Science of the Total Environment 472 (2014) 1070–1081

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

Assessment of water ecological carrying capacity under the two policies

in Tieling City on the basis of the integrated system dynamics model
Shuo Wang a, Ling Xu a, Fenglin Yang a,⁎, He Wang b
a
Key Laboratory of Industrial Ecology and Environmental Engineering (China Ministry of Education), School of Environmental Science and Technology, Dalian University of Technology,
Linggong Road 2, Dalian 116024, PR China
b
Liaoning Province Key Laboratory of Basin Pollution Control, Liaoning Academy of Environmental Sciences, Shenyang 110031, PR China

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Human activities are the producers and

pushers of social development in the
optimistic aspect.
• Social resource was considered into the
model.
• The new model assessed the available
policies (such as penalty and tax).
• Econometrics method is used to correct
the qualitative phase in building a system
dynamic model.

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

Article history: Considering the limitation of the traditional method to assess the ecological carrying capacity and the complexity
Received 19 May 2013 of the water ecological system, we used system dynamics, ANN, and CA–Markov to model a water ecological sys-
Received in revised form 22 November 2013 tem. The social component was modeled according to Granger causality test by system dynamics. The natural
Accepted 24 November 2013
component consists of the water resource and water environmental capacity, which were forecasted through
Available online 17 December 2013
the prediction of precipitation and change in land use cover. The interaction of the social component and the
Keywords:
natural component mainly reflected environmental policies, such as the imposition of an environmental
Carrying capacity fee and environmental tax based on their values. Simulation results showed the different assessments on
Water resource water ecological carrying capacity under the two policies. The population grew (2.9 million), and less pollution
System dynamics (86,632.37 t COD and 2854.5 t NH4N) was observed with the imposition of environmental tax compared with
LULC the imposition of an environmental fee (2.85 million population, 10,8381 t COD and 3543 t NH4N) at the same
Environmental tax GDP level of 585 billion CNY in 2030. According to the causality loop, we discussed the different states under
Tieling City the policies and the reasons that caused the differences in water ecological carrying capacity state. According
to game theory, we explained the limitation of the environmental fee policy on the basis of marginal benefit
and cost. The externality was cleared up by the environmental tax policy.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction

⁎ Corresponding author. Tel./fax: +86 411 84708083. The resource problem caused by an imbalance between supply
E-mail address: yfl6666@yeah.net (F. Yang). and demand has raised the interest of the international scientific

