Journal of Environmental Management 238 (2019) 59–63

Contents lists available at ScienceDirect

Journal of Environmental Management

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

Research article

Combination and placement of sustainable drainage system devices based on T

zero-one integer programming and schemes sampling
Mingming Wanga,∗, Yun Wanga, Xuerui Gaoa, Chris Sweetappleb
a
School of Architectural and Civil Engineering, Anhui University of Technology, Maxiang Rd., Ma'anshan, 243032, Anhui province, PR China
b
Centre for Water Systems, College of Engineering, Mathematics and Physical Sciences, University of Exeter, North Park Rd., Exeter, EX4 4QF, Devon, United Kingdom

ARTICLE INFO ABSTRACT

Keywords: The combination and placement of sustainable drainage systems (SuDS) devices is important for system design,
Device placement but differing site characteristics and device properties can make this a challenging task. Opinion-based and
Sustainable drainage systems (SuDS) optimization-based approaches have the disadvantages of subjectivity and excessive computational burden re-
Zero-one integer programming spectively. This paper presents a new framework for SuDS device combination and placement in system design.
Random sampling
It integrates zero-one integer programming, random sampling, scheme filtering and cost-effectiveness analysis.
Cost-effectiveness analysis
The effectiveness of the framework is tested with a SuDS device placement design case in Chizhou city, Anhui
province, China. The proposed framework will help to objectively choose the best SuDS device combination and
placement scheme for cost-effective implementation.

1. Introduction
Sustainable drainage systems (SuDS) devices such as bioretention
cells (BC), permeable pavements (PP) and green roofs (GR) are im-
portant components of urban drainage systems (Chui et al., 2016; Wang
et al., 2017b). The combination and placement of SuDS devices in a
drainage system play an important role in the performance of the whole
system (Liu et al., 2016a; Sweetapple et al., 2018; Wang et al., 2017a;
Zhen et al., 2004). Due to the wide variety of SuDS devices and the
unique characteristics of each drainage system, however, determining
the most appropriate combination and placement of SuDS devices is a
challenging problem. As such, increasing attention has been paid to the
placement of SuDS devices (Chui et al., 2016; Hu et al., 2015; Lee et al.,
2012; Liu et al., 2016a; Perez-Pedini et al., 2005; Zhang et al., 2013;
Zhen et al., 2004).
There are two kinds of approaches to determine the placement of
SuDS devices in current literature: opinion-based and optimization-
based. The opinion-based approach here refers to subjective placement
of SuDS devices by designers based on a combination of factors, in-
cluding their opinions and experience. SuDS devices are implemented
in this manner in some subcatchments (Hu et al., 2015; Yang and Chui,
2018). Opinion may come into play particularly in high level SuDS
location decisions – for example, Yang and Chui (2018) chose to place
SuDS devices in the first three of four subcatchments. Opinion-based
design decisions may also form the foundation of later detailed design
decisions. For example, at the planning and design stage, a limited
number of potential schemes (usually no fewer than three) that are
expected to meet the required goals are selected according to opinion
and experience. These represent only a limited number of the possible
cases (Hu et al., 2015) and, therefore, the final scheme may be not the
optimum one.
To address this challenge, several different optimization-based ap-
proaches have been developed. Zhen et al. (2004), for example, opti-
mized the placement of stormwater basins using a scatter search algo-
rithm (heuristic optimization technique) and an agricultural non-point
source pollution model (AnnAGNPS). Perez-Pedini et al. (2005) de-
termined the optimal location of infiltration-based best management
practices (BMPs) using a distributed hydrologic model combined with a
genetic algorithm. Lee et al. (2012) developed a design optimization
model, System for Urban Stormwater Treatment and Analysis Integra-
tion (SUSTAIN), to evaluate the optimal location, type, and cost of
BMPs. Zhang et al. (2013) used a multi-objective optimization method,
epsilon -NSGAII combined with Storm Water Management Model
(SWMM) to screen alternative site layouts for low impact development
(LID) technologies. Liu et al. (2016a) presented the multi-level spatial
optimization (MLSOPT) framework, which used the AMALGAM opti-
mization algorithm and the hydrological/water quality model (L-THIA-
LID 2.1) to optimize the selection and placement of low impact devel-
opment practices. In these optimization procedures, the simulation
runtime (many days or even weeks) required by the powerful

