase from upstream to downstream in the range of
cally accounted for pollutant transport via either water or sedi- 0.20e8.90‰. This indicates the impact of tidal effects in the estu-
ments, separately. Some commercially available models, for arine area. Dissolved oxygen and turbidity in water ranged
example, MIKE 11, allows combining both, water and sediment 0.5e6.0 mg/L and 36.7e126.6 NTU, respectively. These results
phases, but the pollutant mass balance calculations do not imply a relatively high variation in water quality along the river due
adequately account for the interactions between pollutants in the to changes in the surrounding environment at each sampling site.
two phases (Benedini and Tsakiris, 2013). The study by Huang The samples collected were transported to the laboratory for
et al. (2015), created models by incorporating pollutant in- further analysis. The sediment samples were analysed for particle
teractions between water and sediment phases in environmental size distribution (PSD) and mineralogical composition, and both,
systems such as large reservoirs. However, these models may not water and sediment samples were analysed for thirteen metals (Al,
be appropriate to investigate the behaviour of pollutants in urban Ti, Mn, Fe, Ni, Cu, Zn, Ba, Pb, Na, Mg, K and Ca) and organic carbon
rivers due to their dynamic nature (continuous flow, currents and (OC) content. The metals, which are ubiquitous in urban areas were
turbulence) compared to reservoirs. selected, such that they represent crustal metals (soil-sourced): Al,
Pollutants potentially exhibit rather complex behaviours within Ti, Mn and Fe; anthropogenic metals (sourced from human activ-
the water and sediment phases such as adsorption and desorption. ities): Ni, Cu, Zn, Ba and Pb; and marine-related metals (sourced
Past research studies show that the interactions between terrestrial from marine/estuarine environments): Na, Mg, K and Ca. The
particulate solids (e.g. soils and road dust) and ionic/molecular testing methods and quality control and quality assurance pro-
forms of pollutants such as metals are influenced by a range of cedures are discussed in the Supplementary Information.
physico-chemical characteristics of water (pH and dissolved
organic matter) as well as sediment particles (organic matter 2.3. Data analysis
content, mineralogical composition and particle size) (Bradl, 2004;
Gunawardana, 2011; Peng et al., 2018; Sposito, 2008). Therefore, The data analyses consisted of three stages: (1) evaluation of the
there is a need for a major step forward in knowledge of changes to variations in metal loads and physico-chemical characteristics of
pollutant loads during transport in the aquatic environment in water and sediments along the river (from upstream to the estua-
order to understand how terrestrial pollution leads to degradation rine environment); (2) identification of the relationships between
of rivers, and in turn marine environments such as estuaries. water and sediment characteristics and metal loads along the river
This research study hypothesised that changes to the physico- and how these relationships are influenced by changes to these
chemical characteristics of water and sediments along the river characteristics; and (3) development of a conceptual model
influence metal transfer between water and sediment phases, and describing the role of interactions between water and sediment
in turn the transport of metals into the estuarine environment. As phases in the transport of metals along the river.
such, the investigation aimed: (a) to understand the relationships
between water and sediment characteristics and metal loads along 3. Results and discussion
the river and how these relationships are influenced by changes to
these characteristics; and (b) to conceptualise pollutant in- 3.1. Variation in metal loads from upstream to estuarine
teractions between water and sediment phases for modelling environment
pollutant transport in urban rivers. The study lays the foundation
for understanding the mechanisms of pollutant transfer between In the context of environmental pollution, the origin of pollut-
water and sediment phases for modelling of pollutant transport ants and their transport pathways are important for effective
along urban rivers. This is a critical need for the design of measures pollution mitigation. Understanding of these factors helps to
to improve urban water quality and to protect the aquatic determine whether to adopt a source-control or transport-control
ecosystems. approach for the design of effective pollution mitigation strate-
gies. Accordingly, in this section, the identification of the patterns of
 B. Wijesiri et al. / Water Research 164 (2019) 114911 3

Fig. 1. Aerial view of sampling locations (S1 / S11) and their surrounding environment along the Shenzhen River. Note: photos do not show exact location of each sampling point,
rather they show typical characteristics of the surrounding area of sampling points within the circled region.

