Water Research 232 (2023) 119720

Contents lists available at ScienceDirect

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

Assessing and optimizing the hydrological performance of Grey-Green

infrastructure systems in response to climate change and non-stationary
time series
Mo Wang a, b, Ming Liu a, Dongqing Zhang c, Jinda Qi d, *, Weicong Fu e, Yu Zhang a, Qiuyi Rao a, b,
Amin E. Bakhshipour f, Soon Keat Tan g
a
College of Architecture and Urban Planning, Guangzhou University, Guangzhou 510006, China
b
Architectural Design and Research Institute of Guangzhou University, Guangzhou 510499, China
c
Guangdong Provincial Key Laboratory of Petrochemical Pollution Processes and Control, School of Environmental Science and Engineering, Guangdong University of
Petrochemical Technology, Maoming, Guangdong 525000, China
d
Department of Architecture, National University of Singapore, 117575, Singapore
e
College of Landscape Architecture, Fujian Agriculture and Forestry University, Fuzhou 350002, China
f
Civil Engineering, Institute of Urban Water Management, University of Kaiserslautern, Kaiserslautern 67663, Germany
g
School of Civil and Environmental Engineering, Nanyang Technological University, 639798, Singapore

A R T I C L E I N F O A B S T R A C T

Keywords: Climate change has led to the increased intensity and frequency of extreme meteorological events, threatening
Urban stormwater management the drainage capacity in urban catchments and densely built-up cities. To alleviate urban flooding disasters,
Green infrastructure strategies coupled with green and grey infrastructure have been proposed to support urban stormwater man­
Coupled Grey-Green system
agement. However, most strategies rely largely on diachronic rainfall data and ignore long-term climate change
Climate change
Regional climate model
impacts. This study described a novel framework to assess and to identify the optimal solution in response to
Antecedent dry days uncertainties following climate change. The assessment framework consists of three components: (1) assess and
process climate data to generate long-term time series of meteorological parameters under different climate
conditions; (2) optimise the design of Grey-Green infrastructure systems to establish the optimal design solutions;
and (3) perform a multi-criteria assessment of economic and hydrological performance to support decision-
making. A case study in Guangzhou, China was carried out to demonstrate the usability and application pro­
cesses of the framework. The results of the case study illustrated that the optimised Grey-Green infrastructure
could save life cycle costs and reduce total outflow (56–66%), peak flow (22–85%), and TSS (more than 60%)
compared to the fully centralised grey infrastructure system, indicating its high superior in economic competi­
tiveness and hydrological performance under climate uncertainties. In terms of spatial configuration, the
contribution of green infrastructure appeared not as critical as the adoption of decentralisation of the drainage
networks. Furthermore, under extreme drought scenarios, the decentralised infrastructure system exhibited an
exceptionally high degree of removal performance for non-point source pollutants.

1. Introduction the hydrological safety of cities and waterlog (Larsen et al., 2016;
Nazari-Sharabian et al., 2019). Aerts et al. (2014) reported that urban
Rapid urbanisation leads to a significant increase of impermeable waterlog caused direct economic losses of $650 billion over the past
surfaces, resulting in increased surface runoff and thereby threatening decade. Conventionally, grey infrastructure (GR) is designed to

Abbreviations: CL-GR-GI, centralised Grey-Green infrastructure; CL-GR, centralised grey infrastructure; GR-GI, Grey-Green infrastructure; DL-GR-GI, decentralised
Grey-Green infrastructure; DL-GR, decentralised grey infrastructure; GA, genetic algorithm; GCM, Global Climate Model; GI, green infrastructure; GR, grey infra­
structure; LCC, life cycle cost; O&M, operation and maintenance; PV, present value; RCPs, Representative Concentration Pathways; RCM, Regional Climate Model;
SWMM, Storm Water Management Model; TSS, total suspended substance.
* Corresponding author at: School of Design and Environment, National University of Singapore, 117356, Singapore.
E-mail address: [email protected] (J. Qi).

https://doi.org/10.1016/j.watres.2023.119720
Received 13 August 2022; Received in revised form 22 January 2023; Accepted 4 February 2023
Available online 8 February 2023
0043-1354/© 2023 Elsevier Ltd. All rights reserved.
M. Wang et al. Water Research 232 (2023) 119720