∗
Corresponding author.
E-mail addresses: wmmahut@163.com, wangmm@ahut.edu.cn (M. Wang).

https://doi.org/10.1016/j.jenvman.2019.02.129
Received 15 August 2018; Received in revised form 17 January 2019; Accepted 28 February 2019
0301-4797/ © 2019 Elsevier Ltd. All rights reserved.
M. Wang, et al.
Fig. 1. Framework for SuDS device placement.
optimization algorithms imposes a practical barrier to optimization
(Butler et al., 2018).
Besides the above opinion-based and optimization-based approaches, is
there another approach that can be used to obtain more cost-effective
schemes and alleviate this high computational burden? To address this, this
paper proposes a new framework for determining the combination and
placement scheme of SuDS devices. The approach is based on zero-one
integer programming, scheme sampling and filtering, and analysis of
cost-effectiveness, integrating assessment of control effectiveness and
cost-effectiveness. To demonstrate the framework, a SuDS device design
case study in Chizhou city, Anhui province, China is presented.
2. Methodology
The proposed framework (Fig. 1) contains three key components:
zero-one integer programming, scheme sampling and filtering, and
cost-effectiveness analysis.
2.1. Zero-one integer programming
An integer programming (IP) problem is a linear programming (LP)
problem used to achieve the best outcome when some or all of the
variables are required to be integer values (Chen et al., 2010). Zero–one
integer programming is the special case of IP where variables are re-
stricted to 0 or 1 (Chen et al., 2010). It has proved useful in different
types of problems in planning, scheduling and distribution (Awumah
et al., 1989; Chen et al., 2010). In this study, zero-one integer pro-
gramming is chosen for generation of combination and placement
schemes for SuDS devices, since this zero-one approach enables a de-
cision to be made as to whether a device is selected for the layout
within each sub-catchment (Awumah et al., 1989). The zero-one integer
programming method is applied as follows:
(1) Divide the catchment into sub-catchment areas according to site
and catchment characteristics.
(2) Identify all feasible SuDS devices according to characteristics of the
site, catchment and SuDS devices.
(3) Number all potential devices in the catchment (to give each an ID),
thus forming a selectable device ID set. The total amount of po-
tential SuDS devices is recorded as N.
(4) Zero-one integer programming model: for each ID (device) in each
subcatchment, there are only two possible implementation states:
implemented or not implemented. Thus a zero-one integer pro-
gramming model is applied, in which ‘0’ indicates that the ID (de-
vice) is not implemented in the subcatchment, and ‘1’ indicates that
it is.
(5) Generation of SuDS device combination and placement schemes:
based on the zero-one integer programming model, different ID
60
Journal of Environmental Management 238 (2019) 59–63
devices in each subcatchment are combined according to their ‘0’ or
‘1’ implementation values. The number of devices is increased from
1 to N, thus forming different combination and placement schemes.
2.2. Scheme sampling and filtering
The specific scheme sampling method is described as follows:
Generate SuDS device scheme samples containing from 0 to N devices,
where N is the total number of potential SuDS devices. The sample size
required in each random sampling round is computed by eq. (1) (Brase
and Brase, 2012; Diao et al., 2016).
NS · p ·(1 p)
nR =
(NS 1)· ( )
CI 2
Z
+ p ·(1 p)
(1)
where nR is the sample size in each sampling round; NS is the total
number of all possible schemes in each sampling round; p is the prob-
ability of success, set as 0.5; CI is the confidence interval, set as ± 5%;
and Za is the normal distribution value of given confidence (e.g.