variations in crustal, anthropogenic and marine-related metal primarily through adsorption and desorption processes. These
loads, and how these three types of metals behave along the river processes are underpinned by three mechanisms: chemisorption,
were analysed. ion exchange and precipitation. Chemisorption occurs when the
Fig. 2 shows total metal loads in water from upstream to the pollutants in the solution (water) forms strong covalent bonds with
estuarine environment. Crustal metals (Al, Ti, Mn and Fe) show a chemically reactive surface functional groups on the surface of solid
decreasing pattern (particularly from sampling location 3 e S3 in particles (sediments) (Bradl, 2004). Ion exchange results in the
Fig. 1) and marine-related metals (Na, Mg, K and Ca) show an replacement of charged ions on solid particles in the solution
increasing pattern (particularly from sampling location 5 e S5 in through relatively weak electrostatic bonds (McBride, 1994). In
Fig. 1) and both patterns are generally consistent for all metals in essence, the above two mechanisms are largely influenced by the
each group. However, anthropogenic metals (Ni, Cu, Zn, Ba and Pb) physico-chemical characteristics of the solid particles (e.g. particle
do not exhibit a consistent pattern. size, specific surface area, mineralogy and surface coatings), while
Fig. 3 shows the total metal loads in sediments at each sampling pollutants in the solution can precipitate on the surface of solid
location. Compared to the patterns in water, metals in sediments particles depending on the pH and concentration of pollutants
exhibit more distinct patterns at different longitudinal sections (Gunawardana, 2011).
along the river. Crustal metal load generally increases from S1 to S5;
then decreases towards S8; and again increases in the proximity of
3.2.1. Implications of the change in particle size from upstream to
the estuarine environment (S9 e S11). Unlike in water (Fig. 2),
downstream
anthropogenic metals in sediments show a consistent increase
While particle size plays the key role in metal adsorption to
from S1 to S4; decrease towards S7; and again show an increase
particulate solids, other influential physico-chemical characteris-
near the estuarine environment. Further, marine related metal load
tics (e.g. particle surface area and surface coatings) are closely
generally increases from upstream to downstream, except for Ca.
related to particle size (Gunawardana et al., 2012, 2014). As it is an
Accordingly, it is evident that even the same group of metals
important parameter, particle size was separately investigated
exhibit rather inconsistent patterns between water and sediment
before other parameters (those discussed in Section 3.2.2).
phases. This, on one hand, underlines the complexity of the
Fig. 4 shows both, volume-based and surface area-based PSD of
behaviour of metals not only between groups that are based on the
sediments collected at the sampling locations along the river. The
origin of metals, but also between individual metals within the
PSDs are consistent with the fact that fine particles have relatively
same group. On the other hand, there is a need for an in-depth
larger surface area, increasing the potential for adsorbing pollut-
analysis of the influence of the physico-chemical characteristics
ants due to the presence of more surface functional groups (Wijesiri
of water and sediments on the variations of metal loads identified.
et al., 2016). However, the most important observation from Fig. 4 is
that particle size decreases (in turn the particle surface area in-
3.2. Interactions between water and sediment phases in metal creases) from upstream towards the estuarine environment
transport (compare PSDs for S1 e S5, PSDs for S6 e S8 and PSDs for S9 e S11).
This is due to the transport of fine sediment particles along the river
As water and sediment phases are in contact, the interactions of as these are more likely to be suspended in water and subject to
ionic and molecular forms of pollutants in each phase would occur currents and turbulence than larger particles. Moreover, as the river
4 B. Wijesiri et al. / Water Research 164 (2019) 114911

Fig. 2. Total metal load in water from upstream to downstream of Shenzhen River. Note 1: crustal metals e Al, Ti, Mn, Fe; anthropogenic metals e Ni, Cu, Zn, Ba; marine-related
metals e Na, Mg, K, Ca. Note 2: metal loads shown as zero were <0.001 ppb as originally detected. Note 3: although Na load at sampling points 9, 10 and 11 are given as the same load
as at sampling point 8, the detection for the samples collected at former three points were ‘out of range’, thus possibly higher than the preceding locations due to the proximity to
the estuarine environment.

approaches the estuarine environment, its bed slope tends to increase in fine particles. Accordingly, it is likely that fine sediment
reduce, and accordingly, the flow velocity tends to reduce and particles would preferentially adsorb crustal and marine related
larger particles will gradually settle on the bed. However, when metals rather than anthropogenic metals.
there is a flood (high flow e high velocity), the bed will be flushed These inconsistencies could be attributed to the contribution of
out. anthropogenic activities along the river and tributaries, which is
Due to the above phenomenon, the metal load associated with not related to the physico-chemical characteristics of water and
sediments in the proximity of the estuarine environment could be sediments as well as metal transport via water and sediments. As
higher than that in upstream locations as the finer particles have a evident from Fig. 1, the estuarine area surrounded by the natural
relatively greater capacity to adsorb cations. In fact, this occurrence environment do not support many anthropogenic activities
is valid for crustal metals and marine related metals (except for Ca) compared to the upstream areas.
in sediments (Fig. 3) which show an increasing trend from S1 to S11.
For crustal metals, this is further supported by the decreasing metal
3.2.2. Implications of the change in mineralogy, organic carbon and
load in water (Fig. 2), which is attributed to the adsorption of metals
pH from upstream to downstream
by fine particles in sediments that are abundantly available in the
Mineralogy is important as it determines the types of surface
estuarine environment. Although marine related metals load in
functional groups protruding from the surface of sediment parti-
water increases (Fig. 2), it could be potentially due to sea water
cles. This influences what species of metals are likely to be adsor-
intrusion (Naþ, Mg2þ, Kþ and Ca2þ in the form of salts) from tidal
bed (Jayarathne et al., 2018; Sutherland et al., 2012). On the other
effects. Moreover, anthropogenic metals in sediments and water do
hand, organic carbon content in sediments indicates the presence
not support the increase in metal loads in sediments with the
of organic coatings around particles which also provide specific
 B. Wijesiri et al. / Water Research 164 (2019) 114911 5