intercept, collect and channel the runoff for effective drainage. How­ RCP8.5 scenarios (total radiative forcing topped at a stable 4.5 W/m2
ever, GR had the side effect of concentrating non-point source pollutants and 8.5 W/m2, respectively, by 2100 at the grid-scale, see Meinshausen
in the catchment area, resulting in extensive deterioration of the water et al. (2011)). El-Housni et al. (2019) made projection of the hydraulic
quality of the recipient water body (Nika et al., 2020; Wang et al., performance of the urban pipe network (2041 to 2070) based on the
2021a). Comparatively, green infrastructure (GI) with natural-based simulation results of the Canadian Regional Climate Model. However,
solutions, including Low Impact Development, Water Sensitive Cities, the majority of the reported studies assessed and optimised the adaptive
Sustainable Urban Drainage Systems, and Sponge City could counter the capacity of GR and GI in response to stationary or event-based climate
pollution issues (Fletcher et al., 2015; Ying et al., 2021). These GIs are scenarios, neglecting the effects of non-stationarity and nonlinearity of
typically incorporated in decentralised structural practices, and manage long-term climate change. Since climate change is dynamic and
the surface runoff at the source. As they are typically designed to mimic nonlinear, optimising the synergistic effects of GR and GI as the means to
the natural hydrological condition before urbanisation, and are adapt to non-stationary climate change needs to be addressed urgently.
aesthetically pleasing, GIs could be strategically integrated as part of the The relevant long-term time series parameters are readily influenced
urban stormwater management scheme and the city’s development plan by climate change. For example, Noor et al. (2022) reported potential
(McFarland et al., 2019; Nickel et al., 2014; Versini et al., 2018). projected changes in rainfall intensity, attributable to climate change, of
Although the advantages of these strategies have been widely discussed − 3.4 − 26.7% (2010− 2039), − 0.1 − 34.5% (2040− 2069) and − 4.3 −
and accepted, GI is recognized as being appropriate for stormwater 96.8% (2070− 2099) from that observed during the base period
management for rainfall of short return periods and in many cases (1971− 2000). Therefore, forecasting reliable long-term time series data
cannot replace the role of GR, especially in the case of densely built-up under various climate changes is also a vital component for accurate
urban environments (Xu et al., 2019). In such an urban environment, GR-GI optimisation. Global Climate Model (GCM) could be used to
over 60% of the land surface is occupied by pavements and roofs and is simulate the data of air temperature, precipitation, and atmospheric
drained through grey infrastructures (Qi et al., 2019). The remaining circulation on a global scale, and is considered to be an essential tool for
open space limits the scope and hydrological capacity of GI in response the projection of climate change (IPCC, 2014; O’Neill et al., 2016).
to extreme storms. For example, Mei et al. (2018) assessed 15 GI sce­ However, the GCM is appropriate at a large scale, due to certain defects
narios subject to rainfall with return periods of 2–100 years in Beijing, and uncertainties of simulation data in geographic space (Willems et al.,
China, and concluded that the hydrological performance of the GIs was 2012). Regional Climate Model (RCM), a downscaling model driven by
not satisfactory and could not eliminate urban flooding even for the case output from a GCM, can be applied to capture topographic and surface
of the most optimistic scenario. Although GI can benefit stormwater characteristics at local minor scales, as well as deviations of boundary
management in multiple ways, including water quality, volumes and conditions, providing parameterisation with observed cumulus, which
frequency of runoff, and floods, one should not expect to achieve high in turn facilitated better simulation of extreme rainfall events (Giorgi
performance in flooding mitigation, particularly for long return periods, and Mearns, 1999). RegCM4.6 (a common RCM) developed by the Na­
by just applying GI only. tional center for Atmospheric Research and Pennsylvania State Uni­
GR and GI could be integrated appropriately and applied as potential versity has been used to perform bias correction on each grid point and
advanced solutions for urban stormwater management (Browder et al., showed high reliability and excellent predictive effect (Gao et al., 2017;
2019). The integrated system leverages on the versatility and ecological Pan et al., 2020; Zhang et al., 2021). Several studies had made use of
aspects of GI and the engineering reliability and practicability of GR these models to perform comparative study of the results of present and
(Wang et al., 2021d, 2017b; Yang and Zhang, 2021; Zhou et al., 2022). future scenarios. However, only a handful of studies highlighted the
To date, several integration schemes have been reported and they dynamic changes in the climate change process (Yu et al., 2022). To
include the Grey-Green infrastructure (GR-GI) system to reduce the life better capture the dynamic changes, some researchers proposed using
cycle cost (LCC) (Bakhshipour et al., 2019a); the System for Urban the Bernaola-Galván algorithm to identify mutation information in
Stormwater Treatment and Analysis Integration to evaluate the optimal long-term time series data. The Bernaola-Galván algorithm is a seg­
location, type, and cost of stormwater best management practices (Lee mentation algorithm originally used to “probe the temporal organisation
et al., 2012); and integrated GR and low impact development to measure of heterogeneities in human heartbeat interval time series”
the technical and operational resilience of best management practices (Bernaola-Galván et al., 2001). Since the algorithm could be used to
(Wang et al., 2021c). Other reported works include assessment of hy­ detect mutations in the climate change process, it might also be used to
draulic reliability and hydrological resilience based on the selected capture nonlinear and non-stationary climate change information (Chen
rainfall events instead of continuous rainfall time series (Da Silva et al., et al., 2019; Jehanzaib et al., 2020; Yu et al., 2022), and provide insights
2018; Mohanty et al., 2020; Sohn et al., 2019; Wang et al., 2020). A on parameters at various stages of climate change. Hence, knowing the
long-term time series analysis for urban stormwater management would temporal variation or time series of the climate change, one would also
have considered multiple climatic variables, such as rainfall, intensity, be able to establish the time series of the various parameters. In short,
duration, and antecedent dry days, to mention a few, as these parame­ both simulation models and segmented time series are crucial to
ters could influence the generation of runoff in the catchment, soil water obtaining realistic and reliable long-term time series parameters for
storage capacity of the underlying surface, and source-accumulation accurately assessing and optimizing the GR-GI performance under
process of non-point source pollutants (Madarang and Kang, 2013; various climate change scenarios since the performance varies signifi­
Rossman and Huber, 2016; Salim et al., 2019). In adopting a single cantly across the time series parameters, such as precipitation amount,
climatic variable (e.g., rainfall intensity) the hydrologic performance of intensity and interval time.
the integrated GR-GI drainage system may be overestimated under This paper presents a novel assessment framework to support the
certain hydrological conditions (Yu et al., 2022), not to mention the selection of GR-GI based on long-term time series information in tandem
inherent limitation of using one single parameter to optimise the loca­ with non-stationary climate change. The specific objectives are: (1) to
tion, type, and cost of urban drainage systems for long-term integrate RegCM 4.6 and Bernaola-Galván algorithm into establishing
applications. the long-term time series of parameters in tandem with non-stationary
Recently, climate models have also been used to assess the hydraulic climate changes. The time series will be used to test the optimised GR-
performance of the GR or GI under various climate scenarios, such as GI design and configuration; (2) to optimise the life cycle cost and hy­
Representative Concentration Pathways (RCPs), to help understand drological performance of GR-GI using the graph-theory-based algo­
better the potential impact of design/planning decisions on urban hy­ rithm and genetic algorithm (GA); and (3) to optimise GR-GI under
drology. Zeng et al. (2021) evaluated the performance of GI in runoff various climate scenarios in Guangzhou, China.
quantity control and water quality improvement under RCP4.5 and

2
M. Wang et al. Water Research 232 (2023) 119720

2. Methodology monsoon climate. The mean annual precipitation is about 1700 mm but
the rainfalls are non-uniformly distributed. Also, Guangzhou is one of
The developmental framework consists of three parts (Fig. 1): (1) the cities with the most waterlogged and severe flood risk amongst 136
Assess and process the climate information to establish the appropriate large coastal cities (Hallegatte et al., 2013). Covering an area of 101.8
climate change time series; (2) establish the optimal drainage design hectares, Zhujiang New Town (Fig. 2) is the most prominent central
strategy; and (3) evaluate the hydrological performance of the drainage business district in Guangzhou, and has the typical characteristics of a
system in response to climate change. The long-term time series of densely built-up city such as high impervious surface, high population
precipitation is obtained by using dynamic downscaling of RegCM based density, and highly developed economy. This study selected Zhujiang
on two RCPs. As for the spatial configuration of urban stormwater New Town as the case study area. Zhujiang New Town consists of 43
infrastructure, the optimal strategies at various levels of decentralised sub-catchments. The Storm Water Management Model (SWMM), an
layouts are obtained through integrating graph theory and genetic al­ open-access hydrological model developed by United States Environ­
gorithms. Then the hydrological performance of the recommended mental Protection Agency, was used to simulate the surface runoff, peak
strategies for various representative climate change scenarios is evalu­ flow, and the total suspended solids (TSS) in an urban catchment
ated using a hydrological model. (Rossman, 2010). The parameters of the sub-catchments in SWMM were
obtained from the Wang et al. (2021c), and are listed in Table S1. The
dynamic wave approach was used for the network confluence compu­
2.1. Case study
tation model, and the Horton model was used to describe the infiltration
process. Appropriate saturation and exponential function were adopted
Guangzhou is located at the Pearl River Basin and has a subtropical

Fig. 1. The framework for hydrological assessment of the optimal GR-GI at various levels of decentralised layouts. Note: GCM - Global Climate Model; LCC - life cycle
cost; GR - grey infrastructure; GI - green infrastructure; and TSS - total suspended substance.

3
M. Wang et al. Water Research 232 (2023) 119720

Fig. 2. The case study area – Zhujiang New Town in Guangzhou, China.

for pollutant accumulation and the wash-off model, respectively. (Dickey and Fuller, 1979). The Augmented Dickey-Fuller is used to
test a time series by detecting the existence of unit root in a time series
sample, so as to ensure that the premise of Bernaola-Galván algorithm is
2.2. Climate information - processing and assessment met. The corresponding procedures of applying Bernaola-Galván algo­
rithm are included in the annex (Text S1).
2.2.1. Climate data simulation
GFDL-ESM-2 M,which had been found to perform well for the Pearl 2.2.3. Data extraction and analysis
River Basin of China (Zhang et al., 2021), was chosen to simulate climate Three sets of data: precipitation, duration, and antecedent dry days
data in this study. The simulation results were used as the initial and have been found significant and relevant in the generation and analysis
lateral boundary conditions for climate projection in this study. of climate change time series for the purpose of drainage design (Gong
RegCM4.6 was adopted for dynamic downscaling (0.25◦ × 0.25◦ ). et al., 2016; Rasheed et al., 2019; Sohn et al., 2019). They were extracted
RCP4.5 and RCP8.5 were selected as climate scenarios as they may be from the simulated long-term time series. A rainfall event is defined as
used to represent similar intermediate and extreme scenarios for climate one with a rainfall depth exceeding 1 mm, and the minimum interval
change, respectively, and had been widely utilised and reported in many between two rainfall events is 6 hr or more (Joo et al., 2014). The
studies (Requena et al., 2019; Yu et al., 2022; Zeng et al., 2021). The probability density function used to describe the variation and distri­
base period (2008–2021) was used, and for a 30-year life cycle of the bution of rainfall is based on a two-parameter gamma distribution
coupled GR-GI system, the target future period was set for 2022–2051. (Chapman, 1997) (Eq. (1)).
− x

