Optimal design of low impact development practices in response to climate change

Journal Pre-proofs Research papers Optimal Design of Low Impact Development Practices in Response to Climate Change Seyed Hamed Ghodsi, Zahra Zahmatkesh, Erfan Goharian, Reza Kerachian, Zhenduo Zhu PII: DOI: Reference:

S0022-1694(19)31001-7 https://doi.org/10.1016/j.jhydrol.2019.124266 HYDROL 124266

To appear in:

Journal of Hydrology

Received Date: Revised Date: Accepted Date:

26 February 2019 8 May 2019 21 October 2019

Please cite this article as: Hamed Ghodsi, S., Zahmatkesh, Z., Goharian, E., Kerachian, R., Zhu, Z., Optimal Design of Low Impact Development Practices in Response to Climate Change, Journal of Hydrology (2019), doi: https:// doi.org/10.1016/j.jhydrol.2019.124266

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier B.V.

Optimal Design of Low Impact Development Practices in Response to Climate Change

Seyed Hamed Ghodsi a,*, Zahra Zahmatkesh b, Erfan Goharian c, Reza Kerachian d, Zhenduo Zhu e

a

Ph.D. Student, Department of Civil, Structural and Environmental Engineering, University at Buffalo, NY 14260,

USA. E-mail: [email protected] b

Post-Doctoral Researcher, Faculty of Engineering, Department of Civil Engineering, McMaster University,

Hamilton, ON L8S4L8, Canada. E-mail: [email protected] c

Assistant Professor, Department of Civil and Environmental Engineering, College of Engineering and Computing,

University of South Carolina, SC 29208, USA. E-mail: [email protected] d

Professor and Vice-Dean in Academic Affairs, School of Civil Engineering, University of Tehran, Tehran, Iran. E-

mail: [email protected] e

Assistant Professor, Department of Civil, Structural and Environmental Engineering, University at Buffalo, NY

14260, USA. E-mail: [email protected] *

Corresponding author

ABSTRACT Urban runoff drainage systems are generally designed based on the analysis of historical hydrologic variables, especially rainfall. Hydrologic analysis, however, has been shown to be affected by climate change. In response to the modified pattern of rainfall, stormwater management practices, for instance, Low Impact Developments (LIDs), need to be applied to successfully respond to the future possible climatic conditions, while meeting the performance criteria based on the current state of the watershed. In this study, a simulation-optimization 1

framework is proposed to find the optimal design of LIDs (type, location, and area) under climate change. Storm Water Management Model (SWMM) is utilized to simulate the rainfallrunoff process. This model is linked with a single-objective optimization routine formulated by Genetic Algorithm (GA) that minimizes runoff volume and the costs associated with the LIDs’ design. To investigate how climate change might affect the future watershed runoff, rainfall projections by several General Circulation Models (GCMs) are used. Change factor methodology is applied to downscale GCMs’ outputs. The framework is applied to the Velenjak watershed, located in the northeast part of Tehran, Iran. Results show that the optimization algorithm provides optimal solutions for LIDs’ design under different extreme climate scenarios by allocating an average of 0.23% of the watershed area (20 % of the maximum potential area considered for LIDs’ implementation), which results in a runoff volume reduction of up to 18%. The results also indicate that the optimal design of LIDs based on the current condition satisfies the watershed management goals and expectations in a future subjected to climate change impacts. Moreover, the results signify that the formulated multi-event optimization algorithm can provide a unique solution meeting the objective functions for all considered climate scenarios at once. Keywords: Climate Change, LID, Urban Runoff Management, Optimization, SWMM

1. INTRODUCTION Rapid urbanization, unplanned development of lands, population growth, aged drainage systems, and climate change adversely affect urban stormwater by degrading runoff quality and altering runoff peak and volume (Zahmatkesh et al. 2014b; Zuo et al. 2016). Stormwater drainage systems are generally designed based on historical statistics of hydrologic variables, 2

which are assumed to be stationary. Nevertheless, drivers such as climate change have been shown to alter patterns of hydrologic variables and extreme events, and in result runoff volume and peak (Ferrer et al. 2012; Andersson-Sköld et al. 2015; Yoon et al. 2015; Putro et al. 2016). Effect of climate change on non-stationarity in hydrologic variables/extremes has been investigated in several studies (e.g., Vaze et al. 2010; Mondal and Mujumdar 2015; Ganguli and Coulibaly 2017). In urban areas, there is a concern that stormwater runoff may be affected by climate change impacts on precipitation amounts, intensities, and frequencies (Fowler and Hennessy 1995). Pyke et al. (2011), for instance, discussed that stormwater runoff is sensitive to changes in land impervious cover, followed by changes in precipitation because of climate change impacts. In climate change studies, precipitation is often the main variable affecting runoff and consequently extreme events. Darand and Sohrabi (2018) identified drought- and floodprone areas of Iran by analyzing spatiotemporal changes in precipitation under climate change impacts. Switzman et al. (2017) showed how rainfall intensity-duration-frequency curves could vary under different climate change models and scenarios, with study sites in southern Ontario, Canada. The variations in rainfall could result in negative effects on urban drainage system and lead to more frequent and severer flooding events (Nilsen et al. 2011). During the recent decades, a projection of future changes in the rainfall-runoff process has received considerable attention all over the world in order to cope with the uncertainties from a changing climate. Zahmatkesh et al. (2014b) show that climate change can result in increases in volume and peak discharge of urban stormwater runoff in the Bronx River watershed in New York City, USA. Frequency analysis on the projected runoff also indicated a considerable increase in the frequency of occurrence of extreme storm events. Yoon et al. (2015) proposed a methodology for the evaluation of the overflow probability of urban streams in the Uicheon Basin of Korea, considering regional climate models’ data. They showed that future 100-year 3

peak flow could be increased by 58.1% compared with the historical conditions. Putro et al. (2016) investigated the impacts of climate and land use change on water quantity and quality for four catchments in England. Their results indicated no trend in annual rainfall but an upward trend in the runoff in the two urban catchments. Zuo et al. (2016) investigated the impacts of land use and climate changes on water yields in the Huangfuchuan River basin of the Loess Plateau in China, by detecting temporal trends and abrupt changes in runoff. Their results indicated decreasing trends in annual runoff and precipitation. The changes in urban stormwater runoff resulted from the climate change impacts may have considerable implications for designing stormwater and flood control measures for watershed management (Zahmatkesh et al. 2014a). Once surface runoff exceeds the drainage system capacity, stormwater management practices such as low impact development and green infrastructure are required to be employed. Implementing runoff management practices in response to climate change can reduce climate-related risks. Adaptation measures are aimed as an integrated approach to address different aspects of watershed management (Andersson-Sköld et al. 2015). Low impact development (LID) techniques are adapted as climate mitigation measures, which also decrease the adverse impacts of urban growth on stormwater to meet drainage performance standards. Permeable pavement, green roofs, bio-retention cells, vegetated swales, and infiltration trench are among the widely used LID measures. These techniques can be economically and environmentally efficient for runoff management in urban areas. Understanding of LIDs in the context of climate change adaptation can be helpful in informing stormwater management decisions to reduce the risk of harmful future impacts (Pyke et al. 2011). Gill et al. (2007), Pyke et al. (2011), Newcomer et al. (2014) and Zahmatkesh et al. (2014a) are among those to investigate how LIDs’ implementation is effective in moderating potential 4