0048-9697/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.scitotenv.2013.11.115
 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081
community and was presented in 1987 in Brundtland's Report in the
context of sustainable development. In the extensive research of sus-
tainability, the carrying capacity is frequently presented in urban areas
(Venkatesan et al., 2011), industrial cluster (Mao et al., 2012), industrial
plant (Zhao et al., 2009) and the smaller area (Åstebøl et al., 2004) as
well as in temporal-scale (Haraldsson and Ólafsdóttir, 2006). The re-
source carrying capacity has been discussed since the Malthus' treatise
on human population (Yue et al., 2008). The formal concept of carrying
capacity originated from the animal ecology, which means the maxi-
mum number of the stock supported by the basic resource (i.e., food).
However, human carrying capacity, which is the maximum quantity of
individuals existing under several specific conditions, was proposed
by Parker and Burgis in the 1920s based on human ecology. In the late
1960s and early 1970s, the concepts of resource and environmental
carrying capacities were discussed in the looming limits of the resource
consumption and environment degradation resulting from excessive
human activities (Ehrlich and Holdren, 1971; Meadows et al., 1972).
Since the 1980s, the concept of carrying capacity has transformed into
the process of ecological carrying capacity. This concept is defined as
the maximum population size that is sustained by multiple hierarchical
levels such as resource consumption, environmental deterioration, and
human activities, accompanied by development with finite resources
(Kessler, 1994; del Monte-Luna et al., 2004).
Along with the concept of carrying capacity, the computation of
carrying capacity improved from the macro statistics to the micro sys-
tematic method. In the early macro statistics method as the Malthus
model, the rates of population growth and resource regeneration were
regarded as constants (Seidl and Tisdell, 1999), which is too simple
to describe the complex social–natural system. Fuzzy assessment and
hierarchical multi-criteria were preferred in the early carrying capacity
micro systematic model because of its availability and intelligence
(Awasthi and Chauhan, 2011; Gao and Hailu, 2012; Zerger et al.,
2011). However, the objectivity of the results is affected by the qualita-
tive criteria and individual opinion during assessment. The gap between
the two methods is narrowed by the development of the methodology.
The concept that a system can be divided into subsystems from the
micro systematic method was applied into the macro statistics method
to calculate separately the human activities and resource or environ-
mental load (Ni et al., 2012), although human activities and load have
no linear relationship. The micro systematic method adopted the quan-
tifiable advantage from the macro statistics method, such as in system
dynamics (Feng et al., 2008; Haraldsson and Ólafsdóttir, 2006; Sušnik
et al., 2012). However, the process of structuring model based on the
experts' experiences, discussion and so on is subjective. In the late
1980s and early 1990s, the calculation and assessment of carrying ca-
pacity have undergone considerable progress. The macro statistics
method absorbed the advantage of accounting, facilitating the calcula-
tion of the carrying capacity with uniform dimension. In the field, Rees
(2000) and Wackernagel and Rees (1999) presented the ecological foot-
print, and Odum (1996) proposed the emergy analysis. Although both
methods are preferred (Stöglehner, 2003; Cerutti et al., 2010; Geng
et al., 2010), the ecological footprint has difficulty in dealing with the
fluidity resource in an open system and the calculation of emergy trans-
formative ratio of the nonrenewable (e.g., ore) resource is based on
an uncertain geological process. The micro systematic method divides
the complex social–natural system into subsystems according to their
functionality. This process generated the drive force–state–response or
pressure–state–response (PRS) model framework. In the model frame-
work, the human activity subsystem is regarded as the driving force
or pressure of the whole system, whereas the resource and environ-
ment condition indicate the state of the system, and the management,
decision, and policy are the responses to the state (Chen et al., 2005).
However, the PSR approach does not accommodate the demands for
decision-making support, because it suggests linearity in the relation-
ships between human activities and the environment (Wolfslehner
and Vacik, 2008).
1071
The concept and method of carrying capacity had undergone great
development, but its wide and accurate application is hindered by
three factors. First, all methods and studies considered human activities
as the negative factor in the complex system. Second, the natural re-
source and environment are considered as unique linear limits of
the social–natural system. Finally, all methods are not sufficiently sup-
ported by the decision maker, that is, all decisions from the traditional
methods only focused on the tasks, such as population control, and
not on the techniques on how to perform tasks, which is the key to
decision making.
To remedy the three limitations of the traditional carrying capacity
method, we presented the new model ‘ECC = F(S, N, I), where ECC
stands for ecological carrying capacity.’ derived from the PSR frame-
work. In the new integrated model, the social component was regarded
as not only the consumer in the entire system but also the resource that
supports the social component itself. We used social resources, such as
GDP per capita, as the support of the society and economy instead of
just natural resources to correct the deviation caused by the sole role
that society plays in traditional assessments. Furthermore, the environ-
mental component was not used as the only basis of the social compo-
nent. In other words, natural resources did not directly support or
limit the society and economy, such as the famous Kiah Papua plateau
deer model. On the basis of the two aforementioned points, the new
model framework considered natural resources as a raw material
input into the production system, with their quantity and quality di-
rectly limiting the production system. Meanwhile, the benefit and cost
of production were affected by natural resources and thus influenced
the economy and society. Therefore, we took actual policies that eco-
nomically affected productive activities and placed them into the new
model to adjust the interaction between society and nature. In tradi-
tional models, the adjustable interaction is considered an external
variable and is not reflective of an endogenous relationship between
society and nature.
2. Methodology
Fig. 1 illustrates that this study was based on reality. We observed a
real and complex integrated ecological system and defined the subsys-
tems that involve environmental and human activities, as well as their
interactions with each other. Then, we abstracted reality in our new
model. Human activities included the socio-economic state. The Grang-
er causality test was then used to analyze the relationships in the social