1.960 at a confidence level of 95%, and 2.575 at 99%).
According to the control target, the sampled schemes are filtered to
leave only those that provide the required level of control.
2.3. Cost-effectiveness analysis
The relevant assessment data include control effectiveness and
costs. These data can be obtained from the design calculation and
drainage system model. The cost-effectiveness ratio (ER) is computed
using eq. (2):
ER = Controleffectiveness/Cost (2)
where control effectiveness represents the degree to which control
targets such as flow volume and pollutant load reduction are met (Lee
et al., 2012; Liu et al., 2016a). The cost includes the construction cost
and annual maintenance cost of SuDS devices, further details are given
in the Supplementary Information.
The results of schemes are ranked from high to low by the cost-
effective ratio according to the cost-effective analysis, therefore en-
abling the final SuDS combination and placement scheme to be de-
termined.
3. Case study
For the purposes of demonstrating the framework, a catchment
(30.56 ha) which is a part of a new residential and commercial devel-
opment area (192 ha) in Chizhou city, Anhui province, China (Wang
et al., 2017b) was selected.
The study area is divided into 7 subcatchments (S01eS07) (Fig. 2).
The storm sewer network was designed for a 3-year return period
storm.
The study case area is relatively flat, with a slope of 0.2–2.0%, and
pre-development was agricultural land. The infiltration capacity of the
soil is sufficient to allow the selection of infiltration SuDS devices, such
as permeable pavements and bioretention cells. Current land use and
vegetation cover vary across the case study area and the placement of
SuDS devices considers land use category (residential area, commercial
area, market area, road, green space) and land cover. Connectivity to a
conventional storm sewer system is available and runoff exceeding the
capacity of the SuDS devices is drained into this. The preferred SuDS
devices, which include bioretention cells (BC), permeable pavements
(PP) and green roofs (GR), were selected based on the local site char-
acteristics. Further details on the SuDS devices’ parameters set and
usage are given in the Supplementary Information (Tables S2 and S3).
The control target of the local site spongy city is an annual runoff
volume capture ratio (CR) of 80% (Wang et al., 2017b). To meet this
target, three preferred SuDS devices are considered in each
M. Wang, et al.
Fig. 2. Layout and properties of study area.
subcatchment.
Setting the coverages of porous pavement and green roof to be at
certain percentages is recommended in some guidelines for spongy city
planning or construction in China (SMCHURD, 2015; SMULRC, 2016).
For example, a porous pavement coverage of 90% for new development
is specified in the Shenzhen Spongy City Planning Key Points and De-
tailed Regulations (SMULRC, 2016), and a green roof coverage of
≥30% for new communities is stipulated in the Shanghai Spongy City
Construction Technical Guidelines (SMCHURD, 2015). Therefore, to
simplify the settings, the permeable pavement and green roof coverages
are set to 90%, 30% respectively in this case study; however, alternative
coverages could also be explored using this methodology. Due to there
being no building in subcatchments S01 and S02 (i.e. roof area is zero),
there are no green roofs are in these areas (the roof areas are zero).
The SuDS device ID Numbers are set from 1 to 21 for BC, PP, GR in
all 7 sub-catchments (ID from 1 to 3 are ID numbers of BC, PP, GR in the
catchment S01, respectively, and so on) (Table 1), forming SuDS device
ID set (total N = 21).
This case study considers the placement of SuDS devices during the
planning and design stage. In this stage, the catchment area is relatively