Fig. 3. Total metal load in sediments from upstream to downstream of Shenzhen River. Note: crustal metals e Al, Ti, Mn, Fe; anthropogenic metals e Ni, Cu, Zn, Ba; marine-related
metals e Na, Mg, K, Ca.

surface functional groups. Organic carbon in water would play a key ‘curse of dimensionality’ (Wijesiri et al., 2018). These dependencies
role in metal adsorption as it influences organic complexation of are firstly identified using expert opinion/state-of-the-art knowl-
metals (Murakami et al., 2008). Further, pH of water influences the edge of the process/system being modelled. The identified re-
surface electrochemical properties of sediment particles (i.e. posi- lationships are then visualised as a graphical structure, which is
tive and negative charge distribution), and in turn the adsorption then fitted with the observed (measured) data. This will result in
and desorption of metals from sediments (Wijesiri et al., 2016). estimations of model parameters, which indicate the strength of
Given that mineralogy, organic carbon content and pH are fac- the relationships between variables, in terms of conditional prob-
tors that influence the interactions between metals and surface abilities (for discrete variables) and conditional regression co-
functional groups on sediment particles, it was necessary to un- efficients (for continuous variables). Chen and Pollino (2012)
dertake a quantitative analysis of the data in order to better un- provide guidelines for good practice in BNs modelling, while
derstand metal transfer between water and sediment phases. Aguilera et al. (2011) and Uusitalo (2007) provide in-depth dis-
Therefore, a Bayesian Network (BN) model was developed, and its cussions on the applications of BNs in environmental modelling.
graphical structure is shown in Fig. 5. The Markov Property of BNs The BN shown in Fig. 5 describes how mineralogy of sediments,
accounts for the conditional dependencies among a given set of organic carbon in water and sediments and pH of water can influ-
random variables, such that a particular variable is dependent only ence the interactions between metals in water and sediment pha-
on its immediate parent variables (Korb and Nicholson, 2010). This ses. It is important to note that: (1) this BN is expected to describe
enables effectively dealing with small data sets by avoiding the only the metal transfer between the two phases under the
6 B. Wijesiri et al. / Water Research 164 (2019) 114911

Fig. 4. Particle size distributions (volume-based and surface area-based) of river sediments from upstream to downstream of Shenzhen River.

Fig. 5. Graphical structure of the Bayesian Network (BN) models of metal transfer between water and sediment phases. Note: conditional density refers to the probability density
function of a particular variable, given its immediate parent variables.
 B. Wijesiri et al. / Water Research 164 (2019) 114911 7

Table 1
Outcomes of the Bayesian Network (BN) model (Model 1) for metal transfer from water to sediments.
a
Origin Metal in sediments Estimated conditional regression coefficients (conditional Gaussian distribution, log Influence of Mw on Ms (%) RMSE
transformed data)

Intercept Mw OC_s Alb Amp Kao Mic Mus Qtz

Crustal Al 4.437 0.00504 0.391 0.00667 0.143 0.262 0.269 0.0480 0.456 0.319 0.00693
Ti 3.250 0.00141 0.0188 0.0244 0.107 0.221 0.205 0.0134 0.00849 0.236 0.0121
Mn 7.625 0.221 0.0494 0.480 1.0713 0.513 0.434 0.904 1.976 3.913 0.0309
Fe 4.446 0.000895 0.106 0.0528 0.0452 0.0461 0.0751 0.0338 0.310 0.134 0.0101
Anthropogenic Ni 13.931 0.0137 3.650 1.615 1.246 2.539 0.798 1.037 0.255 0.123 0.200
Cu 11.487 0.226 2.901 0.617 0.874 1.947 2.351 0.982 1.419 1.999 0.135
Zn 3.515 0.0163 0.573 0.660 1.478 0.272 0.0651 0.0707 0.562 0.441 0.00624
Ba 3.265 0.00483 0.328 0.0237 0.129 0.591 0.128 0.00146 0.134 0.360 0.0101
Pb 2.883 0.000287 0.0775 0.279 0.116 0.126 0.907 0.0496 0.0432 0.0180 0.00833
Marine-related Na 2.200 0.279 0.0746 0.828 0.697 0.320 1.420 0.0515 0.477 6.728 0.0158
Mg 3.839 0.0878 0.169 0.319 0.484 0.252 0.596 0.0302 0.0333 4.454 0.00795
K 5.682 0.00625 0.117 0.0428 0.292 0.0415 0.251 0.0248 0.446 0.512 0.00199
Ca 8.928 0.115 1.231 0.269 0.641 1.363 1.0571 0.0174 0.572 2.184 0.0362

Ms e metal load in sediments; Mw e metal load in water; OC_s e organic carbon content in sediments; Alb e albite; Amp e amorphous; Kao e kaolinite; Mic e microcline;
Mus e muscovite; Qtz e quartz; RMSE e root mean squared error.
a
Calculated by dividing the magnitude of regression coefficient of Ms by the sum of magnitudes of all regression coefficients.