system and prove the hypothesis about the aforementioned relation-
ship between the social component and the environmental component.
Then, we modeled the water ecological system according to the system
dynamics informed by the casual loop from the Granger causality
test. Water quantity and quality were in their natural states, but the
precipitation and land cover that affected the water resources and
the environment were uncontrollable. Therefore, the prediction of
precipitation and land cover refers to external variables that were
not affected by the social component. The forecasting of precipitation
and land cover was computed by RBF net and CA–Markov. Therefore,
the social component was modeled by system dynamics, and the envi-
ronmental component was modeled by RBF net and CA–Markov. Policy
and management were seen in these interactions. We modeled, tested,
and calculated water carrying capacity with the use of various tech-
niques. Finally, two different available policies were assessed and
analyzed according to the modeling result for improved planning and
management.
2.1. Social-economic component
In this component, system dynamics was primarily used to describe
and model the complex human activities. However, we had to adopt the
Granger causality test to analyze the quantitative causality, instead of
the traditional qualitative method first.
1072 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081
Fig. 1. The model technology route map.
2.1.1. Granger causality test
Granger causality test is a statistical hypothesis test for determining
whether a time series is useful for forecasting another (Granger, 1969),
in which x and y are stationary time series. To test the null hypothesis
that x does not Granger-cause y, the proper lagged values of y to be in-
cluded in a univariate autoregression of y were first determined using
the following equation:
yt ¼ a0 þ a1 yt−1 þ a2 yt−2 þ …am yt−m þ residualt :
Then, the autoregression was augmented by including the lagged
values of x, as follows:
yt ¼ a0 þ a1 yt−1 þ a2 yt−2 þ …am yt−m þ b0 þ b1 xt−1 þ b2 xt−2 þ …
þ bm xt−m þ residualt :
In this regression, all lagged values of x that are individually signifi-
cant according to their t-statistics were retained, provided that, collec-
tively, they add explanatory power to the regression according to an
F-test (in which the null hypothesis has no explanatory power jointly
added by the x's). In the notation of the above augmented regression,
p is the shortest and q is the longest, and lag lengths, which are the
lagged values of x, are significant. The null hypothesis that x does not
Granger-cause y was accepted if and only if no lagged values of x were
retained in the regression. Then, casual loops of system dynamics
were constructed according to the result of Granger causality test.
2.1.2. System dynamics
System dynamics is an approach in understanding the behavior of
complex systems over time. This approach deals with internal feedback
loops and time delays that affect the behavior of the entire system. The
interactions among variables of the model can be interpreted using
feedback loops in system dynamics (Roberts et al., 1983). Therefore,
system dynamics was applied in the simulation of the complex system
and impact of the policy, method, and environment (Carbonell et al.,
2000; Feng et al., 2008; Jin et al., 2009).
2.2. Natural component
Contrary to the variable complex process of human activities, the
processes of the water resource and pollution follow the physical prin-
ciple. Rainfall carries the water resource supply and non-point source
(NPS) contaminant via land use cover in a runoff process. The water re-
source and pollution from upstream are the other sources outside the
system. Thus, the model needs the tendency of precipitation and land
use cover change (LULC) to describe the natural component. The radial
basis function (RBF) neutral network was used to predict the precipita-
tion and the cellular automata (CA)–Markov chain model was used to
predict the LULC.
2.2.1. RBF neutral network
RBF networks have been successfully applied in solving dynamic
system problems to predict directly a behavior from the training
data (Han et al., 2011). Therefore, the RBF network is widely adopted
to predict the discrete temporal series because of its intelligible
and reliable features (Casasent and Chen, 2003; Giantomassi et al.,
2011).
An RBF network typically has three layers, namely, an input
layer, a hidden layer with a non-linear RBF activation function,
 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081
and a linear output layer. The output of the network is described as
follows:
N
KðxÞ ¼ ∑ i¼1 α i ρð∥x−ci ∥Þ
where N is the number of neurons in the hidden layer, ci is the center
vector for neuron i, and αi represents the weights of the linear output
neuron. In the basic form, all inputs are connected to each hidden
neuron. The norm is typically considered to be the Euclidean distance,
and the basis function is taken to be Gaussian as follows:
h i
2
ρð∥x−ci ∥Þ ¼ exp −β∥x−ci ∥
where β is the weight of the hidden layer. An RBF network with
sufficient hidden neurons can approximate any continuous function
with arbitrary precision.
2.2.2. CA–Markov model
The CA–Markov model is a combination of CA and Markov chain
land cover prediction method that combines spatial contiguity and
knowledge of the likely spatial distribution of transitions of Markov
chain analysis (Sang et al., 2011). It is a compounded method based
on the different characters and advantages in LULC research. The
Markov chain is focused on the quantity of land use change from the
transfer of land use cover, but not on the spatial distribution and trans-
fer types (Wickramasuriya et al., 2009). By contrast, CA can interpret
accurately the spatial–temporal dynamic evolution of complex space
Fig. 2. Administrative and remote sense maps of Tieling.
1073
systems, such as the LULC based on the suitable transfer rules (Behera
et al., 2012). Hence, CA and Markov complement each other. Markov
chain supplies the transition area of LULC as the transfer rule of the
CA, whereas CA computes and simulates the area, distribution, and
transfer type of LULC.
3. Calculation
3.1. General situation of the research zone
Tieling is a prefecture level city in Liaoning Province, PR China
(Fig. 2). This city has a population of approximately 3,054,000 inhabi-
tants according to the 2008 census and an area of 13,000 km2, including
203 km2 of the city proper. Tieling is situated along the Liao River. This
city is a highly agricultural-industrialized area and has developed into a
thriving center for coal, power, and raw materials.
3.2. Modeling
3.2.1. Model hypothesis
To determine the efficiency and accuracy of the proposed model, the
following six hypotheses were established in this study: 1. the spatial
scope of the model is the boundary of Tieling; 2. the basis time spot
is 2006 and the end time is 2030; 3. the external system is temporal,
uncontrolled, and not interfered by the internal system; 4. people in
urban or rural areas have similar lifestyles, consumptions, and so on;
5. the plants in one industry have no difference in terms of production,
1074 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081