2.2.2. Managing the time series – stage division of data xα− 1 e β

f (x) = (1)
Climate change is a change process with phase regularity. Reliable βα Γ(α)
assessment of the projected climate changes is helpful to identify and 2 2
explore the effects of meteorological factors on the performance of Grey- where α = μσ2 is the shape parameter, β = σμ is the scale parameter, and μ
Green coupled strategy. The Bernaola-Galván algorithm has often been is the mean value of the precipitation, duration or antecedent dry days
used to identify mutation information in long-term time series and has series and σ is the variance.
been found applicable to nonlinear and non-stationary climate change The joint probability distribution function may be used for fitting the
information (Chen et al., 2019; Jehanzaib et al., 2020; Yu et al., 2022). distribution relationship of the two factors (Yu et al., 2022), contrib­
In this study, the Bernaola-Galván algorithm was chosen for stage di­ uting to observing the distribution of precipitation with duration and
vision of the future period. It should be noted that before applying the antecedent dry days from a broader dimension (Eq. (2)).
Bernaola-Galván algorithm, an Augmented Dickey-Fuller test is required
to verify the non-stationarity of the time series to be sub-divided

4
M. Wang et al. Water Research 232 (2023) 119720

− x − y
xαx − 1 yαy − 1 e βx + βy parameters would be hard put given our available resources. After much
f (x, y) = f (x)f (y) = ( ) (2) deliberation, we compromised by adopting an alternative approach in
βαx x βay y Γ(αx )Γ αy
the formulation of the algorithm, and had chosen to use certain discrete
where f(x,y) is the joint probability distribution of the precipitation and design rainfall event for the optimisation instead.
duration, or precipitation and antecedent dry days, where f(x) is the
gamma distribution of precipitation, and f(y) is the gamma distribution 2.3.2. GR strategy optimisation
of duration or antecedent dry days. Initially, the spatial configuration algorithm based on a graph theory
The joint probability density distribution of long-term time series is utilised to search for various decentralised layouts of GR, that is ex­
was employed to assess the characteristics of climate change in different pected to provide suitable and hydraulically adequate basic networks for
stages under various climate scenarios. GR-GI (Bakhshipour et al., 2019a). This algorithm can generate all
possible sewer spatial configurations with various degrees of decen­
2.3. Optimal strategy development tralisation from a base plan (drainage network follows the street align­
ments). It is a connected cyclic graph that includes all drainage
2.3.1. Optimisation objectives and constraints possibilities. In the base plan, vertices represent manholes and edges
In general, the calculation function used for searching the optimal joining the vertices represent sewers. According to Bakhshipour et al.
minimum-cost of a coupled system is expressed as Wang et al. (2021c): (2019a), the first step is to define the degree of decentralisation of the
layout, which is determined by the number of selected and candidate
dopt = argmin[f cost ] (3) outlets across the system:
d∈D

NSO − 1
where, d is the decision variable; D represents all feasible plans for the DDL = × 100% (7)
NCO − 1
system; and dopt is the optimal solution. When GI is included in the
optimisation, d may be extended as Eq. (4): where NSO denotes the selected number of outlets, and NCO represents
d = [DDL, layout parameters, hydraulic parameters, GI parameters] (4) the total number of candidate outlets. The main objective of the study is
to develop a replicable approach for the assessment of GR-GI in response
where, the decision variable d contains the degree of decentralisation of to non-stationary climate change and demonstrate its feasibility and
the drainage layout (DDL), layout parameters (i.e., connections between application. Either fully centralised, i.e., DDL=0%, and fully decen­
different sewers), hydraulic parameters (i.e., pipe diameters and slopes tralised network, DDL=100% are investigated in this study.
that satisfy the hydraulic constraints), and GI parameters of the coupled The optimisation process follows: first a fully centralised layout is
system (type, size, and location of GI). selected (only one outlet out of NCO possible outlets). This is the fully
To characterise the minimum-cost quantitatively, LCC expressed in centralised network (CL-GR). Then a second outlet from the remaining
present value (PV), is an important indicator of the capital urban (NCO – 1) outlets is selected randomly and added to the first network.
infrastructure investment, and operation and maintenance (O&M) cost Including a second outlet means that the first network would be parti­
of the infrastructure Chui et al., 2016; Dos Santos et al., 2021). The tioned to form two sub-networks, each discharges through the respective
optimal strategies at various levels of decentralised layouts for the least outlet. The configuration of the sub-network is selected by using the
LCC is expressed as Eqs. (5) and ((6): algorithm which uses graph theory to assign different parts of the
network to different outlets and generates the decentralised layout.
LCC = CapitalInfrastructure + PVO&M− Infrastructure (5) Then the process is repeated by adding a third, fourth, and fifth outlets
etc. until all NCO outlets are included, and the corresponding sub­
∑
n
PVO&M− Infrastructure = O&M
1
(6) networks were configured using the above mentioned algorithm. The
1
(1 + i)n algorithm needs 2 × Ns decision variables to generate one feasible
layout. Ns is the number of sub-catchments in the base graph. After
where, PVO&M is the PV cost of O&M during the life span; i is the discount generating the layout, the size of the GR network components (e.g., pipe
rate, which is taken as 2% (Dong, 2018); n is the lifespan, and which is diameters and manholes invert elevation) must be designed in such a
taken as 30 years according to Xu et al. (2017). In this study, the capital way that satisfies all hydraulic and technical criteria. For this purpose,
costs were calculated based on local engineering market, and the annual the adaptive sewer design algorithm introduced by Haghighi and
O&M costs of GR and GI in Guangzhou were taken as 10% and 8% of the Bakhshipour (2012) was adopted in this study. The constraints are the
capital cost, respectively, according to Houle et al. (2013). telescopic pattern of sewer diameters, minimum and maximum cover
In addition, we include hydraulic reliability as a constraint in the depths, slopes, flow velocity and hydraulic reliability. Each alternative
optimisation process according to local outdoor drainage design stan­ layout requires 2 × NP decision variables in the genetic algorithm; here
dards (MHURD, 2016), that is, no flooding occurred following the NP represents the total number of pipes. For each pipe, one variable
design rainfall. The design rainfall selected was a 5-yr return period and determines the pipe’s diameter, and other one determines pipe’s slope
6-hr duration, which was 121 mm rainfall depth in Guangzhou. The (by adjusting the invert level of the manhole). Therefore, the total
internal characteristics followed that of a Chicago hyetograph. The number of decision variables for GR strategy and optimisation (gener­
current engineering practice is such that the design of GR is based on a ating a decentralised layout and sizing it) is 2 × (NP+Ns).
specific design rainfall (certain return period and duration). The per­
formance of the so designed infrastructure is subsequently in the 2.3.3. Grey-Green infrastructure system optimisation
assessment of the overall system subject to certain selected events. There Considerable investment is required to construct and maintain the
are obvious advantages to evaluate the system performance based on a conventional GR, and the introduction of GI may somewhat decrease the
series of events, and the ultimate test would be based on a time series of construction demand for GR, and thus reduce the investment as the cost
hydrological and climate change incorporating various relevant pa­ for construction and maintenance of GR-GI is relatively lower (Wang
rameters. Given the nature of the hydrological and climate change et al., 2021c). The general constraints for configuring a Grey-Green
events, the list of parameters would be large and almost always “not infrastructure system include the street network (the sewer pipe runs
exhaustive enough” to describe all aspects of the change. We discovered parallel and at the side of the streets), ground elevation, obstacles, dis­
that an exorbitant amount of computation hardware and software re­ tance of receipient water body from the outlet(s), and space available for
sources is required to perform optimisation for discrete events with a the construction. Moreover, considering the actual construction are
short list of carefully selected parameters. Using a time series of selected