influence of factors that stimulate or restrain adsorption and satisfactorily for all metals. However, the other model (i.e. Model 2)
desorption by sediment particles; (2) for a given metal, this BN performs well only for a limited number of metals (i.e. Mn, Cu, and
model does not account for the potential influence on a particular marine-related metals). This implies that desorption of most crustal
metal from other metals present in the water/sediment matrix (e.g. and anthropogenic metals from sediments to water (Model 2) can
cations competing for functional groups on sediment particles). be influenced by other factors other than pH, mineralogy and
Therefore, the outcomes of the BN model are conditional on the organic carbon content, and also the potential competition be-
model structure shown in Fig. 5. tween cations (which was not accounted for in this study). As the
The proposed model was fitted with observed data (metals in new knowledge on these influential factors becomes available, the
water e mg/L; metals in sediments - mg/g; organic carbon in water structure of the BN model can be improved by adding new pre-
e mg/L; organic carbon in sediments e g/kg; sediment mineralogy dictor variables, resulting in better model performance. Accord-
e %; pH in water) using bnlearn package in the R statistical ingly, the following discussion is based on the metals where the
computing platform (Scutari, 2009). Data on the mineralogical model performs well.
composition of sediments, organic carbon content in sediments As evident from Tables 1 and 2, metals show both, positive and
and water and pH in water are given in Table S1 in the Supple- negative (inverse) relationships between the total metal loads in
mentary Information. The model parameters were estimated for water and sediment phases. The positive relationships mean that
two metal transfer paths (see Table 1: from water to sediments and when the metal load in one phase increases, the metal load in the
see Table 2: from sediments to water). other phase would also increase. On the other hand, the inverse
The more likely path of metal transfer is identified by comparing relationships mean that when metal load in one phase increases,
the relative magnitude of model parameters corresponding to each the metal load in the other phase is expected to decrease. In fact,
path (see % influence column in Tables 1 and 2). It was evident from when considering metal transfer from one phase to another,
the root mean squared error (RMSE) calculated (Tables 1 and 2), obviously those metals with inverse relationships are more likely to
observed vs predicted plots and residual plots (Figs. S2 e S5 in undergo adsorption/desorption. However, the metals with positive
Supplementary Information), that the model relating to metal relationships could also be transferred between water and sedi-
transfer from water to sediments (i.e. Model 1) performs ment phases, while they are more likely to be transported via water

Table 2
Outcomes of the Bayesian Network (BN) model (Model 2) for metal transfer from sediments to water.
a
Origin Metal in water Estimated conditional regression coefficients (conditional Influence of Ms on Mw (%) RMSE
Gaussian distribution, log transformed data)

Intercept Ms OC_w pH

Crustal Al 124.332 13.679 7.335 220.0903 5.673 3.186

Ti 35.195 10.914 3.496 82.734 11.235 1.924
Mn 0.991 0.631 0.0283 0.341 63.0820 0.198
Fe 21.291 5.510 5.728 49.511 9.0701 3.218
Anthropogenic Ni 46.633 0.141 3.523 62.880 0.212 3.500
Cu 28.625 0.163 0.151 36.834 0.439 0.691
Zn 25.861 7.951 2.155 60.886 11.200 3.814
Ba 183.185 15.0760 0.526 269.159 5.294 2.329
Pb 133.358 8.665 1.185 172.224 4.759 3.453
Marine-related Na 11.0623 2.848 0.334 22.511 11.085 0.144
Mg 43.609 3.142 1.0205 63.485 4.645 0.355
K 29.542 3.942 0.0937 53.798 6.816 0.294
Ca 34.531 0.398 0.0211 41.0984 0.959 0.342

Ms e metal load in sediments; Mw e metal load in water; OC_w e organic carbon content in water; RMSE e root mean squared error.
a
Calculated by dividing the magnitude of regression coefficient of Ms by the sum of magnitudes of all regression coefficients.
8 B. Wijesiri et al. / Water Research 164 (2019) 114911