investment policy, discharge efficiency, and so on; and 6. water supply The Granger causality provided the main relationships among the
and water pollution from the upstream is constant. human activities and the proof that the complex social–natural system
was not limited by the sole linear natural source.

3.2.2. Social-economic component
3.2.2.1. Granger causality test. According to the Granger causality test, we
collected the time series data on population (POP), GDP, industrial out-
put (IO), industrial added value (IAV), water demand (WD), household
emission, and industrial emission from 1991 to 2010. The stationarity
test result that was computed using the Augmented Dickey–Fuller test
by E-views software shows that the POP, GDP, IO, IAV, WD, industrial
COD (ICOD), industrial NH4N (INHN), and household NH4N (HNHN)
were second-order stationarities, and that they may have causality
loop among them.
From the stationarity test, we performed a cointegration test be-
tween the variables with the similar orders of stationarity. The results
show that POP, GDP, IO, IAV, WD, HNHN, INHN and ICOD have some
cointegration relationship.
The cointegration test indicates the presence of at least one Granger
causality relationship among the variables with cointegration relation-
ship. According to E-views, if the probability is less than 1%, the
null hypothesis does not exist. Based on this principle, the directions
and quantities of Granger causality among the variables are listed in
Table 1.
From the result of the Granger causality test, the causal loop of
the subsystem clearly described that the GDP caused the change in
the population, and this meant that the economy affected the popula-
tion size. The change in the population size resulted in a change in
the water resource demand and household contaminant emission. In
reality, a population boom would mean a high household water de-
mand and household contaminant emission. Household emission was
the main cause of industrial pollution discharge. Then, the industrial
discharge affected the production output. Meanwhile, the increase in
industrial emission showed that the more raw materials were con-
sumed, the larger was the output of the production system. A large in-
dustrial output indicated a large industrial added value. The increasing
industrial added value increased the GDP and became its main compo-
nent. Therefore, the entire loop showed that the economy was the
cause of the population boom, the population boom was responsible
for the water resource and household pollution emissions, household
emission caused industrial discharge, industrial contamination was
the cause of a large production scale, the production scale resulted in
production benefits, and production benefits were the reason for the
increase in GDP.
3.2.2.2. Population. Population size is the main index of the social state
and carrying capacity. In this study, we adopt the macro view in the
population size tendency using the following function:
POP ðt Þ ¼ POP ðt−1Þ  ðbr þ ir−dr−erÞ  POPPC
where POP is the population size, br is the birth ratio, ir is the immigra-
tion ratio, dr is the death ratio, er is the emigration ratio, and POPPC is
the population policy coefficient. The population size was described as
the other function as follows:
POP ¼ UP þ RP
where UP is the urban population size and RP is the rural population
size.
3.2.2.3. Economy. Economy indicated the energy and impact of the social-
economic component. The GDP is usually used to describe the level of
economic development. In this study, GDP was calculated according to
the expenditure accounting approach, using the formula as follows:
GDP ¼ PC þ GC þ GF þ NEX
where GDP is the gross domestic product, PC is the people total con-
sumption, GC is the government consumption, GF is the gross fixed
capital formation, and NEX is the net export. PC was described as the
sole reason of immigration that affects the population size. GF caused
the investment in production to affect the industrial output and benefit.
However, GC depended on the taxes from the benefit of production,
and NEX was a constant in this study for external disturbances such as
exchange rate, among others.
3.2.2.4. Production. Based on the production structure, the subsystem
was divided into agriculture, industry, and service. Industry was further
divided into building, mining, manufacturing, and energy. According to
the ‘Economy’ section, production scale describes the total output in
every industry using the formula as follows:
TO ¼ INV þ ADV
where TO is the total output, INV is the investment, and ADV is the added
value. The other index that was used to indicate the state of production
was calculated as follows:

MB ¼ ADV=INV
Table 1
The Granger causality test result. where MB is the marginal benefit. According to the recessive unemploy-
ment theory, the marginal benefit affected directly emigration.
Null hypothesis F-statistics Prob.