5
M. Wang et al. Water Research 232 (2023) 119720

located in a high-density city space, the construction scale of GI needs to where TOn and RPn are the ratios of total outflow and peak flow,
be controlled to no larger than 10% of the sub-catchment areas (Eckart respectively; Similarly, the total suspended solids (TSS), a reflection of
et al., 2017). water quality in the recipient water body (Bilotta and Brazier, 2008;
Being one of the most popular types of GI, a bioretention cell com­ Wang et al., 2018, 2017a) may be treated in the same manner:
prises surface vegetation, graded soil, pore space storage, and under­
MRemoval
ground drainage layers, and is widely and flexibly used in high-density RTn = × 100% (12)
MBuildup
cities where open space is scarce (Lim and Lu, 2016; Wang et al., 2021b).
For most practical purposes bioretention cells are mainly designed to where, RTn is the removal rate of TSS, and Mremoval and Mbulidup present
reduce the amount and delay the flow of surface runoff, remove the removed and accumulated mass of TSS, respectively.
non-point source pollutants, and beautify the environment (Liu et al.,
2015). The parameters of the bioretention cell module in SWMM can be 3. Results and discussion
found in Table S2.
After optimising GR networks as explained in Section 2.3.2, GI can be 3.1. Climate simulation results
introduced to the system to formulate coupled GR-GI configuration.
Here, a binary system is used to annotate GA that is utilised to optimise RCP4.5 and RCP8.5 from 2008 to 2051 were adopted to represent the
the coupled system and hence determine the minimum LCC (Ene et al., annual precipitation for the two scenarios. Both time series showed
2016). In this process, the decision variables are encoded as binary upward rising trend, and the upward trend of RCP8.5 (coefficient of
values (i.e., 0 and 1), and a “chromosome-equivalent” composed of these variation = 0.3) was more prominent than that of RCP4.5 (Fig. 3a and
values is used to represent one design scheme (Hsu et al., 2005). In the 3b). These characteristics were consistent with the local future rainfall
present study, there are NS (number of sub-catchments) + 2NP (number predictions (Wu and Huang, 2016; Zhang et al., 2021). Compared with
of pipes) values in each chromosome to reflect the decision variables, RCP8.5, more events were observed in RCP4.5, and showed higher
where NS values control GI in each sub-catchment i and each value is amount and intensity of rainfall (Fig. 3c).
decoded as follow: Based on the results of the Augmented Dickey-Fuller test, the long-
{
1→Subcatchment i has GI term series data were shown to be non-stationary (P = 0.768). The
GIi = (8) Bernaola-Galván algorithm was also used to identify the nodes between
0→Subcatchment i has no GI
the prediction stages (2022–2051). Based on RCP4.5, the future mete­
As GI is introduced into the GR layout, the GR diameter could be and orological stage was sub-divided into two stages: stage 1 (2022–2028, n
will be reduced to achieve minimum LCC, while hydraulic reliability is = 977) and stage 2 (2029–2051, n = 3187); In the case of RCP8.5, the
held at the same level as before. The pipe diameter (D) will be controlled two stages were 2022–2031 (n = 1161) and 2032–2051 (n = 2225).
by 2NP values, represented by a two bits code and define the relative Even though the RCP8.5 scenario contained less rainfall, the annual
diameter of a pipe (j): average rainfall increased more than those of the base period (7% in
⎧
⎪ 11→D = Same as in the GR scheme stage 1 and 17% in stage 2).
⎪
⎨
10→D = One size smaller than in the GR scheme Based on the statistics of independent events obtained (Table 1), an
Dj = (9) average rainfall event during stage 1 in RCP4.5 was 13.9 mm, while the
⎪
⎪ 01→D = Two sizes smaller than in the GR scheme
⎩
00→D = Three sizes smaller than in the GR scheme average in stage 2 was 15.0 mm, while the independent rainfall in
RCP8.5 remained at about 8.7 mm on average. In terms of rainfall
To obtain the minimum cost configuration, these pipe diameters
duration, each stage in RCP4.5 remained at about 13.0 h, while the
need to be iterated through GA to minimize the LCC. Finally, the cen­
corresponding duration for each stage of RCP8.5 was close to 10 h. The
tralised Grey-Green infrastructure (CL-GR-GI) system and decentralised
longer rainfall duration means higher probability of higher rainfall
Grey-Green infrastructure (DL-GR-GI) system alternatives are generated
amount. One may conclude that the frequency of rainstorms was higher
based on the two optimal GR layouts (centralised and fully decentral­
based on the RCP4.5 scenario. Furthermore, the frequency for rainfall
ised) with the corresponding DDL.
exceeding 50 mm increased from 4.7% observed during the base period
to 6.3% in stage 2, and the frequency of rainfall exceeding 100 mm in­
2.4. Hydrological performance assessment creases from 1.6% observed during the base period to 2.3% in stage 2. A
downward trend in the frequency of heavy storms in RCP8.5 was
The ability to control water quantity and quality are two critical observed. It can be seen in Table 1 that the frequency for rainfall
aspects of urban stormwater system assessment (Islam et al., 2021; exceeding 50 mm decreases from 2.2% observed during the base period
Rezaei et al., 2021). As for water quantity, the reducing ratios of total to 1.7% in stage 2, and the frequency of rainfall exceeding 100 mm
outflow and peak flow were chosen as the indicators in this study, which decreases from 0.5% observed during the base period to 0.4% in stage 2.
had been used hydrological performance assessment (Gong et al., 2019; Significant changes of antecedent dry days were observed in RCP8.5,
Jiang et al., 2017; Xing et al., 2016). Here, total outflow refers to the full indicating a more frequent arid climate. Compared to that observed
amount of the final outflow from the end of the network (sum of all during the base period, the antecedent dry days of stage 1 and stage 2
outlet flow); and peak flow refers to the maximum outflow from the end increased by 39.1% and 51.5%, respectively in RCP8.5, with the average
of the drainage system over the whole period. antecedent dry days increased from 1.9 days to 2.9 days.The maximum
The strategy with the highest cost amongst the four optimised stra­ antecedent dry days could reach 150 days. In contrast, the antecedent
tegies was taken as the benchmark. The ratios of the total outflow and dry days of RCP4.5 were relatively constant, showing a mean of 2.1
peak flow relative to the corresponding values for the benchmark, were days. This observation was in line with that reported by Wu and Huang
used as evaluation indexes for the optimised strategy, and the expres­ (2016) and Shi et al. (2020).
sions are as follows: The joint probability distributions of the precipitation amount and
(
Vtotaln
) duration, and the precipitation amount and antecedent dry days are
TOn = 1 − × 100% (10) shown in Fig. 4. In the case of RCP4.5, most of the rainfall events which
VtotalBenchmark
exceed 100 mm are accompanied by long duration (Fig. 4a). Thus, the
( )
Vpeakn overall rainfall intensity was not high. On the other hand, the duration of
RPn = 1− × 100% (11) rainfall events was shorter in RCP8.5, indicating higher rainfall in­
VpeakBenchmark
tensities, especially so for extreme rainfall events. Since urban waterlog

6
M. Wang et al. Water Research 232 (2023) 119720

Fig. 3. Stage division and rainfall events abstraction. (a) Annual precipitation and stage division of RCP4.5, (b) Annual precipitation and stage division of RCP8.5;
and (c) Distribution of rainfall events at various stages of selected scenarios.