Table 3 divided into longitudinal sections based on major locations where

Dominant transfer path and dominant transport phase of metals along the Shenzhen the pollutants are discharged via anthropogenic and natural pro-
River (metals for which the Bayesian Network models had performed well).
cesses. Thus, at a given section (Si) and at a given time (ti), the load
Origin Metal Dominant transfer path Dominant transport phase of a particular pollutant in water (M(w)i) would be a function of the
Crustal Mn Sediments to water Water pollutant load discharged (Ii), load transported in via water from
Anthropogenic Cu Water to sediments Sediments the preceding section (M(wt)i-1), load transferred from sediment at
Marine-related Na Sediments to water Sediments the current section due to adsorption-desorption processes (M(str)i),
Mg Sediments to water Sediments
and load transported out via water towards the next section (M(wt)
K Sediments to water Sediments
Ca Water to sediments Sediments iþ1). Similarly, the load of the same pollutant in sediments (M(s)i)
would be a function of the load transported in via sediment from
the preceding section (M(st)i-1), load transferred from water at the
current section due to adsorption-desorption processes (M(wtr)i),
or sediments. These complex behaviours are discussed in detail
and load transported out via sediment towards the next section
with examples, below.
(M(st)iþ1). Accordingly, Equations (1)e(3) define the load of
As such, more Mn (a crustal metal) is transferred from sedi-
pollutant in water, sediment and total pollutant load at a given
ments to water (63.082% influence e Table 2) than from water to
section. Based on these equations, hypothetical scenarios of
sediments (3.913% influence e Table 1). Thus, the more dominant
pollutant transport along the river can be developed, facilitating
transfer path of Mn is sediments to water, which is also supported
the understanding of pollutant behaviour.
by the inverse relationships between Mn in sediments and water
(conditional regression coefficient 0.631 e Table 2). The inverse
relationship further implies that at a given location, the Mn load 3.3.1. General mass balance
desorbed (transferred) from sediments to water is greater than Mn Pollutant load in water at section Si:
load transported via sediments from upstream.    
Similarly, for K (a marine-related metal), the transfer from MðwÞi ¼ Ii þ MðwtÞi1  MðwtÞiþ1 þ MðstrÞi  MðwtrÞi (1)
sediments to water (6.816% influence e Table 2) is dominant over
the transfer from water to sediments (0.512% influence e Table 1). Pollutant load in sediments at section Si:
However, unlike Mn, K loads in sediments and water are positively    
related when they are transferred between phases (conditional MðsÞi ¼ MðstÞi1  MðstÞiþ1 þ MðwtrÞi  MðstrÞi (2)
regression coefficient 3.942 e Table 2). The implication of this
observation is that at a given location, K in sediments is desorbed to Total pollutant load at section Si:
water. However, the K load desorbed is less than the K load trans-
ported via sediments from upstream. MðwÞi þ MðsÞi ¼ Mi
Contrary to Mn and K behaviour, Cu (an anthropogenic metal)
   
¼ Ii þ MðwtÞi1 þ MðstÞi1  MðwtÞiþ1 þ MðstÞiþ1
exhibits a different behaviour. More Cu in water is adsorbed by
sediments (1.999% influence e Table 1), than what is already (3)
attached to sediments being desorbed into water (0.439% influence
e Table 2). This is further supported by the inverse relationship
between Cu loads in water and sediment phases (conditional
regression coefficient 0.226 e Table 1). This means that Cu load 3.4. Evaluation of hypothetical scenarios of pollutant transport
transferred from water to sediments is larger than Cu load trans-
ported via water from upstream. As outlined below, the hypothetical scenario 1 reduces Equa-
In conclusion, sediment phase plays the key role in K and Cu tions (1) and (2) to Equations (4) and (5), which leads to Case 1 and
transport and more Mn is likely to be transported via water to the Case 2.
estuarine environment. Similarly, the dominant transfer path (i.e.
sediments to water/water to sediments) and dominant transport
3.4.1. Hypothetical scenario 1
phase (i.e. sediments or water) for other metals can be determined
Input pollutant load (I) and net pollutant load transferred be-
using the BN outcomes given in Tables 1 and 2, and are summarised
tween water and sediment phases j(Mstr e Mwtr)j do not vary at
in Table 3, below.
different sections. This implies that physico-chemical properties of
water and sediments, which influence pollutant transfer between
3.3. Conceptualisation of pollutant interactions between water and the two phases, remain unchanged along the river. Thus, Equations
sediment phases (1) and (2) can be written as:
 
As discussed in the Introduction, current modelling approaches MðwÞi ¼ MðwtÞi1  MðwtÞiþ1 þ C1 (4)
to pollutant transport in river systems rely on the movement of
water and sediment and does not account for the changes to  
pollutant load resulting from the interactions between the two MðsÞi ¼ MðstÞi1  MðstÞiþ1 þ C2 (5)
phases. Such changes are typically not influenced by the water and
sediment movements, but rather would be determined by the Where, C1 and C2 are constants.
physico-chemical characteristics of these media as discussed in
Case 1. inward transport is greater than outward transport of
Section 3.2. Accordingly, this section conceptualises the impact of
pollutants
changes to pollutant loads between water and sediment phases on
i.e. M(wt)i-1 > M(wt)iþ1 and M(st)i-1 > M(st)iþ1
the overall pollutant transport along the river towards the estua-
rine environment. Case 2. inward transport is less than outward transport of
The conceptual model, which was developed based on mass pollutants
balance principles, is shown in Fig. 6. Accordingly, the river is i.e. M(wt)i-1 < M(wt)iþ1 and M(st)i-1 < M(st)iþ1
 B. Wijesiri et al. / Water Research 164 (2019) 114911 9

Fig. 6. Conceptual model of pollutant transport via river water and sediments.