large, not the small, community-level area that would be considered for
detailed design. There are two different approaches for placing SuDS
devices within a subcatchment in SWMM (United States Environmental
Protection Agency, 2018). The chosen method will place one or more
controls in an existing subcatchment that will displace an equal amount
of non-LID area from the subcatchment (United States Environmental
Protection Agency, 2018). This approach is adopted because this re-
search is a large, area-wide study where a mix of controls would be
deployed over many different subcatchments (Rossman and Huber,
2016). Details of the placement of SuDS device within a subcatchment
in SWMM are shown in Supplementary Information (Fig. S1).
For each subcatchment, each ID has only two possibilities: im-
plemented or not (0 stands for this device is implemented, 1 stands for
not implemented). Therefore, a zero-one integer programming model is
formed.
Table 1
ID of SuDS device in all 7 subcatchments (S01eS07).
S01 S02 S03 S04 S05 S06
BC ID 1 4 7 10 13 16
PP ID 2 5 8 11 14 17
GR ID 3 6 9 12 15 18
Journal of Environmental Management 238 (2019) 59–63
According to the zero-one model, different SuDS IDs in all 7 sub-
catchments are combined. For each quantity of SuDS devices, from 0 to
21, different combination and placement schemes are formed. The total
21
number of possible schemes is n = 0 C21 n
= 2,097,152.
Due to the large number of schemes generated by zero-one model,
scheme sampling should be carried out to save simulation time and
improve efficiency. In this study case, the maximum sample size for
each round (where round 1 is for no devices, round 2 for 1 device,
round 3 for 2 devices etc.) is calculated as 384, based on eq. (1) and a
confidence level of 95%,. Where the maximum possible number of
schemes for a given number of devices is less than 384, all are con-
sidered. This yields a total of 6608 samples and, therefore, 6608 SuDS
device combination and placement schemes.
Using MATLAB program and the SWMM urban drainage system
model, the 6608 different combination and placement schemes were
simulated and related assessment indicators CR and costs were calcu-
lated. The schemes were then filtered based upon the 80% CR re-
quirement.
For the filtered schemes, cost-effectiveness was analyzed and the
cost-effectiveness ratio (ER) was calculated using eq. (2). In order to
identify the most cost-effective scheme, the calculated results were
ranked from high to low by ER.
4. Results and discussion
Fig. 3 shows the CR and cost results for the 6608 combination and
placement schemes. When there are few SuDS devices (fewer than 9),
few schemes (only 0.24%) meet the 80% CR control target. As the
number of SuDS devices increases, the proportion of schemes meeting
the target gradually improves, but the cost also increases. For schemes
containing a given number of SuDS devices, the results of meeting the
target and the values of CR and cost are vary due to the different
combination and placement of the devices. In this study case, 1474
combination and placement schemes (22.3% of the 6608 evaluated)
met the control target.
Table 2 shows top 10 SuDS schemes, as ranked by their ER value
(lower ranked schemes are not presented due to the large number of
schemes evaluated). The value of ER is correlated with and cost and the
number of SuDS devices (r = −0.967 and −0.633 respectively). There
S07 is no clear correlation between ER and CR in a scheme (r = −0.483, and
|r| is less than the critical value 0.63 at the α = 0.05 level). Schemes
19
providing the same ER value can have different CR values and cost and
20
21 contain different numbers of SuDS devices, as illustrated by Scheme
Nos. 2503 and 2784. In this case, it may be preferable to select the
61
M. Wang, et al. Journal of Environmental Management 238 (2019) 59–63