Table 1
The statistics of climatic characteristics at each stage.
Climate change Scenarios RCP4.5 RCP8.5
Base Period Stage 1 Stage 2 Base Period Stage 1 Stage 2

Average number of events per year 140 140 (− 0.6%) 139 (− 1.3%) 156 116 (− 25.7%) 111 (− 28.8%)
Antecedent dry days Mean (days) 2.1 2.0 (− 2.7%) 2.1 (+0.4%) 1.9 2.6 (+39.1%) 2.9 (+51.5%)
Max (days) 81.5 49.5 (− 39.3%) 112.5 (+38.0%) 112.0 130.8 (+16.7%) 149.8 (+33.7%)
Precipitation Mean (mm) 12.8 13.9 (+8.3%) 15.0 (+17.1%) 8.1 8.0 (− 1.9%) 8.0 (− 2.4%)
Max (mm) 353.4 740.3 (+109.5%) 560.1 (+58.5%) 184.0 213.9 (+16.3%) 262.7 (+42.8%)
x ≤ 50mm n 134 132 (− 1.2%) 130 (− 3.0%) 153 113 (− 25.8%) 109 (− 28.4%)
Ratio 95.3% 94.8% (− 0.6%) 93.7% (− 1.7%) 97.8% 97.6% (− 0.2%) 98.3% (+0.5%)
50<x mm n 7 7 (+10.9%) 9 (+33.0%) 3 3 (− 18.3%) 2 (− 46.0%)
Ratio 4.7% 5.2% (+11.6%) 6.3% (+34.8%) 2.2% 2.4% (+9.9%) 1.7% (− 24.2%)
100<x mm n 2 2 (− 16.1%) 3 (+43.3%) 1 1 (− 30.0%) 0 (− 37.0%)
Ratio 1.6% 1.3% (− 15.6%) 2.3% (+45.3%) 0.5% 0.4% (− 5.8%) 0.4% (− 11.5%)
Duration Mean (h) 13.0 12.9 (− 0.5%) 13.4 (+3.7%) 9.7 10.6 (+9.6%) 10.4 (+7.3%)
Max (h) 132.0 102.0 (− 22.7%) 156.0 (+18.2%) 78.0 84.0 (+7.7%) 84.0 (+7.7%)

is often triggered by rainstorms with high rainfall intensity, therefore, and intense in the study area, which is also consistent with the precip­
we focused on rainfall events with high precipitation (Fig. 4b). It is noted itation in the subtropical monsoon region. According to RCP8.5, ante­
that high precipitation typically coincides with shorter antecedent dry cedent dry days showed a longer duration and hence more frequent
days. From Fig. 4b, it can be seen that rainstorms are usually continuous drought events. However, the antecedent dry days of extreme rainstorm

7
M. Wang et al. Water Research 232 (2023) 119720

Fig. 4. The probability density distribution of rainfall events during various stages of RCP4.5 and RCP8.5. (a) the probability density distribution of precipitation and
duration; and (b) the probability density distribution of precipitation and antecedent dry days.

events appeared an exception as the antecedent dry days were relatively optimised coupled GR-GI system is more attractive than fully GR system
small. Isolated rainstorm mode appeared not changed significantly. in terms of the cost (Fig. 6). Specifically, it is clear that at the same
degree of decentralisation, the LCC savings of a fully centralised and
decentralised GR-GI systems are 18% and 11% compared to the fully GR
3.2. Optimum drainage configuration
system at the same degree of decentralisation, respectively. This
observation also shows that GI has the potential of replacing a signifi­
The optimised configuration of GR and GR-GI at various degrees of
cant proportion of the centralised layout, and lead to significant cost
decentralisation are shown in Fig. 5. Each configuration were run
reduction. Compared to a centralised GR-GI, the decentralised solution
through 320,000 iterations to obtain the optimum design, a process that
produced an additional 16.2% saving of the LCC. This is because of the
took 30 h based on a PC equipped with Intel Core i5, 2.9 GHz dual-core
relatively low cost associated with a highly decentralised layout (Wang
CPU and 16 GB of RAM. Fig. 5 shows the pipe diameter and the corre­
et al., 2021c). It is also noted that large scale GI construction is typically
sponding pipe slope reflected as burial depth of the coupled system. The
not required, which further compresses the construction investment of
pipe diameters are significantly smaller than that designed for the single
the decentralised GR-GI. Therefore, an optimised and fully decentralised
grey system. In the coupled system, the GI area in a fully centralised GR-
Grey-Green design is most economical and competitive while satisfying
GI (CL-GR-GI) accounts for 6.6% of the drainage area, while the area in a
the hydraulic reliability requirement.
fully decentralised GR-GI (DL-GR-GI) accounts for 5.8% of the catch­
ment area. A CL-GR-GI design contained a larger proportion of GIs. An

8
M. Wang et al. Water Research 232 (2023) 119720

Fig. 5. Spatial allocation of optimised strategies (Wang et al., 2022):. (a) the spatial allocation of optimised CL-GR; (b) the spatial allocation of optimised DL-GR; (c)
the spatial allocation of optimised CL-GR-GI; and (d) the spatial allocation of optimised CL-GR-GI.

3.3. Hydrological performance retained at source, and the outflows were not affected significantly with
climate change. This result demonstrates that as an annualised index,
The hydrological performances of the optimised strategies were reduction of annual outflow is not sensitive with climate change. Similar
evaluated using a catchment-runoff model, SWMM, and included con­ results have been reported by Kim et al. (2018) and Yang and Zhang
siderations of reducing the outflow, peak flow, and TSS. The simulation (2021).
was calculated for 12 h simulation time. Fig. 7b shows that DL-GR-GI achieved the most effective perfor­
Since the LCC of the centralised GR is the least ideal, it was used as a mance in reducing peak flow, with a score between 71% and 85%, which
benchmark strategy to evaluate the systems’ performance of runoff is significantly different from the outflow discharge. It is worth noting
volume. Based on the findings of the simulation, it is found that, with that the performance of decentralised GR was second only to DL-GR-GI
regards to outflow discharge, GR-GIs could achieve reduction ratios of and was significantly higher than CL-GR-GI. The reason is the strong role
56− 66% (Fig. 7a). In terms of hydrological performance, the CL-GR-GI of decentralised networks in attenuating peak flow (Bakhshipour et al.,
system scored the best. As a result of the configuration of larger-scale GI 2019b; Wang et al., 2021c). There is also the added positive character­
implemented in CL-GR-GI, most of the light rainfalls were effectively isrtics of a decentralised layout with multiple outlets which could

9
M. Wang et al. Water Research 232 (2023) 119720

Fig. 6. Life cycle cost of various optimisation strategies (Wang et al., 2022).