According to Case 1, the load of pollutants transported decreases Thus, Equations (1) and (2) remains the same. Therefore, Case 1
along the river. Thus, compared to pollutant loads found in the and Case 2 in scenario 1 can be elaborated by incorporating the
upstream, largely reduced pollutant loads can be released into the possibilities of metal transfer between water and sediment phases.
estuarine environment. In fact, crustal metals in water (Fig. 2) and
Case 1.1. inward transport is greater than outward transport of
Ca in sediments (Fig. 3) show similar patterns. In order for this to be
pollutants, given that metal load transferred from water is greater
a real-world case, there should be factor/s that would constrain the
than the metal load transferred from sediments.
movement of ionic and molecular forms of pollutants together with
i.e. M(wt)i-1 > M(wt)iþ1 and M(st)i-1 > M(st)iþ1
water or sediment mass. However, hypothetical scenario 1 is
M(wtr)i > M(str)i
inadequate to investigate the causation for such decrease in metal
discharge into the estuarine environment. This is because of the Case 1.2. inward transport is greater than outward transport of
assumption that pollutant input and transfer between water and pollutants, given that metal load transferred from water is less than
sediment phases do not vary at different sections. On the other the metal load transferred from sediments.
hand, Case 2 describes pollutant behaviour, such that the pollutant i.e. M(wt)i-1 > M(wt)iþ1 and M(st)i-1 > M(st)iþ1
load can increase from upstream to downstream. This is more likely M(wtr)i < M(str)i
than Case 1, because pollutants would move as the water and
In Case 1.1 and Case 1.2, pollutant load transported is expected
sediment masses move along the river. In fact, Case 2 is similar to
to decrease from upstream to downstream. However, Case 1.1 will
most modelling approaches undertaken in past studies which have
be likely only for pollutant transport via water as these pollutants
not accounted for pollutant transfer between water and sediment
are continuously adsorbed by sediments (e.g. Cu, see discussion in
phases. This will be explored in hypothetical scenario 2 outlined
Section 3.2.2). Case 1.2 will be likely only for pollutant transport via
below.
sediments as these pollutants attached to sediment particles are
continuously desorbed into water (e.g. Mn and K, see discussion in
3.4.2. Hypothetical scenario 2 Section 3.2.2).
Input pollutant load (I) does not vary at different sections, but
net pollutant load transferred between water and sediment phases Case 2.1. inward transport is less than outward transport of pol-
j(Mstr e Mwtr)j varies. This implies that the physico-chemical lutants, given that metal load transferred from water is greater than
properties of water and sediments, which influence pollutant the metal load transferred from sediments.
transfer between the two phases are subject to change at different i.e. M(wt)i-1 < M(wt)iþ1 and M(st)i-1 < M(st)iþ1
locations along the river. M(wtr)i > M(str)i
10 B. Wijesiri et al. / Water Research 164 (2019) 114911