POP does not Granger cause GDP 2.98793 0.0776

3.2.2.5. Water consumption and contaminant emission. In the water con-
GDP does not Granger cause POP 37.7597 4.00E−06
GDP does not Granger cause IAV 0.09916 0.9062
sumption subsystem, the household-production-ecology framework
IAV does not Granger cause GDP 4.456 0.0318 was considered into the model. Water consumption was computed as
GDP does not Granger cause IO 0.76329 0.4223 follows:
IO does not Granger cause GDP 27.5595 0.0033
IO does not Granger cause IAV 6.79005 0.0189 WC ¼ HWC þ AWC þ IWC þ EWC
IAV does not Granger cause IO 4.29845 0.054
POP does not Granger cause WD 0.34897 0.7413
where WC is the water consumption, HWC is the household water con-
WD does not Granger cause POP 11.5494 0.0797
WD does not Granger cause HNHN 4.47728 0.088 sumption, AWC is the agricultural water consumption, IWC is the indus-
HNHN does not Granger cause WD 8.61755 0.0324 trial water consumption, and EWC is the ecological water consumption.
HNHN does not Granger cause INHN 1.22595 0.3186 EWC was calculated by the function as follows:
INHN does not Granger cause HNHN 8.28145 0.0347
INHN does not Granger cause ICOD 9.87079 0.0256 EWC ¼ 0:12  WRS
ICDO does not Granger cause INHN 0.11315 0.7502
ICDO does not Granger cause IO 5.40008 0.0502 where 0.12 is the ecological water resource coefficient, as suggested by
IO does not Granger cause ICOD 5.95139 0.0419
the World Commission on Environment and Development, and WRS is
 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081
the water resource supply. Analogous to the water consumption,
the contaminant subsystem was also divided into several sections
according to the pollution source, using the formula as follows:
WP ¼ HWP þ IWP þ NPS
where WP is the water pollution, HWP is the household water pollution,
IWP is the industrial water pollution, and NPS is the non-point source
pollution. The NPS consists of the runoff and livestock and is computed
as follows:
NPS ¼ UPC  UA þ GPC  GA þ UCC  UCA þ PAC  PA þ BLC  BL
þ SLC  SL
where UA is urban area, GA is green and dry field area, UCA is uncover
area, PA is paddy area, and BL is the quantity of large livestock involving
cow, pig, and so on. SL is quantity of small stock (e.g., duck, chicken, and
rabbit). UPC, GPC, UCC, PAC, BLC, and SLC were non-point source pollu-
tion intensities of different sources that were suggested by the research
in Water Pollution Control and Treatment Major Project.
3.2.3. Natural component
3.2.3.1. Water resource supply. In water resource supply, the similar func-
tion of the NPS is used as follows:
WRS ¼ PREC  ðUWC  UA þ GWC  GA þ UCWC  UCA þ PWC  PA
þWAÞ þ UPWRS−EV
where PREC is the precipitation, WA is the water area, UPWRS is the up-
stream water resource supply, and EV is evaporation. UWC, GWC, UCWC,
and PWC are runoff coefficients of different sources that were presented
by Liu and Zhang (2007).
According to the above function, UPWRS is the uncontrolled value
from the external system, and EV is regarded as the constant in long
time series. Thus, the precipitation and LULC are the keys in the water
resource supply.
3.2.3.2. Precipitation. We predicted the precipitation of Tieling by the RBF
approach based on the precipitation data from 1981 to 2010 in Fig. 3.
We built the RBF net and used the data from 1981 to 2010 to train
the net and analyze the accuracy of the net by comparing the observed
and simulated data in Fig. 3.
From Fig. 3, the accuracy of the prediction method in precipitation
was accepted in the error analysis. We forecasted the precipitation
t ¼ ðZ  f þ M Þ=W
Fig. 3. Simulated and observed precipitation data of Tieling from 1981 to 2010.
1075
data from the 2011 to 2030 by the RBF net. Then, we input the predic-
tion data into the system dynamics model.
3.2.3.3. LULC. We undertook the land use classifications in 2006, 2007,
and 2010 (Figs. 4) based on the LANDSAT 7 images by maximum likeli-
hood method via ENVI software.
We analyzed the transition area matrices between the classifications
by Markov-chain. Then, we used the matrix as the transforming rule in
the CA method to predict the geographic distribution in 2010. A total of
500 random points were set in the prediction and observed maps. The
predicted and observed data were compared to test the error of the
CA–Markov method as applied in this study. The test shows 75% accura-
cy from the results obtained. CA–Markov was applied to forecast the
LULC from 2015 to 2030 (Figs. 5). Then, we input the prediction data
of LULC into the model.
Fig. 5 shows that the change in urban land use is not obvious in
terms of either quantity or spatial location from 2015 to 2030. Dry
field and forest land use are the most dominant land uses, and changes
in area and location are not obvious during this period. No obvious
change in the area and location of water land use can be observed
from 2015 to 2030. This result means that water land use did not con-
vert the other land usage. Uncovered land was mainly located in the
north and west of Tieling City along the river, but no obvious increases
or decreases were observed in the area of uncovered land use. The
paddy located east of the research zone and the area around it did
not obviously change from 2015 to 2030. According to the count on
different land uses in the area, the urban area is 561 km2 in 2015 and
522 km2 in 2030. The descent area of small urban land uses converts
to the other land use in Fig. 5. The area of dry land and forests is
10,388 km2 in 2015 and 10,391 km2 in 2030. In contrast with these
two values, the change in area of dry fields and forests is not obvious.
The area of water is 180 km2 in 2015 and 179 km2 in 2030, and
no change in water land use is observed. The area of uncovered land
is 980 km2 in 2015 and 979 km2 in 2030. No change in the use of un-
covered land is observed. The area of the paddy is 853 km2, and it has
the same area in 2030. The area of paddy land use increases from 2015
to 2030.
3.2.3.4. Water environment capacity. We used the one-dimensional
water environment quality model to calculate the COD and NH4N envi-
ronment capacity as the natural environment limit of the complex sys-
tem, according to the nature of Liao River in Tieling.
3.2.4. Interaction component
In this component, we modeled the two different management
policies to stimulate the state of the natural–social system and analyze
the different impacts.
3.2.4.1. Pollutant discharge fee policy. This policy is the current environ-
mental management policy in the study area. Discharge fee comprises
real fee and fine. The discharge fee is 0.7 CNY per equivalent contami-
nant, and the fine is twice the fee. The policy states that the pollution
which is less than the water environment capacity is included in the
discharge fee, whereas the excess is included in the fine.
3.2.4.2. Pollutant discharge tax policy. The tax policy is being discussed in
China so this policy is currently uncertain. However, we still presented
the policy case according to the Pigou tax theory. The tax policy case
was set as follows:
¼ f½ Z  nð CODÞ  pð CODÞ þ ½Z  nðNH4 NÞ  P ð NH4 NÞ þ ðM Þg=W
where t is the tax, Z is the total fee in recovering the water environment
or treating the contaminant, f is the pollution coefficient, M is the total
fee in managing the policy, W is the total pollution equivalent as the
1076 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081

Fig. 4. Land use classification maps of Tieling in 2006, 2007 and 2010.
S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081 1077

Fig. 5. Stimulated LUCC maps of Tieling in 2015, 2020, 2025 and 2030.
1078 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081

Table 2
The error test result of F(S, N, I).