Fig. 7. Performance of long-term time series simulation of each strategy compared with CL-GR for each stage of RCP4.5 and RCP8.5. (a) The reducing ratios of total
outflow; (b) The reducing ratios of peak flow; and (c) The reducing ratios of TSS.

effectively moderate the impacts of the discharge even when the storage retention property of GI on TSS. In GR-GI, the removal efficiency of
capacity of GI is reached. Meanwhile, there is observed that peak flow TSS is unstable under RCP8.5. In RCP8.5, antecedent dry days are
reduction in DL-GR was generally maintained at 65% for various climate increased significantly, potentially results in increased TSS accumula­
scenarios. In comparison, the capacity for peak flow reduction in the tion in the catchment. Consequently the removal efficiency of TSS in GR-
case of GR-GIs was highly sensitive to climate change. In the case of GI would decrease. While a larger scale GI is anticipated to retain the
RCP4.5, the performance of GR-GI decreases to certain degrees from TSS, intriguingly, the DL-GR-GI with lesser extent of GI was observed to
stage 1 to stage 2, and the reducing rate of CL-GR-GI decreases from 50% show better performance in removing TSS. Plausibly the decentralised
to 22%. This is primarily due to more intense extreme rainfall events spatial topology may have increased the contact opportunities between
being observed during stage 2. The hydrological performance of runoff and GI, leading to better retention of TSS in the catchment.
CL-GR-GI was less favourable as the GI portion of the drainage system
does not perform well during extreme rainfall. Similar results have been
obtained in studies of independence GI performance (Bae and Lee, 2020; 3.4. Limitations and future work
Hu et al., 2019). The centralised layout lacks sufficient resilience, and GI
is usually only used to manage runoff at source. When extreme events Certain limitations of this study could be improved and carried out in
generate more runoff than the upper retention limit, GI basically fails to future work. First, with the updating of GCMs, more possibilities for
limit flood peaks (Zeng et al., 2021). climate change have been proposed, which may not be able to be fully
In terms of the removal rate of TSS, the GR-GIs depicted a rate of covered by a single climate model. In the future, the ensemble of multi-
more than 60% (Fig. 7c), which is largely dependent on the pre- climate models can be considered to provide a varied and reliable pre­
diction of the non-stationarity of future climate change and to achieve a

10
M. Wang et al. Water Research 232 (2023) 119720

more reasonable layout optimisation. Second, the simulation process is References

time-consuming. An improvement of the optimisation algorithm could
be carried out to refine the search for an optimum solution. Third, this Aerts, J.C.J.H., Botzen, W.J.W., Emanuel, K., Lin, N., de Moel, H., Michel-Kerjan, E.O.,
2014. Evaluating flood resilience strategies for coastal megacities. Science 344
study employed GA optimisation and focused on economic performance (6183), 472–474.
in the optimisation, which may not be comprehensive and adequate for Bae, C., Lee, D.K., 2020. Effects of low-impact development practices for flood events at
addressing real-life applications with multiple objectives, such as more the catchment scale in a highly developed urban area. Int. J. Disast. Risk Res. 44.
Bakhshipour, A.E., Bakhshizadeh, M., Dittmer, U., Haghighi, A., Nowak, W., 2019a.
non-point source pollution indicators (e.g., Nitrogen compounds and Hanging gardens algorithm to generate decentralized layouts for the optimization of
Chemical Oxygen Demand). More advanced optimisation algorithms urban drainage systems. J. Water Res. Plan. Man. 145 (9).
could be developed and used to tackle multi-objective problems in the Bakhshipour, A.E., Dittmer, U., Haghighi, A., Nowak, W., 2019b. Hybrid green-blue-gray
decentralized urban drainage systems design, a simulation-optimization framework.
future. Finally, the framework presented in this paper considered only J. Environ. Manage. 249, 109364.
bioretention cells as a GI example, and may not be universally applicable Bernaola-Galván, P., Ivanov, P.C., Amaral, L.A.N., Stanley, H.E., 2001. Scale invariance
for GI allocation. Future applications could include more GI types to in the nonstationarity of human heart rate. Phys. Rev. Lett. 87 (16), 168105.
Bilotta, G.S., Brazier, R.E., 2008. Understanding the influence of suspended solids on
support comprehensive and universally applicable decision-making
water quality and aquatic biota. Water Res 42 (12), 2849–2861.
considerations. Browder, G., Ozment, S., Rehberger Bescos, I., Gartner, T., Lange, G.M., 2019.
Integrating Green and Gray: Creating Next Generation Infrastructure. World Bank
4. Conclusions and World Resources Institute.
Chapman, T.G., 1997. Stochastic models for daily rainfall in the Western Pacific. Math.
Comput. Simulat. 43 (3–6), 351–358.
A novel assessment framework for an urban Grey-Green infrastruc­ Chen, P., Tan, G.M., Deng, J.Y., Xu, Q.X., Tang, R.X., 2019. Mutation analysis of annual
ture system is described in response to climate change. The current sediment discharge at Wu Long station in Wu Jiang River Basin from 1960 to 2016.
PLoS ONE 14 (12), e0225935.
framework leverages the synergistic effect of GR and GI within a coupled Chui, T.F.M., Liu, X., Zhan, W.T., 2016. Assessing cost-effectiveness of specific LID
drainage system to produce a superior hydrological performance at the practice designs in response to large storm events. J. Hydrol. 533, 353–364.
least life cycle cost. It highlights the impacts of non-stationary time se­ Da Silva, C.V.F., Schardong, A., Garcia, J.I.B., Oliveira, C.D.M, 2018. Climate change
impacts and flood control measures for highly developed urban watersheds. Water-
ries rather than stationary or event-based climate scenarios emphasised Sui 10 (7).
in previous studies on GR-GI optimisation. Applying the framework to Dickey, D.A., Fuller, W.A., 1979. Distribution of the estimators for autoregressive time
optimize GR-GI in Guangzhou, China, with respect to over 30 years of series with a unit root. J. Am. Stat. Assoc. 74 (366a), 427–431.
Dong, Y., 2018. Performance assessment and design of ultra-high performance concrete
climate change shows that: First, compared to the conventional grey (UHPC) structures incorporating life-cycle cost and environmental impacts. Constr.
infrastructure drainage solution, the decentralised and centralised GR- Build. Mater. 167, 414–425.
GI are economically competitive with a saving of 11% and 18% of the Dos Santos, M.F.N., Barbassa, P., Vasconcelos, A.F, 2021. Low impact development
strategies for a low-income settlement: balancing flood protection and life cycle costs
whole life cycle cost compared with that of conventional GR, respec­
in Brazil. Sustain. Cities. Soc. 65.
tively; Second, the degree of decentralisation of GR-GI has higher im­ Eckart, K., McPhee, Z., Bolisetti, T., 2017. Performance and implementation of low
pacts than the areal proportion of GI in GR-GI on hydrological and impact development - a review. Sci. Total Environ. 607, 413–432.
economic performance. The less extent of GI in a decentralised GR-GI El-Housni, H., Duchesne, S., Mailhot, A., 2019. Predicting individual hydraulic
performance of sewer pipes in context of climate change. J. Water Res. Plan. Man.
depicted high performance in removing TSS in long-term time series 145 (11), 04019051.
simulations; Finally, the decentralised GR-GI shows superior perfor­ Ene, S., Küçükoğlu, İ., Aksoy, A., Öztürk, N., 2016. A genetic algorithm for minimizing
mance in scenarios with extremely long antecedent dry days. Therefore, energy consumption in warehouses. Energy 114, 973–980.
Fletcher, T.D., Shuster, W., Hunt, W.F., Ashley, R., Butler, D., Arthur, S., Trowsdale, S.,
in addition to the factors of precipitation and rainfall duration, ante­ Barraud, S., Semadeni-Davies, A., Bertrand-Krajewski, J.L., Mikkelsen, P.S.,
cedent dry days should be taken into consideration to evaluate the im­ Rivard, G., Uhl, M., Dagenais, D., Viklander, M., 2015. SUDS, LID, BMPs, WSUD and
pacts of climate change on optimised GR-GI strategies in a more more - The evolution and application of terminology surrounding urban drainage.
Urban Water J 12 (7), 525–542.
comprehensive way. The assessment methods developed in this study Gao, X.J., Shi, Y., Han, Z.Y., Wang, M.L., Wu, J., Zhang, D.F., Xu, Y., Giorgi, F., 2017.
have the potential to provide a reference for the planning and decision- Performance of RegCM4 over major river basins in China. Adv. Atmos. Sci. 34 (4),
making of GR-GI in high-density urban areas. The proposed framework 441–455.
Giorgi, F., Mearns, L., 1999. Introduction to special section: regional climate modeling
can not only assess and optimize GR-GI under non-stationarity time revisited. J. Geophys. Res. 104, 6335–6352.
series but also be a powerful support for stormwater management in Gong, Y., Yin, D., Li, J., Zhang, X., Wang, W., Fang, X., Shi, H., Wang, Q., 2019.
response to climate change. Performance assessment of extensive green roof runoff flow and quality control
capacity based on pilot experiments. Sci. Total Environ. 687, 505–515.
Gong, Y.W., Liang, X.Y., Li, X.N., Li, J.Q., Fang, X., Song, R.N., 2016. Influence of rainfall
Declaration of Competing Interest characteristics on total suspended solids in urban runoff: a case study in Beijing.
China. Water-Sui 8 (7).
Haghighi, A., Bakhshipour, A.E., 2012. Optimization of sewer networks using an
None.
adaptive genetic algorithm. Water Resour. Manage. 26 (12), 3441–3456.
Hallegatte, S., Green, C., Nicholls, R.J., Corfee-Morlot, J., 2013. Future flood losses in
Data availability major coastal cities. Nat. Clim. Change. 3 (9), 802–806.
Houle, J.J., Roseen, R.M., Ballestero, T.P., Puls, T.A., Sherrard, J., 2013. Comparison of
maintenance cost, labor demands, and system performance for LID and conventional
Data will be made available on request. stormwater management. J. Environ. Eng. 139 (7), 932–938.
Hsu, C.-.M., Chen, K.-.Y., Chen, M.-.C., 2005. Batching orders in warehouses by
minimizing travel distance with genetic algorithms. Comput. Ind. 56 (2), 169–178.
Hu, M.C., Zhang, X.Q., Li, Y., Yang, H., Tanaka, K., 2019. Flood mitigation performance
Acknowledgement of low impact development technologies under different storms for retrofitting an
urbanized area. J. Clean. Prod. 222, 373–380.
This work was supported by the National Natural Science Foundation IPCC, 2014. Climate Change 2013 – The Physical Science Basis: Working Group I
Contribution to the Fifth Assessment Report of the Intergovernmental Panel on
of China [grant number 51808137], Natural Science Foundation of Climate Change. Intergovernmental Panel On Climate, C. (ed). Cambridge University
Guangdong Province [grant number 2019A1515010873], and Science Press, Cambridge, pp. 741–866.
and Technology Program of Guangzhou, China [grant number Islam, A., Hassini, S., El-Dakhakhni, W., 2021. A systematic bibliometric review of
optimization and resilience within low impact development stormwater
202201010431]. management practices. J. Hydrol. 599.
Jehanzaib, M., Shah, S.A., Yoo, J., Kim, T.W., 2020. Investigating the impacts of climate
Supplementary materials change and human activities on hydrological drought using non-stationary
approaches. J. Hydrol. 588, 125052.
Jiang, C., Li, J., Li, H., Li, Y., Chen, L., 2017. Field performance of bioretention systems
Supplementary material associated with this article can be found, in for runoff quantity regulation and pollutant removal. Water Air and Soil Pollut. 228
the online version, at doi:10.1016/j.watres.2023.119720. (12).