impacts of climate change on surface runoff in urban areas. Although LIDs have been mentioned and used as promising practices to control stormwater runoff volume and pollution in the urban areas, experimental and modeling efforts are required to test LID characteristics and propose an adequate guideline for optimizing LID management (Baek et al. 2015). In optimizing LIDs’ application, objectives such as improving runoff quality as well as decreasing runoff volume and costs of the implementation of LIDs can be incorporated in the modeling effort. To find the most desired LID management options for urban runoff management, single and multi-objective optimization tools have been widely used during the recent years. Lee et al. (2015) used System for Urban Stormwater Treatment and Analysis Integration (SUSTAIN) for evaluating the optimal location, type, and cost of stormwater management practices to meet water quality and quantity goals. Baek et al. (2015) proposed a methodology to optimize the sizes of different types of LIDs by conducting intensive stormwater monitoring and numerical modeling in a commercial site in Korea. Ghodsi et al. (2016-b) used Non-dominated Sorting Genetic Algorithm II (NSGA-II) to determine the LIDs’ type, area, and location for runoff quality and quantity control in Tehran urban watershed, Iran. Macro et al. (2018) also developed a new open-source multi-objective SWMM optimization tool, OSTRICH-SWMM, to illuminate trade-offs between the cost of rain barrel placement and the reduction of combined sewer overflows (CSOs) in Buffalo, New York. Only a few studies investigated how optimal design of climate adaptation measures based on historical rainfall would perform in future based on rainfall modified by climate change impacts (e.g., Zhou et al. 2018 and Loiola et al. 2018). This study aims to present an integrated framework for the optimal design of LID practices under the potential impacts of climate change. An urban watershed (Velenjak) in Tehran, Iran, modeled with the Storm Water Management Model (SWMM) for rainfall-runoff simulation, is selected as a real-world case study to implement the proposed methodology. Moreover, future 5

hydrologic conditions under climate change impacts are projected using the General Circulation Models (GCMs), downscaled by a change factor methodology. LID controls are utilized to decrease urban runoff volume (as a stormwater management method) while minimizing the costs associated with the LIDs’ implementation. Optimization algorithms are formulated for optimal design of the LIDs under different rainfall scenarios. The scenarios are defined to be the representatives of minimum, maximum, and average rainfall possible conditions for the watershed. The focus of this research is on the changes to single storm events (here 2- and 20year return period) intensities, rather than climate change impacts on storms’ frequency and/or duration. The objectives of the study are to answer two main questions: I) Does the optimum solution for the historical rainfall scenario (defined based on single storm events with return periods of 2- and 20-year) successfully work for the possible future rainfall scenarios under climate change impacts? and II) Compared to one optimum solution corresponding to each rainfall scenario, how efficient would it be to develop a multi-event optimization approach, which considers all rainfall scenarios at once? The second question has been raised because, given the uncertainty in climate change impacts, it is assumed to be more practical to develop one unique solution (LIDs’ type, area, and location), which can meet the runoff management goals for different future rainfall scenarios. The solutions for these two questions would finally be compared to find the most desirable solution for urban runoff management considering climate change impact.

2. STUDY AREA To implement the framework for urban runoff quantity management, Velenjak watershed, in the northeast part of Tehran, Iran, is selected. Tehran, the capital city of Iran, is located at 35°41’21’’ N and 51°23’20’’ E and has an area of 730 km2 (Fig. 1). The population of Tehran in 6

the urban area is 8.8 million. Tehran has a semi-arid climate with mountains to its north and central desert to the south. It is mild in spring and autumn, and cold in winter. Summers in Tehran are long, hot and dry with very light rain (in average 10 mm/month), and low relative humidity. More rainfall occur from early-winter (~30 mm in December) to mid-spring (~50 mm in March). http://www.maphill.com

6

Iran

2 1 4 3 5

8

7 9

11 13

10 12 14

15

16 19

18 17 21

Chitgar lake

Velenjak watershed

20

24

Tehran

23

22

25 Rainfall Station Runoff collection system 0 1 28.4214.03 3 5

potential sub-catchments for LID implementation

10 km 28.06

0 100 300 500

26

27

1000 m

Fig. 1. Study area (Velenjak watershed) in the northeast part of Tehran, Iran.

Because of the population growth, urban expansion and increase in the impervious areas, the drainage system in Tehran is not able to capture the total runoff volume produced from heavy rainfall events. Moreover, groundwater level as one of the main sources of supplying water in Tehran is dropping down. One reason is the increase in impervious areas, which prevents stormwater runoff to infiltrate and recharge the groundwater.

7

The region of focus in this study has 27 sub-catchments with the total area of 19.77 (Fig. 1). Sub-catchments’ areas vary between 0.162 km2 and 1.745 km2. Lower and upper values of impervious coefficients and slopes are 0 and 82%; and 2% and 80%, respectively. Land use in this region is classified into three categories: undeveloped, residential with the low-density population and residential with the high-density population. Topographic data along with length and roughness of channels are obtained from Soltani (2009). In addition, rainfall time series (every 6 hours) from 1998 to 2015 are collected by the Shomal Tehran synoptic station, located at 35°480´N and 51°290´E and elevation of 1549.1 m (red circle in Fig. 1). The duration of the design rainfall is 24-hours based on the technical reports on the study area discussing the design standards of the local drainage system (Mahab Ghods Consulting Engineering Company, 2012). The rainfall values and distribution are provided later in the result section.

3. METHODOLOGY In this research, an integrated framework is proposed for the simulation and management of urban runoff considering climate change impacts. SWMM is used for rainfall-runoff modeling. LID techniques are suggested to control runoff quantity. An optimization routine is formulated for runoff management based on a defined objective function associated with the runoff quantity and implementation cost of LIDs. This is accomplished through developing a model using the Genetic Algorithm (GA) linked with the SWMM. Type, location and area of LIDs are considered as the decision variables in the optimization model. To investigate climate change impacts on the watershed runoff, various GCMs are selected. A change factor methodology is used to downscale raw GCM projections to a finer-grid resolution, which will be useful for the future assessment of impacts of extreme events on urban runoff. Design for LIDs to mitigate adverse 8

impacts of climate change on runoff quantity is then investigated. The most appropriate management solution for future and historical conditions of the watershed are compared. Moreover, an optimization model is formulated for optimal design of a unique solution for LIDs considering multiple rainfall scenarios. Flowchart of the proposed methodology is shown in Fig. 2. In the following sections, details of the different steps of the methodology along with the applied tools and models are presented. Analyzing climate change impacts on watershed runoff management 1- Data collection and processing 1-1- Drainage system characteristics

1-2- Historical rainfall

1-3- Determination of design rainfall

1-4- Historical runoff

1-5- Information of LIDs

2- Rainfall-runoff model SWMM for runoff quantity simulation

3- Urban runoff management Selecting appropriate LID controls

4- Climate change impact modeling 4-1- GCM outputs for rainfall

4-2- Rainfall downscaling

4-3- Identifying extreme climate scenarios

5- Optimization model 5-1- Objective function, constraints and decision variables

5-2- Genetic Algorithm (GA) optimization model

Fig. 2. Flowchart of the proposed methodology for urban runoff management under climate change impacts

9

3.1. Data collection To build the methodology and apply it to a real-world case study, several data sets and different information are collected and used. The data include observed historical rainfall and runoff, topographic and hydrologic characteristics, and LID controls’ initial information for the study watershed. To incorporate the climate change assessment, GCMs products for all the provided models and scenarios are obtained from http://www.ipcc-data.org. The raw and large-scale products are downloaded for the historical simulation for a grid which contains the rain gauge station. In this study, the data of 14 GCMs under the climate change scenarios A1, A1B, and B2 are used. The list

of

models

and

scenarios

can

be

found

in

(http://www.ipcc-data.org/cgi-

bin/ddc_nav/dataset=ar4_gcm). Also, future projected information of these models and scenarios are downloaded for the period of 2046-2065 (considering data availability) for the same grid as data of the historical simulations. 3.2. Rainfall-runoff simulation model The SWMM hydrologic model, developed by the United States Environmental Protection Agency (USEPA) (Rossman 2015), is employed for the simulation of the rainfall-runoff process. SWMM has been widely used for urban hydrologic modeling all around the world. Examples are Kovács and Clement (2009), Karamouz and Nazif (2013), Zahmatkesh et al. (2014a, b), Ghodsi et al. (2016- a,b) and Yang and Chui (2018). It should be noted that the model for the study watershed is previously calibrated by Soltani (2009) based on the sub-catchments’ characteristics including total area, impervious area, width and curve numbers, channels length and roughness as well as the observed historical rainfall and runoff.