Time POP (10,000 capita) GDP (10,000 CNY)

Stimulated Observed Error (%) Stimulated Observed Error (%)

2006 304.54 304.5 0.013 2827001.64 3139300 −9.948

2007 305.94 305.4 0.177 3644242.07 4039700 −9.789
2008 307.14 305.9 0.405 4851927.10 5363000 −9.530
2009 308.17 306.1 0.676 5723939.66 6057065 −5.500
2010 309.01 305.15 1.265 7921219.00 7221291 9.693

Time IVA (10,000 CNY) WD (109 m3)

Stimulated Observed Error (%) Stimulated Observed Error (%)

2006 1,197,642.00 1,263,844 −5.238 11.38 12.24 −7.026

2007 1,318,959.16 1,308,221 0.821 11.38 10.64 6.955
2008 1,454,570.56 1,564,007 −6.997 12.48 10.98 13.661
2009 1,606,153.65 1,675,161 −4.119 12.25 11.02 11.162
2010 1,775,521.43 2,021,919 −12.186 14.63 13.3 10.000

Time HCOD (t) HNHN (t)

Stimulated Observed Error (%) Stimulated Observed Error (%)

2006 19,733.40 19,691.3 0.214 11.38 12.24 −7.026

2007 20,535.11 18,935.08 8.450 11.38 10.64 6.955
2008 20,811.10 19,564.33 6.373 12.48 10.98 13.661
2009 20,946.81 22,291.29 −6.031 12.25 11.02 11.162

Time ICOD (t) INHN (t)

Stimulated Observed Error (%) Stimulated Observed Error (%)

2006 3736.5 4326.5 −13.637 221.35 221.57 −0.099

2007 4143.34 4741.03 −12.607 259.31 285.2157 −9.083
2008 4384.8 4430.78 −1.038 231.6 175.1 32.267
2009 4613.2 4493.77 2.658 297.42 406.48 −26.830

former policy, n(COD) is the discharge weight of COD, n(NH4N) is the
discharge weight of NH4N, p(COD) is the hazard weight of COD, and
p(NH4N) is the hazard weight of NH4N. Contrary to the pollutant dis-
charge fee policy, the tax is not constant but a variable related to the
fee in treating the contaminant to narrow the gap between the private
and social costs.
4. Results
4.1. Error test
For the accuracy of the model, the simulation results of POP, GDP,
IVA, WD, and contaminant production were compared with the ob-
served data from the statistical data. The results from the error analysis
are presented in Table 2.
Table 2 shows the error rates of the six subsystems, which verify the
reliability of the model. The majority of the errors were mostly con-
trolled in the 10% level.
policy that encouraged or supported the birth rate. However, a low
value of the coefficient of population indicates the limitations of
the population policy. The population policy indicates the effect of
entophytic change on the population of the entire system. The economy
reflects the effects of stimulations to it. When the value of the economy
coefficient is high, consumption decreases and investment increases.
Therefore, the economy coefficient reflects the states that are pre-
ferred by economic development. Production is the coefficient value
of product policy. The higher the value of production is, the greater
the industrial investment is. When the value is low, investment is pre-
ferred over agriculture and service. Therefore, the coefficient shows
the impact of product policy on the entire system. The environment is
regarded as the coefficient value of environmental policy, which refers
to the ratio of the environmental input to the entire system. When
this value is high, the environmental investment is large, and the pure
product investment is low. We tested the impact of environmental
policy on the model.
Fig. 6 shows that the sensitivity of the model was acceptable.

4.2. Sensitivity test 4.3. Results and analysis

The controller parameters are typically matched to the process char- The model was applied to simulate the reality of the complex social–
acteristics. Given the process may change, choosing the controller pa- natural system for the water carrying capacity. Simulation results under
rameters in such a way that the closed loop system is not sensitive to different policies of water resource, water environment, and social-
variations in the process dynamics is important. In system dynamic economy are presented in Fig. 7.
model, sensitivity testing is the process of changing the assumptions
about the value of constants in the model and examining the resulting
Table 3
output. In this research, internal constant did not exist for the complex
The values of constants in the sensitivity test.
relationship between the variables and external coefficient, denoting
that the natural factor was uncontrolled. Thus, the policy coefficient Policy Range Case 1 Case 2 Case 3 Case 4
was chosen for the sensitivity test. The values of constants are listed in Population [0,2] 1.26 0.224 0.346 1.621
Table 3. Economy [0,1] 0.631 0.122 0.173 0.655
Population reflects the effect of population policy on birth control. Production [0,3] 1.26 0.224 1.016 0.897
Environment [0,1] 0.631 0.112 0.373 0.855
The value of coefficient of population was used to express the state of
 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081 1079

Fig. 6. The sensitivity of the model.

The tendencies of the complex system under the two policies are evi-
dently shown in Fig. 7. In terms of population, the difference between pol-
icies is mainly in quantity. The number under the fee policy is
2.85 million, and the other condition is predicted to be 2.9 million in
2030. The size reductions in the death ratio are caused by aging, lower
birth ratio and higher immigration ratio caused by the higher GDP, and
the emigration ratio caused by the unemployment in different
levels under the two policies. The GDP is expected to increase to
582,097,210,000 CNY in 2030 under the fee policy, and the tax policy
will cause the increase to 585,952,340,000 CNY in 2030. Industrial
production gained the most similar benefits with different investments
under the two policies. Under the fee policy, 151,573,960,000 CNY
of IAV was caused by 380,787,930,000 CNY in investment. Under
the tax policy, it costs 268,055,910,000 CNY in investment for the
157,063,090,000 CNY of added value. The total WD increased to
14.42 × 109 m3 and 14.39 × 109 m3 in fee and tax policies, respectively.
From Fig. 7, the imbalance between the water supply and water demand
is predicted to frequently exist from 2006 to 2030 such that water
resource is not enough to support the system carrying capacity. In the
environmental aspect, the discharge quantity is mostly different under
the two policies. The point source COD emission to the river is
105,381 t and NH4N is 3543 t under the environmental fee. The COD
emission is 86,632.37 t and the NH4N is 2854.5 t under the environ-
mental tax. The water environment carrying capacity of COD can afford
the natural–social system, but NH4N capacity is the key limitation to the
sustainable development.
5. Discussion
Simulation results showed that the Granger causality test was
reasonable and reliable before it was used in the system dynamics
modeling process. The causal loop based on the quantitative analysis
1080 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081