11
M. Wang et al. Water Research 232 (2023) 119720

Joo, J., Lee, J., Kim, J.H., Jun, H., Jo, D., 2014. Inter-event time definition setting stormwater runoff quality from urban and agricultural catchment. Membr. Water
procedure for urban drainage systems. Water-Sui 6 (1), 45–58. Treat. 10 (1), 45–55.
Kim, J., Lee, J., Song, Y., Han, H., Joo, J., 2018. Modeling the runoff reduction effect of Shi, C., Jiang, Z.H., Zhu, L.H., Zhang, X.B., Yao, Y.Y., Li, L., 2020. Risks of temperature
low impact development installations in an industrial area, South Korea. Water-Sui extremes over China under 1.5◦ C and 2◦ C global warming. Adv. Clim. Chang. Res. 11
10 (8). (3), 172–184.
Larsen, T.A., Hoffmann, S., Luthi, C., Truffer, B., Maurer, M., 2016. Emerging solutions to Sohn, W., Kim, J.H., Li, M.H., Brown, R., 2019. The influence of climate on the
the water challenges of an urbanizing world. Science 352 (6288), 928–933. effectiveness of low impact development: a systematic review. J. Environ. Manage.
Lee, J.G., Selvakumar, A., Alvi, K., Riverson, J., Zhen, J.X., Shoemaker, L., Lai, F.-h., 236, 365–379.
2012. A watershed-scale design optimization model for stormwater best Versini, P.A., Kotelnikova, N., Poulhes, A., Tchiguirinskaia, I., Schertzer, D., Leurent, F.,
management practices. Environ. Modell. Softw. 37, 6–18. 2018. A distributed modelling approach to assess the use of Blue and Green
Lim, H.S., Lu, X.X., 2016. Sustainable urban stormwater management in the tropics: an Infrastructures to fulfil stormwater management requirements. Landscape Urban
evaluation of Singapore’s ABC Waters Program. J. Hydrol. 538, 842–862. Plann. 173, 60–63.
Liu, Y.Z., Ahiablame, L.M., Bralts, V.F., Engel, B.A., 2015. Enhancing a rainfall-runoff Wang, J., Cheng, Q.Y., Xue, S.G., Rajendran, M., Wu, C., Liao, J.X., 2018. Pollution
model to assess the impacts of BMPs and LID practices on storm runoff. J. Environ. characteristics of surface runoff under different restoration types in manganese
Manage. 147, 12–23. tailing wasteland. Environ. Sci. Pollut. Res. 25 (10), 9998–10005.
Madarang, K., Kang, J.H., 2013. Accuracy of a stormwater monitoring program for urban Wang, J., Liu, G.H., Wang, J.Y., Xu, X.L., Shao, Y.T., Zhang, Q., Liu, Y.C., Qi, L., Wang, H.
landuses. Water Environ. Res 85 (9), 815–822. C., 2021a. Current status, existent problems, and coping strategy of urban drainage
McFarland, A.R., Larsen, L., Yeshitela, K., Engida, A.N., Love, N.G., 2019. Guide for using pipeline network in China. Environ. Sci. Pollut. Res. 28 (32), 43035–43049.
green infrastructure in urban environments for stormwater management. Environ. Wang, M., Zhang, Y., Bakhshipour, A.E., Liu, M., Rao, Q., Lu, Z., 2022. Designing coupled
Sci-Wat. Res. 5 (4), 643–659. LID–GREI urban drainage systems: Resilience assessment and decision-making
Mei, C., Liu, J.H., Wang, H., Yang, Z.Y., Ding, X.Y., Shao, W.W., 2018. Integrated framework. Science of The Total Environment 834, 155267.
assessments of green infrastructure for flood mitigation to support robust decision- Wang, M., Zhang, D.Q., Su, J., Trzcinski, A.P., Dong, J.W., Tan, S.K., 2017a. Future
making for sponge city construction in an urbanized watershed. Sci. Total Environ. scenarios modeling of urban stormwater management response to impacts of climate
639, 1394–1407. change and urbanization. Clean-Soil Air Water 45 (10).
Meinshausen, M., Smith, S.J., Calvin, K., Daniel, J.S., Kainuma, M.L.T., Lamarque, J.F., Wang, M., Zhang, D.Q., Wang, Z.L., Zhou, S.Q., Tan, S.K., 2021b. Long-term performance
Matsumoto, K., Montzka, S.A., Raper, S.C.B., Riahi, K., Thomson, A., Velders, G.J.M., of bioretention systems in storm runoff management under climate change and life-
van Vuuren, D.P.P., 2011. The RCP greenhouse gas concentrations and their cycle condition. Sustain. Cities. Soc. 65, 102598.
extensions from 1765 to 2300. Clim. Change 109 (1–2), 213–241. Wang, M., Zhang, Y., Zhang, D.Q., Zheng, Y.S., Li, S., Tan, S.K., 2021c. Life-cycle cost
Ministry of Housing and Urban-Rural Development (MOHURD). 2016. Code for design of analysis and resilience consideration for coupled grey infrastructure and low-impact
outdoor wastewater engineering. Beijing, People’s Republic of China: MOHURD. development practices. Sustain. Cities. Soc. 75, 103358.
Mohanty, M.P., Vittal, H., Yadav, V., Ghosh, S., Rao, G.S., Karmakar, S., 2020. A new Wang, M., Zhang, Y., Zhang, D.Q., Zheng, Y.S., Zhou, S.Q., Tan, S.K., 2021d. A Bayesian
bivariate risk classifier for flood management considering hazard and socio- decision model for optimum investment and design of low-impact development in
economic dimensions. J. Environ. Manage. 255, 109733. urban stormwater infrastructure and management. Front. Env. Sci-Switz. 9, 713831.
Nazari-Sharabian, M., Taheriyoun, M., Karakouzian, M., 2019. Surface runoff and Wang, M.M., Sweetapple, C., Fu, G.T., Farmani, R., Butler, D., 2017b. A framework to
pollutant load response to urbanization, climate variability, and low impact support decision making in the selection of sustainable drainage system design