10

The developed SWMM is used for dynamic single-event simulation of surface hydrology. The one-day design rainfall based on the historical time series of data is used as the model input. Within the model, the hydrologic component receives rainfall and generates runoff. The routing time steps is 30 seconds. The hydraulic module then transports the generated runoff through channels to the watershed outfall. The quantity of flow in the watershed outlet is tracked and recorded for further analysis.

3.3. Urban runoff management LID controls are designed based on the characteristics of the sub-catchments, namely land use, area, slope, and impervious area. Using the SWMM, combinations of LIDs can be included while hydrologic modeling is performed for the watershed. To investigate LIDs’ effect on runoff quantity, several types of LIDs including permeable pavement (PrPv), bio-retention cell (BRc), vegetative swale (VeS) and infiltration trench (InTr) are considered. PrPvs are hard layers filled with gravel and paved with porous concrete. They allow rainfall to penetrate in the underneath base. They have benefits for reducing surface streamflow quantity and pollutant loading (Kuo et al. 2007; Jia et al. 2015). BRcs are depressed layers of vegetation, soil mixture with a gravel bed. They are able to provide storage, infiltration, and evaporation of rainfall and surface runoff. BRcs are designed to remove dissolved pollutants and particulates from the surface stormwater. Depending on the infiltration rate, BRcs can capture approximately 92% of stormwater (Lucas 2008). VeSs are networks of canals with sloping sides enclosed by vegetation. They slow down the movement of runoff and allow it to percolate the soil beneath the ground. InTrs are ditches filled with gravel. They capture streamflow and provide capacity and time for it to infiltrate the soil. The characteristics’ of LIDs’ have been taken from SWMM user manual (Rossman 2015)

11

and Mahab Ghods Consulting Engineering Company (2012), which are presented in Table A1 (See supplement section). To effectively integrate the abovementioned LIDs in the watershed runoff management, the sub-catchments with high runoff volume and impervious area are identified as the potential locations for LIDs’ implementation. For this purpose, an index is calculated for each subcatchment by multiplying the total runoff volume to the corresponding impervious coefficient. 3.4. Climate change impact modeling Understanding of climate impacts on stormwater runoff can be derived from monitoring the changes in rainfall projection of GCMs. These models have around 100–250km horizontal resolution. Several approaches have been demonstrated to scale projected values for higher spatial resolution and reduce the biases between GCM and local projected time series. Here, the change factor methodology (CFM) is used due to its previous successful applications, simplicity, and its computationally economical value in comparison to other downscaling methods. This method has been widely used by researchers and applied in different climate change impact assessment studies on water supply and availability (Conway and Hulme 1996; Karamouz et al. 2014; Goharian et al. 2015), ecology (Buckley et al. 2015), climate variability and hydropower generation (Mehta et al. 2011), hydrological applications (Arnell and Reynard 1996; Seidel et al. 1998), and flood analysis (Panagoulia and Dimou, 1997; Reynard et al. 2001; Prudhomme et al. 2002; Zahmatkesh et al. 2014a,b). Here, after downscaling the raw GCMs’ precipitation products, the scenarios which represent the extreme cases (i.e., wet and dry), as well as normal conditions, are identified. Then, these scenarios are used to obtain design rainfall for the future conditions of the watershed under climate change impacts. List of the climate models and scenarios considered to identify extreme cases is provided in Table 1. 12

Table 1. List of the climate models used to define the rainfall scenarios Original resolution Model number

Model name

Scenarios

Country Long × Lat levels (L)

1

CSIRO Mk3.0

A2, A1B, B1

Australia

192×96 L18

2

ECHOG-G

A2, A1B

Germany

96×48 L19

3

GFDL CM2.0

A2, A1B, B1

United States

144×90 L24

4

GFDL CM2.1

A2, A1B, B1

United States

144×90 L24

5

GISS-ER

A2, B1

United States

72×46 L20

6

HadGEM1

A2, A1B

United Kingdom

192×145 L38

7

INM-CM3.0

A2, A1B, B1

Russia

72×45 L21

8

IPSL-CM3

A2, A1B, B1

France

96×72 L19

9

MIROC3.2

A2, A1B, B1

Japan

128×64 L20

10

ECHAM5

A2, A1B, B1

Germany

192×96 L31

11

MRI-CGCM2.3.2

A2, A1B, B1

Japan

128×64 L30

12

CCSM3

A2, A1B, B1

United States

156×128 L26

13

PCM

A2, A1B

United States

128×64 L26

14

HadCM3

A2, A1B, B1

United Kingdom

96×73 L19

3.4.1. Rainfall downscaling CFM incorporates information of local observed rainfall to provide direct scaling of the GCMs. Change factors (or scaling values) should be first calculated based on the differences or ratios of the historical and future projections of GCMs. Then, CFM develops additive or multiplicative relationships to scale the historical local observations to the future rainfall projections. The historical local observations and GCM projections for the baseline period should be equal in length to the corresponding future time period. Change factors are different in terms of their temporal resolution, domain, and scales, and also the mathematical formations. 13

How to better select an appropriate time period, temporal domain, scale, number of change factors, and type of change factors may vary depending on the meteorological characteristics and purpose of the study. More information in this regard can be found in Anandhi et al. (2011) and Hansen et al. (2017). In this study, 12 monthly multiplicative change factors are used to scale the historically observed rainfall time series to the future possible rainfall projections. These change factors are calculated as follows: CFi,m = Pi,m PHis i,m

Fut

(1)

Obs LPFut i,t = CFi,m × LPi,t

(2)

where CFi,m is the monthly change factor for each GCM model and specific scenario (i) and month (m). Pi,mFut and Pi,mHis are the future and baseline values of precipitation for each time increment (m) and model (i). LPi,mFut and LPi,mObs are local-scaled future value and the corresponding local historical observed value of rainfall for the tth time-step. 3.4.2. Rainfall scenarios based on climate change projections In order to find the extreme rainfall conditions from the future projections of GCMs, a method developed by Karamouz et al. (2014) is used. Based on this method, the projections of different GCMs and their respective change factors are calculated for each model under different climate change scenarios. In order to select the minimum, maximum, and average models (Min P, Max P, and Mean P), the following equation is used:

{

MDCFJsel = Min For j = 1,2, …,n:

[∑12i= 1|CFi,m ― f(CFi)|]}

(3)

where Jsel stands for the model projection in which its change factors have the minimum difference with the selected function (MDCF). f is the function that considers the selection of desired scenarios, which is minimum, maximum, or average for Min P, Max P, or Mean P, respectively (the corresponding rainfall scenarios hereafter will be referred as Min P, Max P, and 14

Mean P). This metric determines the total difference between the CFs of the selected GCM with the extreme CFs, based on the desired function (f). In addition, three other scenarios are defined which represent the mean, max, and min precipitation scenarios using absolute mean, max, and min CFi,m. These scenarios are synthetic scenarios and do not represent any GCM projection. They are generated in this study to represent extreme cases. The corresponding rainfall scenarios hereafter will be referred as A-Min P, A-Max P, and A-Mean P. Identifying extreme scenarios using the change factors, results in the representation of the possible range of future runoff variation under climate change impacts. 3.5. Optimizing LIDs’ design Determining the most suitable type, area and location of LIDs is the main concern related to using LIDs for urban runoff management. In response to this concern, in this study, a singleobjective optimization model (Optimization algorithm I) is formulated using GA to find the optimal possible solution for the decision variables according to the type of constraints and the objective function. Total runoff volume in the outfall is preferred to be minimized. This leads to mitigate the dimensions of channels and construction costs. In addition, the implementation cost of LIDs is favored to be minimized. These two aims are combined to build the single objective function (𝐹 (𝑥)): Optimization algorithm I 𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒

𝐹(𝑥) = [𝑤1 × 𝑂𝑏𝑗 1 + 𝑤2 × 𝑂𝑏𝑗 2]

𝑉(𝑥) ― 𝑉𝑚𝑖𝑛

𝐶(𝑥) ― 𝐶𝑚𝑖𝑛

𝑂𝑏𝑗 1 = 𝑉𝑚𝑎𝑥 ― 𝑉𝑚𝑖𝑛 , 𝑂𝑏𝑗 2 = 𝐶𝑚𝑎𝑥 ― 𝐶𝑚𝑖𝑛 Subject to: lower bound 𝑖 ≤ xi ≤ upper bound 𝑖 & 𝑤1 + 𝑤2 = 1

15

(4)

where 𝐶(𝑥) is the cost for the implementation of LIDs, 𝑉(𝑥) is the runoff volume. 𝑉𝑚𝑎𝑥 , 𝑉𝑚𝑖𝑛, 𝐶𝑚𝑎𝑥, and 𝐶𝑚𝑖𝑛 are the maximum and minimum values of runoff volume and cost, respectively. 𝑂𝑏𝑗 1 and 𝑂𝑏𝑗 2 are functions defined to standardize minimizing the runoff volume and cost of LID’s implementation. The value of the objective function (𝐹(𝑥)), as well as 𝑂𝑏𝑗 1 and 𝑂𝑏𝑗 2 is bounded between 0 and 1. “x” represents the areas and location of LIDs as the decision variables. In addition, 𝑤1 and 𝑤2 in Equation 4 are the coefficients representing the relative importance of the objective function components, namely runoff quantity and cost, respectively (𝑤1 + 𝑤2 = 1). Constraints of the optimization model are the minimum and maximum values of the LIDs’ areas in each sub-catchment, which are defined, based on the subcatchments characteristics. “i” in the constraint is a counter to represent the decision variables. Finally, the developed optimization model is driven by defining the population size, total generation number, crossover and mutation parameters, as well as the constraints. Furthermore, a second optimization model (referred to as Optimization algorithm II) is formulated for optimal design of LIDs considering multiple possible rainfall events all at once: Optimization algorithm II 𝑠 = 14

𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 ∑𝑠 = 1 𝐺(𝑥) = [𝑤1 × 𝑂𝑏𝑗1𝑠 + 𝑤2 × 𝑂𝑏𝑗2𝑠 ]

(5)

where “s” represents the rainfall scenario and 𝐺(𝑥) is the objective function. 𝑂𝑏𝑗1𝑠 and 𝑂𝑏𝑗2𝑠 are the values of 𝑂𝑏𝑗 1 and 𝑂𝑏𝑗 2 (as defined in Equation 4) corresponding to scenario s. The algorithm will minimize the summation of the 𝐺(𝑥) for all defined rainfall scenarios and result in the optimal solution for LIDs’ areas and locations.

16

4. RESULTS 4.1. Design rainfall In order to estimate the design storms, frequency analysis is performed on the historical annual maximum series of rainfall. 24-hour rainfalls with return periods of 2-year and 20-year (cumulative values of 36.6 mm, and 54 mm, respectively) are consequently selected for the analysis. The duration of the design storm as well as the 2- and 20-year return periods are adopted from the technical reports on the study area discussing the design standards of the local drainage system (Mahab Ghods Consulting Engineering Company, 2012). For defining the rainfall distribution, the rainfall hyetograph of the study area is used (Ghodsi et al. 2016-a). The distribution of the design rainfalls with the two return periods (2-year and 20-year) is shown in Fig. 3. 12 20-year

Rainfall (mm/h)

10

2-year

8 6 4 2 0 0

2

4

6

8

10

12

14

16

18

20

22

24

Time (h) Fig. 3. Design rainfalls hyetographs with return period of 2-year and 20-year (Ghodsi et al. 2016-a)

Hyetographs of the corresponding design storms are used as the SWMM input representing the baseline scenario (without considering climate change impacts).

17

4.2. Potential sub-catchments for LIDs’ implementation To identify potential sub-catchments for LIDs’ implementation, an index (columns 4 and 8 in Table 2) is calculated by multiplying the total runoff volume in each sub-catchment to the corresponding impervious coefficient. To obtain the total runoff in each sub-catchment, the simulation model is driven by the design rainfall with return period of 2-year (refer to Fig. 3.) without the use of LIDs. The produced runoff in the 27 sub-catchments of the study area is shown in Table 2. Land use and impervious area are the main factors contributing to the runoff for the sub-catchments (for more details see Ghodsi et al. 2016-a). Table 2. Total runoff, impervious area, and the defined index for the sub-catchments in the initial condition, regarding finding the potential sub-catchments to implement LIDs Runoff Sub*

Runoff % Imperv**

Index***

Sub

(mm)

% Imperv

Index

(mm)

1

0

0

0

15

19.08

62

1183

2

0

0

0

16

11.39

37

421

3

0

0

0

17

22.02

72

1585

4

0

0

0

18

20.56

67

1378

5

0

0

0

19

11.38

37

421

6

0

0

0

20

25.14

82

2061

7

0

0

0

21

2.47

32

79

8

0

0

0

22

25

82

2050

9

0

0

0

23

1.98

32

63

10

2.17

7

15

24

1.97

32

63

11

0

0

0

25

21.84

72

1572

12

22.17

72

1596

26

11.43

37

423

13

20.59

67

1380

27

20.5

67

1374

14

23.66

77

1822

* Sub: Sub-catchment

18

** % Imperv: Percent of the land area (not including any LIDs) which is impervious. ***Index = Runoff × %Imperv

The first 11 sub-catchments, located in undeveloped areas, are completely pervious and do not contribute to runoff generation. Therefore, the value of the index for these sub-catchments is negligible. Based on the results, the value of the index for sub-catchments 12, 13, 14, 15, 17, 18, 20, 22, 25 and 27 (Bolded in Table 2) is significantly larger than the others. Thus, they are selected as the potential sub-catchments for the implementation of LIDs (green shaded areas in Fig. 1.) 4.3. Climate change projection Frequency analysis is performed on the downscaled rainfall and consequently, updated design rainfalls under climate change impacts are obtained. Table 3 presents the value of the design rainfalls based on historical and future data. Based on the monthly change factors calculated for the GCMs, the absolute (as explained in the methodology section) minimum, average, and maximum precipitation scenarios (S2-S4 for 2-year design rainfall and S9-S11 for 20-year design rainfall) are formed. This table shows that MRCGCM (B1), HADCM3 (A1B), and GFCM21 (B1) climate models project the minimum, average, and maximum possible rainfall in the future (scenarios S5-S7 and S12-S14). The baseline design rainfall scenarios (S1 and S8) are also shown in Table 3, which correspond to the BAU (business as usual) state when climate change impacts are not incorporated in the analysis. The GFCM21 (B1) projects about 35%

increase in 2-year design rainfall in the future compared with the historical rainfall (comparing S7 with S1), while under the same model and scenario, the 20-year design rainfall can be increased by almost twice (68%) (comparing S14 with S8). It should be noted the absolute scenarios are not the real representations of GCM projections. They are formed based on the 19

combinations of GCMs and project the extreme conditions in each month. The interesting point is that the combination of GCMs, which represents the absolute averaged condition in this study (scenario S3), shows just about 1% changes compared to the historical observed 2-year design rainfall (scenario S1). It shows that although the GCM projections are diverse, the average change of all their projections suggests a really small difference with the historical rainfall. The most moderate model (with projection close to historical condition) is the HADCM3 (A1B). This model projects about 3mm increase (8%) in 2-year design rainfall. In terms of the 20-year design rainfall, the MRCGCM (B1) suggests about 9% decrease in the design rainfall for the future (i.e., 33.4 mm). For the 20-year design rainfall, all future climate scenarios (except A. Min P), projects an increase in the rainfall. In conclusion, the results show that, based on the Mean P model, there is a tendency in the GCM projections for an increase in the 2-year and 20-year design rainfalls for the study area. According to Table 3, an increase of 1% to 68% could be expected for future design rainfall under climate change impacts for this region. Table 3. Rainfall scenarios; cumulative amount of rainfall (mm) for design scenarios based on the current status of the watershed (BAU) and under the impact of climate change Future scenarios