Fig. 3. CR and cost of the 6608 combination and placement schemes. A minimum acceptable control level of CR threshold of 80% is shown by the red dotted line. (For
interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Table 2 NSGA-II or AMALGAM optimization algorithms with SWMM is 100

Ranking results of schemes (TOP 10) based on ER. population *100 generation = 10,000 SWMM simulations (Yang and
Scheme No. CR (%) Cost ( × 104 ER ( × 10−5) SuDS IDs
Chui, 2018), and 200 population (default) in MATLAB Optimization
CNY) Tool (Genetic Algorithm) for more than five variables. In this case
study, there are 6608 SWMM simulations using the proposed method -
1549 80.25 1354.16 59.3 2,5,8,14,17,18 approximately 33% fewer than required for the two optimization al-
1928 80.35 1400.97 57.4 2,5,7,8,11,14,20
2897 80.16 1411.58 56.8 5,6,8,11,12,13,15,17,20
gorithms. Each simulation may take 180 s with a one year rainfall time
3216 81.76 1448.98 56.4 3,4,5,8,12,14,15,17,19,20 series with one minute interval – therefore, this approach would save
2503 81.92 1470.91 55.7 5,8,10,14,15,16,17,20 19.6 h of computational time.
2784 80.55 1446.59 55.7 2,5,8,9,12,14,15,17,19 The case study area of this paper is relatively large due to it being at
2493 81.51 1474.48 55.3 5,6,8,14,16,17,20,21
the planning and design stage. From the point of view of the method
2193 81.60 1483.67 55.0 1,2,5,8,13,14,17,21
2636 80.34 1469.15 54.7 1,3,5,7,8,11,13,17,20 itself, there are no barriers to applying this method to small areas at the
2910 81.40 1491.67 54.6 5,8,11,12,14,15,17,18,21 community level; however, this would be subsequent to initial planning
… … … … … and design at a larger scale. Further verification work will be carried
out in small areas at the community level in future, providing addi-
tional support to the conclusions.
lowest cost scheme.
According to the ranking results based on ER, the proposed SuDS
combination and placement scheme is determined. In this study case, 5. Conclusions
the top 2 schemes are: No. 1549 scheme “2,5,8,14, 17, 18” and No.
1928 scheme “2,5,7,8,11, 14, 20”. These are illustrated in Fig. 4. In this paper, a new framework is used to determine the combina-
The framework provided here has the following features: tion and placement of individual devices in the design of sustainable
The approach for determining combination and placement of SuDS urban drainage systems (SuDS). The zero-one integer programming and
devices is based on the zero-one integer programming model, random random sampling techniques are utilized to generate and filter SuDS
sampling, solution filtering and cost-effectiveness analysis, and in- combination and placement schemes. In this study, 6608 schemes are
tegrated control effectiveness and cost assessment. It is flexible to simulated, based on a 95% degree of confidence, and 1474 solutions
choose/determine an appropriate scheme sampling size because the meet the control target (CR 80%). The final scheme is selected after the
method is based on the degree of confidence. At the same time the cost- cost-effectiveness analysis, which provided cost-effectiveness and ob-
effectiveness ratio is considered, so the feasibility of implementation of jectivity. The results indicate that the proposed approach is capable of
the selected scheme is supported. The method can be used to guide the guiding the combination and placement planning and design of SuDS
combination, placement and design of SuDS devices. devices, and better serves low impact development (LID) implementa-
An opinion-based approach places SuDS devices based on a com- tion.
bination of factors, including the opinions and experience of the de- It should be noted that other indicators, such as flooding volume,
signers. In the opinion-based approach, a limited number of potential flooding duration, rainwater utilization, pollutant indicators, etc. (in
schemes that are expected to meet the predefined goals are selected at addition to CR, Cost and ER) can also be incorporated in this metho-
the planning and design stage. However, these schemes represent only a dology. However, as the number of indicators considered increases
limited number of the possible cases (Hu et al., 2015), and the number determination of the preferred SuDS scheme would become more
of samples is far from enough to be sure of capturing the optimum. The complex because weights would need to be considered.
method proposed in this paper, therefore, is based on random sampling
of all combination schemes, which excludes subjectivity to some extent.
Due to considering a large number of potential schemes with the Acknowledgment
random combinations and zero-one integer programing, this assessment
of SuDS device combination and placement schemes is more compre- This work was supported by the National Natural Science
hensive and scientific than finite artificial schemes. Foundation of China (Grant No. 51208001); and the Key Program for
It also provides another way and thinking to determine SuDS device Overseas Academic Visitors of Outstanding Young Teachers at
combination and placement schemes, beyond the evolutionary opti- Universities in Anhui Province, China (Grant No. gxfxZD2016056)
mization algorithms that have been used previously (Lee et al., 2012; awarded to the first author.
Liu et al., 2016b; Zhang et al., 2013). The common configuration of