Case 2.2. inward transport is less than outward transport of pol- content in water and sediments, and pH, salinity, dissolved oxygen
lutants, given that metal load transferred from water is less than and turbidity in water from upstream to estuarine environment,
the metal load transferred from sediments. and the outcomes of observed vs predicted plots and residual plots
i.e. M(wt)i-1 < M(wt)iþ1 and M(st)i-1 < M(st)iþ1 in BN modelling are provided as Supplementary Information.
M(wtr)i < M(str)i Supplementary data to this article can be found online at
https://doi.org/10.1016/j.watres.2019.114911.
In Case 2.1 and Case 2.2, both water and sediment masses are
expected to transport an amount of pollutants that increases to-
wards the estuarine environment. Similar to Cases 1.1 and 1.2, these References
two cases will also be likely either in water or sediment phase, such
Aguilera, P.A., Fern
andez, A., Ferna
ndez, R., Rumí, R., Salmero
n, A., 2011. Bayesian
that Case 2.1 for sediments (e.g. Mn and K) and Case 2.2 for water
networks in environmental modelling. Environ. Model. Softw 26 (12),
(e.g. Cu). 1376e1388. https://doi.org/10.1016/j.envsoft.2011.06.004.
More importantly, it is possible to replicate real-world systems Ani, E.-C., Wallis, S., Kraslawski, A., Agachi, P.S., 2009. Development, calibration and
of pollutant transport by combining the cases discussed in the evaluation of two mathematical models for pollutant transport in a small river.
Environ. Model. Softw 24 (10), 1139e1152. https://doi.org/10.1016/j.envsoft.
hypothetical scenarios above. For example, Case 1.1 and Case 2.1 2009.03.008.
and Case 1.2 and Case 2.2 can be combined as each set has the Benedini, M., Tsakiris, G., 2013. Water Quality Modelling for Rivers and Streams.
same system of pollutant transfer between water and sediment Springer Science & Business Media.
Bradl, H.B., 2004. Adsorption of heavy metal ions on soils and soils constituents.
phases. Accordingly, in the system formed by Case 1.1 and Case 2.1, J. Colloid Interface Sci. 277 (1), 1e18.
sediment mass would play a key role in transporting pollutants Brady, J.P., Ayoko, G.A., Martens, W.N., Goonetilleke, A., 2014a. Enrichment, distri-
towards the estuarine environment as pollutant loads transported bution and sources of heavy metals in the sediments of Deception Bay,
Queensland, Australia. Mar. Pollut. Bull. 81 (1), 248e255. https://doi.org/10.
have a decreasing trend in water and an increasing trend in sedi- 1016/j.marpolbul.2014.01.031.
ments. Similarly, water mass contributes most to the pollutant Brady, J.P., Ayoko, G.A., Martens, W.N., Goonetilleke, A., 2014b. Temporal trends and
transport in the system formed by Case 1.2 and Case 2.2. bioavailability assessment of heavy metals in the sediments of Deception Bay,
Queensland, Australia. Mar. Pollut. Bull. 89 (1), 464e472. https://doi.org/10.
1016/j.marpolbul.2014.09.030.
4. Conclusions Chen, S.H., Pollino, C.A., 2012. Good practice in Bayesian network modelling. En-
viron. Model. Softw 37, 134e145.
Ciszewski, D., Grygar, T.M., 2016. A review of flood-related storage and remobili-
This research study investigated the key role of metal transfer zation of heavy metal pollutants in river systems. Water, Air, & Soil Pollution
between water and sediment phases in an urban river in relation to 227 (7), 239. https://doi.org/10.1007/s11270-016-2934-8.
metal transport from upstream to the estuarine environment, and Gunawardana, C., Egodawatta, P., Goonetilleke, A., 2014. Role of particle size and
composition in metal adsorption by solids deposited on urban road surfaces.
conceptualised the incorporation of metal transfer processes for Environ. Pollut. 184 (0), 44e53. https://doi.org/10.1016/j.envpol.2013.08.010.
metal transport modelling. The research outcomes indicated that Gunawardana, C., Goonetilleke, A., Egodawatta, P., Dawes, L., Kokot, S., 2012. Source
metals from different origins (crustal, anthropogenic and marine- characterisation of road dust based on chemical and mineralogical composition.
Chemosphere 87 (2), 163e170.
related) generally exhibit consistent patterns in water and sedi- Gunawardana, C.T.K., 2011. Influence of Physical and Chemical Properties of Solids
ment phases along the river. It was noted that different metals on Heavy Metal Adsorption. PhD. Queensland University of Technology.
exhibit different behaviour during the transfer between water and Huang, L., Fang, H., Reible, D., 2015. Mathematical model for interactions and
transport of phosphorus and sediment in the Three Gorges Reservoir. Water
sediment phases, and the physico-chemical characteristics of water
Res. 85, 393e403. https://doi.org/10.1016/j.watres.2015.08.049.
and sediments influence the adsorption and desorption processes Islam, M.S., Ahmed, M.K., Raknuzzaman, M., Habibullah -Al- Mamun, M.,
that underpin metal transfer between the two phases. The results Islam, M.K., 2015. Heavy metal pollution in surface water and sediment: a
preliminary assessment of an urban river in a developing country. Ecol. Indicat.
also show that each metal has a dominant transfer path (sediments
48, 282e291. https://doi.org/10.1016/j.ecolind.2014.08.016.
to water/water to sediments), and this dominant path determines Jayarathne, A., Egodawatta, P., Ayoko, G.A., Goonetilleke, A., 2018. Intrinsic and
the phase that transports the largest amount of metals to the extrinsic factors which influence metal adsorption to road dust. Sci. Total En-
estuarine environment. The conceptualisation of the behaviour of viron. 618 (Suppl. C), 236e242. https://doi.org/10.1016/j.scitotenv.2017.11.047.
Kennish, M.J., 2017. Practical Handbook of Estuarine and Marine Pollution. CRC
metals in water and sediment phases led to the creation of several press.
hypothetical scenarios which helps in understanding the complex Korb, K.B., Nicholson, A.E., 2010. Bayesian Artificial Intelligence, 2 ed. CRC press.
real-world scenarios observed from the field data. Accordingly, it Landrigan, P.J., Fuller, R., Acosta, N.J., Adeyi, O., Arnold, R., Balde
, A.B., Bertollini, R.,
Bose-O'Reilly, S., Boufford, J.I., Breysse, P.N., 2018. The Lancet Commission on
was revealed that the dominant transfer path (sediments to water/ pollution and health. The Lancet 391 (10119), 462e512.
water to sediments) and dominant transport phase (sediment/ Li, P., Qian, H., Howard, K.W., Wu, J., Lyu, X., 2014. Anthropogenic pollution and
water) are different for different metals, irrespective of their origin. variability of manganese in alluvial sediments of the Yellow River, Ningxia,
northwest China. Environ. Monit. Assess. 186 (3), 1385e1398.
These conceptualisations will contribute to enhancing the model- Markus, A.A., Parsons, J.R., Roex, E.W.M., de Voogt, P., Laane, R.W.P.M., 2016.
ling of pollutant transport in urban waterways and thereby assist in Modelling the transport of engineered metallic nanoparticles in the river Rhine.
informed decision-making in relation to river pollution control. Water Res. 91, 214e224. https://doi.org/10.1016/j.watres.2016.01.003.
McBride, G.B., Rutherford, J.C., 1984. Accurate modeling of river pollutant transport.
J. Environ. Eng. 110 (4), 808e827. https://doi.org/10.1061/(ASCE)0733-
Acknowledgement 9372(1984)110:4(808).
McBride, M.B., 1994. Environmental Chemistry of Soils. Oxford university press.
Merritt, W.S., Letcher, R.A., Jakeman, A.J., 2003. A review of erosion and sediment
We thank National Natural Science Foundation of China transport models. Environ. Model. Softw 18 (8), 761e799. https://doi.org/10.
(41601510) and China Postdoctoral Science Foundation 1016/S1364-8152(03)00078-1.
(2018M643194) for the financial support. Further, this work was Murakami, M., Nakajima, F., Furumai, H., 2008. The sorption of heavy metal species
by sediments in soakaways receiving urban road runoff. Chemosphere 70 (11),
enabled by use of the Central Analytical Research Facility hosted by 2099e2109.
the Institute for Future Environments at QUT. Nakagawa, H., 2017. Turbulence in Open Channel Flows. Routledge.
Peng, L., Liu, P., Feng, X., Wang, Z., Cheng, T., Liang, Y., Lin, Z., Shi, Z., 2018. Kinetics of
heavy metal adsorption and desorption in soil: developing a unified model
Appendix A. Supplementary data based on chemical speciation. Geochem. Cosmochim. Acta 224, 282e300.
https://doi.org/10.1016/j.gca.2018.01.014.
The details of sample collection and handling, details of labo- Peri
an~ ez, R., Casas-Ruíz, M., Bolívar, J.P., 2013. Tidal circulation, sediment and
pollutant transport in Ca
diz Bay (SW Spain): a modelling study. Ocean. Eng. 69,
ratory testing and quality control and quality assurance procedures, 60e69. https://doi.org/10.1016/j.oceaneng.2013.05.016.
The mineralogical composition of sediments, organic carbon Rügner, H., Schwientek, M., Mila cic, R., Zuliani, T., Vidmar, J., Paunovi
c, M.,
 B. Wijesiri et al. / Water Research 164 (2019) 114911 11