Baseline Return period scenarios

Min P A-Min P

A-Mean P

Mean P

Max P

A-Max P MRCGCM (B1)

HADCM3 (A1B)

GFCM21 (B1)

S1 (BAU)

S2

S3

S4

S5

S6

S7

36.6

24.5

37.1

55.0

33.4

39.6

49.3

S8 (BAU)

S9

S10

S11

S12

S13

S14

54.0

37.6

58.1

90.6

56.5

63.6

90.6

2-year

20-year

20

4.4. Optimizing LIDs’ implementation in the watershed 4.4.1. Optimization model: genetic algorithm Considering the 10 potential sub-catchments (i.e., sub-catchments 12, 13, 14, 15, 17, 18, 20, 22, 25, 27 as explained in section 4.2) and four different types of LIDs (i.e., VeS, BRc, PrPv, and InTr), 40 decision variables are identified for the optimization model, which represent the type, location, and area of LIDs in different sub-catchments. The definition of the decision variables and the constraints for the optimization models are provided in Ghodsi et al. (2016-b). As a brief explanation, decision variables x1, x2, x3 and x4 represents VeS, BRc, PrPv and InTr, respectively, which are related to sub-catchment 12; similarly, x5, x6, x7, and x8 correspond to sub-catchment 13. It goes in a way that all potential sub-catchments (presented in Table 2) would be covered. Therefore, the last four variables (i.e., x37, x38, x39, and x40) are related to subcatchment 27. The lower bound of the decision variables is set to zero. The upper bounds are determined based on the sub-catchments’ land use and area. The unit implementation cost for LIDs are obtained from Fallahi Zarandi (2013) and Ghodsi et al. (2016-a): 5.77, 14.46, 7.65 and 11.07 Dollar/m2, for implementing vegetative swale (VeS), bio-retention cell (BRc), permeable pavement (PrPv) and infiltration trench (InTr), respectively. For calculating the maximum and minimum values of runoff volume and cost (to be used in the objective function), the developed SWMM model is driven for all values of the design rainfall considering two conditions: I) without LIDs’ implementation, II) with the maximum possible area of LIDs in the target sub-catchments. The minimum cost is zero and the maximum cost is obtained $1,796,555. The minimum and maximum values of runoff volume for the rainfall scenarios are presented in Table 4. The minimum volume is expected when the maximum possible LID areas are implemented, while the maximum runoff volume will obtain when no LID is used in the sub-catchments. In addition, equal values of weights to minimize 21

runoff volume and implementation cost of LIDs are considered in this analysis. However, to expand the current study, effect of uncertainty in these weights on the optimization results could be investigated. Table 4. The maximum and minimum values of runoff volume for each rainfall scenario Rainfall

Rainfall

Vmin

Vmax

Rainfall

Rainfall

Vmin

Vmax

scenario

(mm)

(106 L)

(106 L)

scenario

(mm)

(106 L)

(106 L)

S1

36.6

214.1

263.7

S8

54

355.9

413.1

S2

24.5

120.3

160.3

S9

37.6

222

272.3

S3

37.1

218

267.9

S10

58.1

389.2

448

S4

55

363.7

421.6

S11

90.6

662.2

727.7

S5

33.4

189

236.6

S12

56.5

376.5

434.9

S6

39.6

238.6

289.9

S13

63.6

434.6

495.4

S7

49.3

317.4

372.9

S14

90.6

662.2

727.7

4.4.2. LIDs optimal design Following estimating the maximum and minimum runoff volume and cost for LIDs’ implementation, the optimization model is solved for design rainfall scenarios. An interface in MATLAB is written in which the optimization algorithm is linked with the SWMM hydrologic model. The model is driven with design rainfall as input while the algorithm automatically changes the values of the decision variables to optimize the objective function. In this process, the SWMM model is driven for more than 100,000 realizations. The optimization model is driven for 14 different rainfall scenarios (the computational time is 20 hours approximately with a desktop computer, 32.0 GB RAM, CPU E5-268W v4 @ 3.00 GHz). Variation of the objective function value for the rainfall scenarios is shown in Fig. 4.

22

Objective function value F(x)

0.315

S2

0.31 0.305

S1

0.3

S5 S3

0.295

S9

S7

S6

S11, S14 S4 S10 S8 S13 S12

0.29 0.285 20

30

40

50

60

70

80

90

100

Total rainfall (mm)

Fig. 4. Objective function value (F(x) in Equation 4) for each rainfall scenario

The figure shows that the difference in the objective function values corresponding to different rainfall scenarios is not significant (the maximum difference is observed between scenarios S2 and S13 as 0.022). It means that by changing the amount of precipitation, the optimization model finds the optimal solution so that although the cost and volume of runoff are obtained different, the values of the objective function are still very close. The corresponding runoff volume, LID implementation cost and volume reduction rate designed for different rainfall scenarios based on the simulation-optimization process are shown in Fig. 5. For each scenario, the percentage of the volume reduction is calculated by dividing the runoff volume after the optimized implementation of LIDs, to the maximum runoff volume (when no LID is used).

23

705

600 400 S2

200

S5

S3 S6 S1 S9

S7

S11, S14

S13

S8 S12

Cost (103 $)

Volume (106 lit)

800

S4 S10

700 695

S9

690 685

S3

70

120

S8 S10 S4

20

S13

70

Total rainfall (mm)

120

Total rainfall (mm)

(a) Volume reduction ratio (%)

S11, S14

S6

S5

675 20

S7

S12

680

0

S1

S2

(b)

20% S2

S1 S9

15%

S5 S3

S6

S7

S8 S12

10%

S13

S4 S10

S11, S14

5%

0% 20

30

40

50

60

70

80

90

100

Total rainfall (mm) (c) Fig. 5. a) Runoff volume (106 lit), b) Cost of LIDs’ implementation (103$), c) Runoff volume reduction rate (%) associated with each rainfall scenario

As expected, the total volume of runoff is increasing by the increase in the amount of the rainfall. This increase follows a linear trend (Fig. 5. a). In contrast, by increasing the value of rainfall, the runoff volume reduction rate decreases, with a sharper slope for scenarios of less rainfall value to slighter slope for scenarios of more rainfall (Fig. 5. c). More details for the LIDs type and area from the simulation-optimization results are provided in Table 5. Table 5. Specifications of LIDs type and area for each rainfall scenario; output of the optimization process LID area (m2)

Percent of the LIDs’ area to the:

Scenario S1

VeS*

BRc*

PrPv*

InTr*

Total

Area I **

Area II ***

Area III ****

2245

40453

1231

1105

45034

0.23

0.45

20.8

24

S2

8014

38246

699

1734

48693

0.25

0.49

22.5

S3

1278

40277

697

1409

43661

0.22

0.44

20.2

S4

1289

39993

1112

1282

43677

0.22

0.44

20.2

S5

1435

39473

892

1582

43382

0.22

0.43

20.0

S6

1479

40618

904

1386

44386

0.22

0.44

20.5

S7

1809

40619

1807

1206

45441

0.23

0.45

21.0

S8

2556

40088

629

1380

44652

0.23

0.45

20.6

S9

1738

40479

1042

1328

44588

0.23

0.45

20.6

S10

1830

39762

1372

1706

44670

0.23

0.45

20.6

S11

1221

40796

1343

1223

44583

0.23

0.45

20.6

S12

1182

39916

724

1576

43398

0.22

0.43

20.1

S13

1130

39851

983

1252

43217

0.22

0.43

20.0

S14

1221

40796

1343

1223

44583

0.23

0.45

20.6

Average (%)

4.5

90.1

2.4

3.1

100

0.23

0.45

20.6

* VeS: Vegetative swale, BRc: Bio-retention cell, PrPv: Permeable pavement, InTr: Infiltration trenches ** Area (I): watershed area=19,776,000 m2 *** Area (II): 10 potential sub-catchments areas= 9,991,000 m2 **** Area (III): maximum possible area of LID implementation= 216400 m2

For S1 scenario (without incorporation of climate change impacts to update the design rainfall intensity), by implementation of LIDs on only an area of 0.23% of the total area of the watershed (0.45% of the 10 potential sub-catchments’ areas, or 20.8% of the maximum possible area of LID implementation), with a budget of $697000, the runoff volume could be decreased by 14%. There is also a small difference for the implemented LIDs’ area among the rainfall scenarios, which is because of the definition of objective function and the range of area (lowerand upper-bound) that each LID can have in any subcatchments. The optimization algorithm allocated less than one percent of the area of the sub-catchments to LIDs implementation, which corresponds to almost 20% of the maximum area available for 25

using LIDs. These findings are of great importance indicating how effective is dedicating a small fraction of the watershed area to cost-effective LID practices for stormwater management and further, potential flooding reduction. The last row of Table 5 shows the average percentage of LID types obtained from the optimization model for different scenarios. Insight into the results of the LIDs’ optimal design considering different scenarios shows that the use of bio-retention cells is significantly more than the other types of LIDs. The reason that the model tends to choose bio-retention cells is the low cost of implementation as well as the ability of volume reduction per unit area for this LID technique. So far, the optimization algorithm has been used for obtaining the optimal design of the LIDs in the watershed for different rainfall scenarios taking into account climate change impacts (i.e., different designs for different scenarios). Each rainfall scenario represents a possible future state of the watershed, nevertheless, the probability of occurrence of each scenario under climate change is not known. This brings up the issue of which optimal design (i.e., LIDs’ area and location) should be implemented to ensure meeting the modeling objectives and stormwater management goals in the future. In response to this issue, the following questions are raised and investigated: I)

Does the optimal design based on the analysis of historical rainfall (rainfall scenarios S1 and S8) work well for future rainfall scenarios defined under climate change impacts (single-event optimization approach)?

II)

Compared to the optimum solution based on a single historical rainfall scenario, how efficient would it be to develop a multi-event optimization approach, which considers the possible future rainfall scenarios at once?

26

4.4.3. Single-event optimization approach To answer the first question, the simulation model is driven considering a fixed design of LIDs for the sub-catchments, for all defined rainfall scenarios. This design corresponds to the optimum design (LIDs’ area and location obtained from the optimization model) for the watershed which is obtained based on the current status of rainfall (scenarios S1 and S8, which correspond to 2-year and 20-year historical design rainfalls, respectively, and are hereafter referred to as Design I and Design II). The resultant values of the objective function for each scenario are shown in Fig. 6. This figure also indicates the value of the objective function (optimal solution I) from the optimum designs by solving the optimization model individually for each scenario (as shown in Fig. 4).

Objective function value F(x)

0.300

Optimal Solution I Design I Design II Multi-event optimization

0.250 0.200 0.150

S1

0.100 S8

0.050 0.000 20

30

40

50

60

70

80

90

100

Rainfall (mm) Fig. 6. Scenario S1 and S8 optimum solution (Design I and Design II) comparison with the optimal solution I (the objective function F(x) corresponds to the optimal solution for each rainfall scenario)

As expected, the figure shows that the values of the objective function from the optimum solutions are less than those assuming the historical rainfall (Design I and Design II) for the future. 27

The percentage of decrease/increase in the runoff volume and LIDs’ implementation cost considering LIDs constructed based on Design I and Design II to the optimal design of LIDs

Change in LIDs' cost (%)

from solving the optimization model for each scenario, is shown in Fig. 7. 6% 5% 4% 3% 2% 1% 0% -1% -2% -3% S1

S2

S3

S4

S5

S6

S7

S8

S9

S10

S11

S12

S13

S14

S11

S12

S13

S14

Rainfall scenario

Change in runoff volume (%)

Design I

Design II

Multi-event optimization

0.5% 0.4% 0.3% 0.2% 0.1% 0.0% -0.1% -0.2% -0.3% -0.4% S1

S2

S3

S4

S5

S6

S7

S8

S9

S10

Rainfall scenario Design I

Design II

Multi-event optimization

Fig. 7. Change in LIDs’ cost and runoff volume from optimal solution I (LIDs’ optimum design for each scenario separately), for Design I (LIDs’ optimum design based on rainfall scenario 1), Design II (LIDs’ optimum design based on rainfall scenario 8) and the multi-event optimization (LIDs’ optimal design considering all rainfall scenario together)

28

As observed in this figure, for Design I and Design II, the maximum increase in the runoff volume is 0.33% and 0.44%, and in the cost of LIDs’ implementation is 2.68% and 1.84%, respectively. It means that if Design I is utilized, the maximum changes in the runoff volume for the possible rainfall scenario in the future would be just 0.33 percent of the optimal solution I (the best solution for each rainfall scenario). These differences in the results for the objectives of the optimization model are negligible. Therefore, it can be concluded that for this watershed, designing LIDs by optimizing the objective function considering the analysis obtained from the historical rainfall (scenarios S1 and S8), provides successful results satisfying the possible changes arising from modified rainfall in future due to climate change. This also highlights the adequacy of the developed simulation-optimization model for optimal designing of LIDs to be used for a range of future possible rainfalls taking into account climate change impacts. 4.4.4. Multi-event optimization approach In response to the second research question (exploring that compared to the optimum solution based on a single historical rainfall scenario, how efficient would it be to develop a multi-event optimization approach, which considers the possible future rainfall scenarios at once) optimization algorithm II defined by Equation 5 is used. In this model, the summation of obj1 and obj2 for different rainfall scenarios is minimized. Consequently, the optimal area and location of the LIDs are shown in Fig. 8.