Fig. 7. Simulated result of the model.

was closer to reality than the traditional sole qualitative causal loop.
Meanwhile, system dynamics were used to build the microcosmic eco-
logical carrying capacity model based on the loop. By running the
model, we simulated contaminant emission and resource consumption
under the development of the society and economy. For accurate
results, we adopted the RBF net to forecast precipitation and used the
CA–Markov to predict the LULC for the natural impact on resources
and environment supply. In the simulation, we added the environmen-
tal tax and environmental fee policies into the model. The simulation
results show that the social–ecological carrying capacity under the tax
policy is more sustainable than that under the fee policy. Obviously,
a low investment is involved in terms of natural resources and social
capital for systems with a similar or large size. The fee policy gradually
abates in favor of low private costs and thus creates a prisoner's
dilemma in the industry. Raising private costs to purchase high unit
prices for increased benefit is the optimal policy for the entire industry.
However, competition and gap between private and public costs exist
because having a low private cost to achieve a superior standing in the
market is the optimal policy for the private industry. Therefore, optimal
private opinion is not beneficial for the entire industry. For example,
discharging contaminants at a low cost and not treating these contam-
inants at a high cost is the main choice according to research under
the fee policy. By contrast, the tax policy narrows the gap between the
discharging cost and the treatment cost and thus causes some private
industries to adopt an effective treatment method with a low private
cost. However, a high cost causes a low productive investment and a
high unit price for increased benefit. The marginal benefit then rises
to reduce recessive unemployment, which causes immigration, and
to maintain the size of the population. Compared with the traditional
single method, our integrated model is closer to reality in three aspects.
First, the society and economy are regarded as negative factors in the
carrying capacity system in traditional research. However, we consider
the society and economy as the consumers and producers of the entire
system. Reasonable development is the driver of carrying capacity.
 S. Wang et al. / Science of the Total Environment 472 (2014) 1070–1081
Second, we discussed the reason and the changing state of water carry-
ing capacity under different policies. Finally, we considered the natural
subsystem as a variable, but in traditional research, it is considered a
constant. These characteristics make our work reflective of the relation-
ships among society and economy, nature, and their interaction.
6. Conclusion
The integrated system dynamic model of carrying capacity under
the different policies presented in this paper is reliable and reasonable
in practice and theory. The model can predict the temporal status of
the carrying capacity state of a complex system in water resource
and environment under nonlinear management policies. The difference
in the tendencies of water carrying capacity can be interpreted by
the cause loop of a system dynamic model, according to the statistics
and game theory. This paper presents the carrying capacity function,
RC = F(S, N, I), which considers human activities as the resource in
the system and the modeling method based on the integrated method
involving system dynamics, econometrics, and RS-GIS.
However, the large quantity of data limited the application of
the model and the similitude to reality. Moreover, the production and
natural components are still macroscopic. Thus, these two aspects will
be the focus of future studies.
Acknowledgment
The authors acknowledge the financial support of the National
Science and Technology Major Project (Water Pollution Control and
Treatment Major Project (NO. 2009ZX07526-006)) and Fundamental
Research Funds for the Central Universities.
References
Åstebøl SO, Hvitived-Jacobsen T, Simonsen Ø. Sustainable stormwater management at
Fornebu—from an airport to an industrial and residential area of the city of Oslo,
Norway. Sci Total Environ 2004;334–335:234–49.
Awasthi A, Chauhan SS. Using AHP and Dempster–Shafer for evaluating sustainable trans-
port solutions. Environ Model Softw 2011;26(2):787–96.
Behera MD, Borata SN, Panda SN, Behera PR, Roy PS. Modelling and analyzing the water-
shed dynamics using Cellular Automata (CA)–Markov model —a geo-information
based approach. J Earth Syst Sci 2012;121:1011–24.
Carbonell G, Ramos C, Pablos MV, Ortiz JA, Tarazona JV. A system dynamic model for the
assessment of different exposure routes in aquatic ecosystems. Sci Total Environ
2000;247:107–18.
Casasent D, Chen X. New training strategies for RBF neural networks for X-ray
agriculatural product inspection. Pattern Recogn 2003;36:535–47.