62
M. Wang, et al.
Fig. 4. Top 2 SuDS device combination and placement schemes.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.jenvman.2019.02.129.
References
Awumah, K., Bhatt, S.K., Goulter, I.C., 1989. An integer programming model for layout
design of water distribution networks. Eng. Optim. 15, 57–70. https://doi.org/10.
1080/03052158908941142.
Brase, C.H., Brase, C.P., 2012. Chapter 7: Estimation in Understandable Statistics:
Concepts and Methods, Tenth. ed. Brooks/Cole, Cengage Learning.
Butler, D., Digman, C.J., Makropoulos, C., Davies, J.W., 2018. Urban Drainage, Fourth.
ed. Taylor & Francis, CRC Press.
Chen, D.S., Batson, R.G., Dang, Y., 2010. Applied Integer Programming: Modeling and
Solution. John Wiley and Sons.
Chui, T.F.M., Liu, X., Zhan, W., 2016. Assessing cost-effectiveness of specific LID practice
designs in response to large storm events. J. Hydrol. 533, 353–364.
Diao, K., Sweetapple, C., Farmani, R., Fu, G., Ward, S., Butler, D., 2016. Global resilience
analysis of water distribution systems. Water Res. 106, 383–393. https://doi.org/10.
1016/j.watres.2016.10.011.
Hu, A., Ren, X., Ding, N., Tang, W., 2015. LID facilities layout and optimization in an area
in Shenzhen based on SWMM. China Water & Wastewater 31, 6–10.
Lee, J.G., Selvakumar, A., Alvi, K., Riverson, J., Zhen, J.X., Shoemaker, L., Lai, F.H., 2012.
A watershed-scale design optimization model for stormwater best management
practices. Environ. Model. Softw 37, 6–18. https://doi.org/10.1016/j.envsoft.2012.
04.011.
Liu, Y., Cibin, R., Bralts, V.F., Chaubey, I., Bowling, L.C., Engel, B.A., 2016a. Optimal
selection and placement of BMPs and LID practices with a rainfall-runoff model.
Environ. Model. Softw 80, 281–296.
Liu, Y., Theller, L.O., Pijanowski, B.C., Engel, B.A., 2016b. Optimal selection and place-
ment of green infrastructure to reduce impacts of land use change and climate change
63
Journal of Environmental Management 238 (2019) 59–63
on hydrology and water quality: an application to the Trail Creek Watershed. Indiana
Sci. Total Environ. 553, 149–163. https://doi.org/10.1016/j.scitotenv.2016.02.116.
Perez-Pedini, C., Limbrunner, J.F., Vogel, R.M., 2005. Optimal location of infiltration-
based best management practices for storm water management. J. Water Resour.
Plann. Manag. 131, 441–448. https://doi.org/10.1061/(ASCE)0733-9496(2005)
131:6(441).
Rossman, L.A., Huber, W.C., 2016. Storm Water Management Model Reference Manual
Volume III – Water Quality. National Risk Management Laboratory, U.S.
Environmental Protection Agency, Cincinnati, OH.
SMCHURD, 2015. Shanghai Spongy City Construction Technical Guidelines. Shanghai
Municipal Commission of Housing and Urban-Rural Development.
SMULRC, 2016. Shenzhen Spongy City Planning Key Points and Detailed Regulations.
Urban Planning Land and Resources Commission of Shenzhen Municipality.
Sweetapple, C., Fu, G., Farmani, R., Meng, F., Ward, S., Butler, D., 2018. Attribute-based
intervention development for increasing resilience of urban drainage systems. Water
Sci. Technol. 77, 1757–1764. https://doi.org/10.2166/wst.2018.070.
United States Environmental Protection Agency, 2018. Storm Water Management Model
Version 5.1.013 Help.
Wang, M., Sun, Y., Sweetapple, C., 2017a. Optimization of storage tank locations in an
urban stormwater drainage system using a two-stage approach. J. Environ. Manag.
204, 31–38. https://doi.org/10.1016/j.jenvman.2017.08.024.
Wang, M., Sweetapple, C., Fu, G., Farmani, R., Butler, D., 2017b. A framework to support
decision making in the selection of sustainable drainage system design alternatives. J.
Environ. Manag. 201, 145–152. https://doi.org/10.1016/j.jenvman.2017.06.034.
Yang, Y., Chui, T.F.M., 2018. Optimal sizing of roadside bioretention cell treatment train
for runoff volume control. In: 2018 International Spongy City Conference, pp.
485–493.
Zhang, G., Hamlett, J.M., Reed, P., Tang, Y., 2013. Multi-objective optimization of low
impact development designs in an urbanizing watershed. Open J. Optim. 2013,
95–108.
Zhen, X.Y.J., Yu, S.L., Lin, J.Y., 2004. Optimal location and sizing of stormwater basins at
watershed scale. J. Water Resour. Plan. Manag. 130, 339–347. https://doi.org/10.
1061/(ASCE)0733-9496(2004)130:4(339).