5 7 9 10

11 13 16

19

3000 2000

20 22

3000 2000 1000 0

Sub-catchment 12

24

25

4000

VeS BRc PrPv InTr

18 17

23

4000

0

14

21

5000

1000

12 15

5000 Area (m2)

8

Area (m2)

6

2 1 4 3

29

VeS BRc PrPv InTr

Sub-catchment 13

5000

4000

4000

Area (m2)

Area (m2)

5000 3000 2000 1000

3000 2000 1000 0

0

VeS BRc PrPv InTr

VeS BRc PrPv InTr

2500

2000

2000

2000

1500 1000 500

Area (m2)

2500 1500 1000

1000 0

0

VeS BRc PrPv InTr

1500 500

500

0

VeS BRc PrPv InTr

VeS BRc PrPv InTr

Sub-catchment 17

Sub-catchment 18

Sub-catchment 20

5000

5000

4000

4000

4000

3000 2000 1000

Area (m2)

5000 Area (m2)

Area (m2)

Sub-catchment 15

2500 Area (m2)

Area (m2)

Sub-catchment 14

3000 2000 1000

0

2000 1000

0 VeS BRc PrPv InTr

3000

0 VeS BRc PrPv InTr

VeS BRc PrPv InTr

Sub-catchment 22 Sub-catchment 25 Sub-catchment 27 Fig. 8. Multi-event optimization results, representing LIDs’ type (VeS: Vegetative swale, BRc: Bioretention cell, PrPv: Permeable pavement, InTr: Infiltration trenches) and area in each sub-catchment

The cost of the LIDs’ implementation for this optimum solution is $712,000. Fig. 8 shows that the tendency for using bio-retention cells is more than that for other LIDs. It also indicates that the area of bio-retention cells is close to their maximum values in all sub-catchments. The bio-retention cell’s cost and design parameters defined in the SWMM model, as well as the objective function and constraints defined in the optimization model, are the main reasons for that. Back to Fig. 6, it also shows that how the application of multi-event optimization model (optimization algorithm II), results in finding a unique optimum solution, which satisfies multiple scenarios for different possible rainfall events. The solution (value of the objective 30

function) from multi-event optimization (Yellow line in Fig. 6) is more than those corresponding to the optimal solutions (optimal solution I), and less than those from Design I and Design II (related to scenario 1 and 8), as shown in Fig. 6. It is worth mentioning that solving the multievent optimization model is computationally expensive (12 -days runtime). It means that the multi-event optimization model would propose a unique solution, which is better than design I and II for the all-possible future rainfall scenarios but it will need a running time from 20-hours to 12-days. In addition, Fig. 7. further indicates the variations in LIDs’ design (runoff volume and implementation cost) when Design I and Design II, as well as the design from the multi-event optimization model are used for all scenarios, in comparison to the optimal designs obtained earlier (optimal solution I) for each scenario. Implementing the unique design from the multievent optimization model instead of optimal design individually for each scenario, results in less runoff volume and allocating more budget to build the LIDs. However, these changes are not significant. Maximum change in the runoff volume and implementation cost of LIDs from the solution of multi-event optimization model with those obtained by solving Optimization algorithm I individually for each scenario (optimal solution I), is -0.33% and 4.94%, which is negligible. This further justifies the fact that driving the model (Optimization algorithm II) is time-intensive, although, this time could be decreased using desktops of high-speed configuration. In addition, the solution form the multi-event model suggests more LIDs with more implementation cost and more runoff volume reduction in comparison with the solutions from designs I and II. Results show that the proposed multi-event optimization algorithm can successfully provide a unique solution, which satisfies a combination of different rainfall scenarios considering extreme conditions expected from climate change.

31

5. CONCLUSION This study investigated an optimal design of low impact development practices for urban runoff management considering climate change impacts. A framework was developed for simulation of a watershed in a hydrologic model and then optimization of LIDs’ design to reduce the runoff volume while minimizing the cost of LIDs’ implementation. Both single-event and multi-event optimization algorithms are formulated using Genetic Algorithm for optimal design of the LIDs. This study showed that designing LIDs considering the baseline scenarios obtained from frequency analysis of the historical rainfall, could acceptably result in meeting the watershed management goals and expectations in a future subjected to climate change impacts. The results also indicated that the developed multi-event optimization algorithm is capable of providing a unique optimal solution satisfying the objective functions for multiple climate scenarios all at once. This model provides a unique solution that integrates a variety of extreme climate conditions expected for the watershed from climate change impacts. The results showed that even if no change in rainfall is expected because of climate change, LIDs play an important role in runoff reduction (i.e., about 14% runoff volume reduction by implementing the LIDs in less than 1% of the watershed area) while still functioning properly if rainfall intensity increases due to climate change impacts. It should be mentioned that although recent climate projections from CMIP5 (Coupled Model Inter-comparison Project Phase 5) are available through the IPCC, since the corresponding outputs are not downscaled for the study area (while the CMIP3 downscaled data were already available from another research by the authors), this study benefits from the downscaled projections of CIMP3. However, the climate scenarios (min, max and absolute minimum and maximum) are defined to be representative of possible extreme changes in the 32

watershed. Still, it is recommended to use CMIP5 downscaled information to update/expand this work when the data is available for the study area. This study does not address the impact of changes in extreme events’ durations or frequency, but rather the impacts on single storm events. Moreover, although there is no clear answer to the probability of occurrence of climate change scenarios in the future, absolute minimum and maximum climate change scenarios are incorporated in the analysis to address the uncertainty associated with climate change impacts. Therefore, the optimal solutions designed in this study, address a future expected from the considered climate change models and scenarios.

33

REFERENCES

1. Andersson-Sköld, Y., Thorsson, S., Rayner, D., Lindberg, F., Janhäll, S., Jonsson, A., ... & Granberg, M. (2015). An integrated method for assessing climate-related risks and adaptation alternatives in urban areas. Climate Risk Management, 7, 31-50. 2. Arnell, N. W., & Reynard, N. S. (1996). The effects of climate change due to global warming on river flows in Great Britain. Journal of hydrology, 183(3-4), 397-424. 3. Baek, S. S., Choi, D. H., Jung, J. W., Lee, H. J., Lee, H., Yoon, K. S., & Cho, K. H. (2015). Optimizing low impact development (LID) for stormwater runoff treatment in urban area, Korea: Experimental and modeling approach. Water research, 86, 122-131. 4. Buckley, L. B., Ehrenberger, J. C., & Angilletta, M. J. (2015). Thermoregulatory behaviour limits local adaptation of thermal niches and confers sensitivity to climate change. Functional Ecology, 29(8), 1038-1047. 5. Conway D, Hulme M (1996) The impacts of climate variability and future climate change in the Nile basin on water resources in Egypt. Water Resour Dev 12:277–296. 6. Darand, M., & Sohrabi, M. M. (2018). Identifying drought-and flood-prone areas based on significant changes in daily precipitation over Iran. Natural Hazards, 90(3), 1427-1446. 7. Fallahi Zarandi, A., 2013. Selection of optimized combination for best management practices considering economic issues for improving Tehran urban surface runoff quality. Master Thesis. Kharazmi University, Tehran, Iran (in Persian). 8. Ferrer, J., Pérez-Martín, M. A., Jiménez, S., Estrela, T., & Andreu, J. (2012). GIS-based models for water quantity and quality assessment in the Júcar River Basin, Spain, including climate change effects. Science of the Total Environment, 440, 42-59. 9. Ganguli, P., & Coulibaly, P. (2017). Does nonstationarity in rainfall require nonstationary intensity– duration–frequency curves?. Hydrology and Earth System Sciences, 21(12), 6461-6483.

34

10. Ghodsi, S. H., Kerachian, R., Estalaki, S. M., Nikoo, M. R., & Zahmatkesh, Z. (2016-a). Developing a stochastic conflict resolution model for urban runoff quality management: application of info-gap and bargaining theories. Journal of Hydrology, 533, 200-212. 11. Ghodsi, S. H., Kerachian, R., & Zahmatkesh, Z. (2016-b). A multi-stakeholder framework for urban runoff quality management: Application of social choice and bargaining techniques. Science of The Total Environment, 550, 574-585. 12. Goharian, E., Burian, S. J., Bardsley, T., & Strong, C. (2015). Incorporating potential severity into vulnerability assessment of water supply systems under climate change conditions. Journal of Water Resources Planning and Management, 142(2), 04015051. 13. Jia, H., Yao, H., Tang, Y., Shaw, L. Y., Field, R., & Tafuri, A. N. (2015). LID-BMPs planning for urban runoff control and the case study in China. Journal of environmental management, 149, 65-76. 14. Karamouz, M., & Nazif, S. (2013). Reliability-based flood management in urban watersheds considering climate change impacts. Journal of Water Resources Planning and Management, 139(5), 520-533. 15. Karamouz, M., Zahmatkesh, Z., Goharian, E., & Nazif, S. (2014). Combined impact of inland and coastal floods: Mapping knowledge base for development of planning strategies. Journal of Water Resources Planning and Management, 141(8), 04014098. 16. Lee, J. Y., Lee, M. J., & Han, M. (2015). A pilot study to evaluate runoff quantity from green roofs. Journal of environmental management, 152, 171-176. 17. Loiola, C., Mary, W., & da Silva, L. P. (2018). Hydrological performance of modular-tray green roof systems for increasing the resilience of mega-cities to climate change. Journal of Hydrology.