Cerutti AK, Bagliani M, Beccaro GL, Bounous G. Application of Ecological Footprint Analy-
sis on nectarine production: methodological issues and results from a case study in
Italy. J Clean Prod 2010;18:771–6.
Chen CH, Liu WL, Liaw SL, Yu CH. Development of a dynamic strategy planning theory and
system for sustainable river basin land use management. Sci Total Environ 2005;346:
17–37.
del Monte-Luna P, Brook BW, Zetina-Rejón MJ, Cruz-Escalona VH. The carrying capacity of
ecosystems. Glob Ecol Biogeogr 2004;13:485–95.
Ehrlich PR, Holdren JP. Impact of population growth. Science 1971;171:1212–7.
1081
Feng LH, Zhang XC, Luo GY. Application of system dynamics in analyzing the carrying ca-
pacity of water resources in Yiwu City, China. Math Comput Simul 2008;79:269–78.
Gao L, Hailu A. Ranking management strategies with complex outcomes: an AHP-fuzzy
evaluation of recreational fishing using an integrated agent-based model of a coral
reef ecosystem. Environ Model Softw 2012;31(12):3–18.
Geng Y, Zhang P, Ulgiati S, Sarkis J. Emergy analysis of an industrial park: the case of
Dalian, China. Sci Total Environ 2010;408:5273–83.
Giantomassi A, Ippoliti G, Longhi S, Bertini I, Pizzuti S. On-line steam production prediction
for a municipal solid waste incinerator by fully tuned minimal RBF neural networks.
J Process Control 2011;21:164–72.
Granger CWJ. Investigating causal relations by econometric models and cross-spectral
methods. Econometrica 1969;37(3):424–38.
Han H, Chen Q, Qiao J. An efficient self-organizing RBF neural network for water quality
prediction. Neural Netw 2011;24:717–25.
Haraldsson HV, Ólafsdóttir R. A novel modelling approach for evaluating the preindustrial
natural carrying capacity of human population in Iceland. Sci Total Environ 2006;372:
109–19.
Jin W, Xu LY, Yang ZF. Modeling a policy making framework for urban sustainability:
incorporating system dynamics into the Ecological Footprint. Ecol Econ 2009;68:
2938–49.
Kessler JJ. Usefulness of the human carrying capacity concept in assessing ecological sus-
tainability of land-use in semi-arid regions. Agric Ecosyst Environ 1994;48:273–84.
Liu Q, Zhang B. Study on land use/cover change, environmental effect and the optimal
regulation of Liaohe River Basin. Beijing: Science Press; 2007122.
Mao IF, Chen MR, Wang L, Chen ML, Lai SC, Tsai CJ. Method development for determining
the malodor source and pollution in industrial park. Sci Total Environ 2012;437:270–5.
Meadows DH, Meadows DL, Randers J, Behrens III WW. Limits to growth: a report for the
club of Rome's project on the predicament of mankind. New York: Universe Books;
1972.
Ni X, Wu YQ, Wu J, Liu J, Wilson PC. Scenario analysis for sustainable development of
Chongming Island: water resources sustainability. Sci Total Environ 2012;439:
129–35.
Odum HT. Environmental accounting: emergy and environmental decision making.
New York: John Wiley and Sons; 1996370.
Rees WE. Eco-footprint analysis: merits and brickbats. Ecol Econ 2000;32(3):371–4.
Roberts N, Anderssen DF, Deal RM, Garet MS, Shaffer WA. Introduction to computer
simulation. Portland: Productivity Press; 1983 [562 pp.].
Sang L, Zhang C, Yang J, Zhu D, Yun W. Simulation of land use spatial pattern of towns and
villages based on CA–Markov model. Math Comput Model 2011;54:938–43.
Seidl I, Tisdell CA. Carrying capacity reconsidered: from Malthus' population theory to
cultural carrying capacity. Ecol Econ 1999;31:395–408.
Stöglehner G. Ecological footprint —a tool for assessing sustainable energy supplies.
J Clean Prod 2003;11:267–77.
Sušnik J, Vamvakeridou-Lyroudia LS, Savić DA, Kapelan Z. Integrated system dynamics
modelling for water scarcity assessment: case study of the Kairouan region. Sci
Total Environ 2012;440:290–306.
Venkatesan AK, Ahmad S, Johnson W, Batista JR. Systems dynamic model to forecast
salinity load to the Colorado River due to urbanization within the Las Vegas Valley.
Sci Total Environ 2011;409:2616–25.
Wackernagel M, Rees WE. Perceptual and structural barriers to investing in natural
capital: economics from an ecological footprint perspective. Ecol Econ 1999;20(1):
3–24.
Wickramasuriya R, Bregt A, van Delden H, Hagen-Zanker A. The dynamics of shifting
cultivation captured in an extended constrained cellular automata land use model.
Ecol Model 2009;220:2302–9.
Wolfslehner B, Vacik H. Evaluating sustainable forest management strategies with the
analytic network process in a pressure–state–response framework. J Environ
Manag 2008;88:1–10.
Yue TX, Tian YZ, Liu JY, Fan ZM. Surface modeling of human carrying capacity of terrestrial
ecosystems in China. Ecol Model 2008;214:168–80.
Zerger A, Warren G, Hill P, Robertson D, Weidemann A, Lawton K. Multi-criteria assess-
ment for linking regional conservation planning and farm-scale actions. Environ
Model Softw 2011;26:103–10.
Zhao N, Liu Y, Chen JN. Regional industrial production's spatial distribution and water
pollution control: a plant-level aggregation method for the case of a small region in
China. Sci Total Environ 2009;407:4946–53.