Laschou, S., Kalogianni, E., Skoulikidis, N.T., Diamantini, E., Majone, B., Bellin, A., WB_DRC, 2013. China 2030: Building a Modern, Harmonious, and Creative Society.
Chiogna, G., Martinez, E., Lo
pez de Alda, M., Díaz-Cruz, M.S., Grathwohl, P., 2019. World Bank and the Development Research Center of the State Council, P. R.
Particle bound pollutants in rivers: results from suspended sediment sampling China, Washington, DC. https://doi.org/10.1596/978-0-8213-9545-5.
in Globaqua River Basins. Sci. Total Environ. 647, 645e652. https://doi.org/10. Wijesiri, B., Deilami, K., McGree, J., Goonetilleke, A., 2018. Use of surrogate in-
1016/j.scitotenv.2018.08.027. dicators for the evaluation of potential health risks due to poor urban water
Scutari, M., 2009. Learning Bayesian Networks with the Bnlearn R Package arXiv quality: a Bayesian Network approach. Environ. Pollut. 233, 655e661. https://
preprint arXiv:0908.3817. doi.org/10.1016/j.envpol.2017.10.076.
Sposito, G., 2008. The Chemistry of Soils. Oxford University Press. Wijesiri, B., Egodawatta, P., McGree, J., Goonetilleke, A., 2016. Understanding the
Sutherland, R.A., Tack, F.M.G., Ziegler, A.D., 2012. Road-deposited sediments in an uncertainty associated with particle-bound pollutant build-up and wash-off: a
urban environment: a first look at sequentially extracted element loads in grain critical review. Water Res. 101, 582e596. https://doi.org/10.1016/j.watres.2016.
size fractions. J. Hazard Mater. 225e226, 54e62. https://doi.org/10.1016/j. 06.013.
jhazmat.2012.04.066. Xiao, R., Bai, J., Huang, L., Zhang, H., Cui, B., Liu, X., 2013. Distribution and pollution,
UNDESA, 2016. The World's Cities in 2016. The Population Division of the Depart- toxicity and risk assessment of heavy metals in sediments from urban and rural
ment of Economic and Social Affairs of the United Nations. rivers of the Pearl River delta in southern China. Ecotoxicology 22 (10),
UNDESA, 2018. World Urbanization Prospects 2018. The Population Division of the 1564e1575.
Department of Economic and Social Affairs of the United Nations. Retrieved 28/ Xu, Y., Sun, Q., Yi, L., Yin, X., Wang, A., Li, Y., Chen, J., 2014. The source of natural and
11/2018, from. https://population.un.org/wup/. anthropogenic heavy metals in the sediments of the Minjiang River Estuary (SE
Uusitalo, L., 2007. Advantages and challenges of Bayesian networks in environ- China): implications for historical pollution. Sci. Total Environ. 493, 729e736.
mental modelling. Ecol. Model. 203 (3e4), 312e318. http://doi.org/10.1016/j. https://doi.org/10.1016/j.scitotenv.2014.06.046.
ecolmodel.2006.11.033. Zhou, J., Broodbank, N., 2014. Sediment-water interactions of pharmaceutical resi-
Van Rijn, L.C., 1993. Principles of Sediment Transport in Rivers, Estuaries and Coastal dues in the river environment. Water Res. 48, 61e70. https://doi.org/10.1016/j.
Seas, vol. 1006. Aqua publications, Amsterdam. watres.2013.09.026.