developments - a case study. Water Supply 19 (8), 2410–2421. alternatives. J. Environ. Manage. 201, 145–152.
Nickel, D., Schoenfelder, W., Medearis, D., Dolowitz, D.P., Keeley, M., Shuster, W., 2014. Wang, Z.L., Zhou, S.Q., Wang, M., Zhang, D.Q., 2020. Cost-benefit analysis of low-impact
German experience in managing stormwater with green infrastructure. J. Environ. development at hectare scale for urban stormwater source control in response to
Plann. Man. 57 (3), 403–423. anticipated climatic change. J. Environ. Manage. 264, 110483.
Nika, C.E., Gusmaroli, L., Ghafourian, M., Atanasova, N., Buttiglieri, G., Katsou, E., 2020. Willems, P., Arnbjerg-Nielsen, K., Olsson, J., Nguyen, V.T.V., 2012. Climate change
Nature-based solutions as enablers of circularity in water systems: a review on impact assessment on urban rainfall extremes and urban drainage: methods and
assessment methodologies, tools and indicators. Water Res. 183, 115988. shortcomings. Atmos. Res. 103, 106–118.
Noor, M., Ismail, T., Shahid, S., Asaduzzaman, M., Dewan, A., 2022. Projection of rainfall Wu, C.H., Huang, G.R., 2016. Projection of climate extremes in the Zhujiang River basin
intensity-duration-frequency curves at ungauged location under climate change using a regional climate model. Int. J. Climatol. 36 (3), 1184–1196.
scenarios. Sustain. Cities. Soc. 83, 103951. Xing, W., Li, P., Cao, S.-b., Gan, L.-l., Liu, F.-l., Zuo, J.-e., 2016. Layout effects and
O’Neill, B.C., Tebaldi, C., van Vuuren, D.P., Eyring, V., Friedlingstein, P., Hurtt, G., optimization of runoff storage and filtration facilities based on SWMM simulation in
Knutti, R., Kriegler, E., Lamarque, J.F., Lowe, J., Meehl, G.A., Moss, R., Riahi, K., a demonstration area. Water Sci. Eng. 9 (2), 115–124.
Sanderson, B.M., 2016. The scenario model intercomparison project (ScenarioMIP) Xu, C.Q., Hong, J.L., Jia, H.F., Liang, S.D., Xu, T., 2017. Life cycle environmental and
for CMIP6. Geosci. Model. Dev. 9 (9), 3461–3482. economic assessment of a LID-BMP treatment train system: a case study in China.
Pan, X.D., Zhang, L., Huang, C.L., 2020. Future climate projection in Northwest China J. Clean. Prod. 149, 227–237.
with RegCM4.6. Earth. Space Sci. 7 (2). Xu, C.Q., Tang, T., Jia, H.F., Xu, M., Xu, T., Liu, Z.J., Long, Y., Zhang, R.R., 2019. Benefits
Qi, J.-.D., He, B.-.J., Wang, M., Zhu, J., Fu, W.-.C., 2019. Do grey infrastructures always of coupled green and grey infrastructure systems: evidence based on analytic
elevate urban temperature? No, utilizing grey infrastructures to mitigate urban heat hierarchy process and life cycle costing. Resour. Conserv. Recycl. 151, 104478.
island effects. Sustain. Cities Soc. 46, 101392. Yang, W.Y., Zhang, J., 2021. Assessing the performance of gray and green strategies for
Rasheed, A., Egodawatta, P., Goonetilleke, A., McGree, J., 2019. A novel approach for sustainable urban drainage system development: a multi-criteria decision-making
delineation of homogeneous rainfall regions for water sensitive urban design-a case analysis. J. Clean. Prod. 293, 126191.
study in Southeast Queensland. Water-Sui 11 (3). Ying, J., Zhang, X.J., Zhang, Y.Q., Bilan, S., 2021. Green infrastructure: systematic
Requena, A.I., Burn, D.H., Coulibaly, P., 2019. Estimates of gridded relative changes in literature review. Econ. Res-Ekon. Istraz.
24-h extreme rainfall intensities based on pooled frequency analysis. J. Hydrol. 577, Yu, Y., Chen, L., Xiao, Y.C., Chang, C.C., Zhi, X.S., Shen, Z.Y., 2022. New framework for
123940. assessing urban stormwater management measures in the context of climate change.
Rezaei, A.R., Ismail, Z., Niksokhan, M.H., Dayarian, M.A., Ramli, A.H., Yusoff, S., 2021. Sci. Total Environ. 813, 151901.
Optimal implementation of low impact development for urban stormwater quantity Zeng, J.J., Lin, G.S., Huang, G.R., 2021. Evaluation of the cost-effectiveness of Green
and quality control using multi-objective optimization. Environ. Monit. Assess. 193 Infrastructure in climate change scenarios using TOPSIS. Urban For. Urban Green.
(4). 64, 127287.
Rossman, L.A, 2010. Storm Water Management Model User’s Manual, Version 5.0. Zhang, H., Yang, Z.F., Cai, Y.P., Qiu, J., Huang, B.S., 2021. Impacts of climate change on
National Risk Management Research Laboratory, Office of Research and urban drainage systems by future short-duration design rainstorms. Water-Sui 13
Development, US Environmental Protection Agency. (19), 2718.
Rossman, L.A., Huber, W.C., 2016. Storm Water Management Model Reference Manual Zhou, S., Liu, Z., Wang, M., Gan, W., Zhao, Z., Wu, Z., 2022. Impacts of building
Volume I–Hydrology. US Environmental Protection Agency. configurations on urban stormwater management at a block scale using XGBoost.
Salim, I., Paule-Mercado, M.C., Sajjad, R.U., Memon, S.A., Lee, B.Y., Sukhbaatar, C., Sustain. Cities Soc. 87, 104235.
Lee, C.H., 2019. Trend analysis of rainfall characteristics and its impact on

12