18. Macro, K., Matott, L. S., Rabideau, A., Ghodsi, S. H., & Zhu, Z. (2018). OSTRICH-SWMM: A New Multi-Objective Optimization Tool for Green Infrastructure Planning with SWMM. Environmental Modelling & Software. 19. Mahab Ghods Consulting Engineering Company, 2012. Integrated Plan of Tehran Surface Runoff Management. Tehran, Iran (in Persian). 35

20. Mondal, A., & Mujumdar, P. P. (2015). Modeling non-stationarity in intensity, duration and frequency of extreme rainfall over India. Journal of Hydrology, 521, 217-231. 21. Newcomer, Michelle E., Jason J. Gurdak, Leonard S. Sklar, and Leora Nanus. "Urban recharge beneath low impact development and effects of climate variability and change." Water resources research 50, no. 2 (2014): 1716-1734. 22. Panagoulia, D., & Dimou, G. (1997). Sensitivity of flood events to global climate change. Journal of Hydrology, 191(1-4), 208-222. 23. Prudhomme, C., Reynard, N., & Crooks, S. (2002). Downscaling of global climate models for flood frequency analysis: where are we now?. Hydrological processes, 16(6), 1137-1150. 24. Reynard, N. S., Prudhomme, C., & Crooks, S. M. (2001). The flood characteristics of large UK rivers: potential effects of changing climate and land use. Climatic change, 48(2-3), 343-359. 25. Rossman, L.A., 2015. Storm Water Management Model User’s Manual, Version 5.1. National Risk Management Laboratory Office of Research and Development, U.S. Environmental Protection Agency. 26. Seidel, K., Ehrler, C., & Martinec, J. (1998). Effects of climate change on water resources and runoff in an Alpine basin. Hydrological Processes, 12(10), 1659-1669. 27. Soltani, M., 2009. Quality based modeling of inland channels. Master Thesis. Sharif University of Technology, Tehran, Iran (in Persian). 28. Switzman, H., Razavi, T., Traore, S., Coulibaly, P., Burn, D. H., Henderson, J., ... & Ness, R. (2017). Variability of future extreme rainfall statistics: comparison of multiple IDF projections. Journal of Hydrologic Engineering, 22(10), 04017046. 29. Vaze, J., Post, D. A., Chiew, F. H. S., Perraud, J. M., Viney, N. R., & Teng, J. (2010). Climate nonstationarity–validity of calibrated rainfall–runoff models for use in climate change studies. Journal of Hydrology, 394(3-4), 447-457.

36

30. Yang, Y., & Chui, T. F. M. (2018). Optimizing surface and contributing areas of bio-retention cells for stormwater runoff quality and quantity management. Journal of Environmental Management, 206, 1090-1103. 31. Yoon, S. K., Kim, J. S., & Moon, Y. I. (2015). Urban stream overflow probability in a changing climate: Case study of the Seoul Uicheon Basin, Korea. Journal of Hydro-environment Research. 32. Zahmatkesh, Z., Burian, S. J., Karamouz, M., Tavakol-Davani, H., & Goharian, E. (2014a). Lowimpact development practices to mitigate climate change effects on urban stormwater runoff: Case study of New York City. Journal of Irrigation and Drainage Engineering, 141(1), 04014043. 33. Zahmatkesh, Z., Karamouz, M., Goharian, E., & Burian, S. J. (2014b). Analysis of the effects of climate change on urban storm water runoff using statistically downscaled precipitation data and a change factor approach. Journal of Hydrologic Engineering, 20(7), 05014022. 34. Zhou, Q., Leng, G., & Huang, M. (2018). Impacts of future climate change on urban flood volumes in Hohhot in northern China: benefits of climate change mitigation and adaptations. Hydrology and Earth System Sciences, 22(1), 305-316. 35. Zuo, D., Xu, Z., Yao, W., Jin, S., Xiao, P., & Ran, D. (2016). Assessing the effects of changes in land use and climate on runoff and sediment yields from a watershed in the Loess Plateau of China. Science of The Total Environment, 544, 238-250.

37

Complementary: LID information The LIDs’ parameters and information, which have been used in the developed SWMM model, are presented in Table A1. Table A1. LIDs’ parameters in the SWMM model

LID type Bio-retention cell

Layer Surface

Soil

Parameter Berm Height: 120 mm

Vegetation Volume Fraction: 0.2

Surface Roughness: 0

Surface Slope: 0

Thickness: 1200 mm

Porosity: 0.5

Field Capacity: 0.2

Wilting Point: 0.1

Conductivity: 44 mm/hr

Conductivity slope: 10

Suction Head: 90 mm Storage

Drain

Thickness: 800 mm

Void Ratio: 0.5

Seepage Rate: 20 mm/hr

Clogging Factor: 0

Flow Coefficient: 0

Flow Exponent: 0.5

Offset Height: 0 mm

Vegetative Swale

Surface

Berm Height: 750 mm

Vegetation Volume Fraction: 0.2

Surface Roughness: 0.24

Surface Slope: 1 percent

Swale Side Slope (run/rise): 5

Permeable Pavement

Surface

Pavement

Berm Height: 0 mm

Vegetation Volume Fraction: 0.1

Surface Roughness: 0.14

Surface Slope: 2 percent

Thickness: 150 mm

Void Ratio (voids/solids): 0.15

Impervious Surface Fraction: 0

Permeability: 2540 mm/hr

Clogging Factor: 0 Soil

Thickness: 0 mm

Porosity: 0.5

Field Capacity: 0.2

Wilting Point: 0.1

Conductivity: 0.5 mm/hr

Conductivity slope: 10

Suction Head: 3.5 Storage

Drain

Thickness: 300 mm

Void Ratio: 0.75

Seepage Rate: 5 mm/hr

Clogging Factor: 0

Flow Coefficient: 0

Flow Exponent: 0.5

38

Offset Height: 0 mm

Infiltration Trench

Surface

Storage

Drain

Berm Height: 0

Vegetation Volume: 0

Surface Roughness: 0.14

Surface Slope: 0 percent

Thickness: 200 mm

Void Ratio (voids/solids): 0.75

Seepage Rate: 5 (mm/hr)

Clogging Factor: 0

Flow Coefficient: 0

Flow Exponent: 0.5

Offset Height: 0 mm

http://www.maphill.com

6

Iran

2 1 4 3

8 9 11 13

5 7 10 12 14

15

16 19

18 17 21

Chitgar lake

Velenjak watershed

20

24

Tehran

23

22

25 Rainfall Station Runoff collection system 0 1 28.4214.03 3 5

potential sub-catchments for LID implementation

10 km 28.06

0 100 300 500

1000 m

Fig. 9. Study area (Velenjak watershed) in the northeast part of Tehran, Iran.

39

26

27

Analyzing climate change impacts on watershed runoff management 1- Data collection and processing 1-1- Drainage system characteristics

1-2- Historical rainfall

1-3- Determination of design rainfall

1-4- Historical runoff

1-5- Information of LIDs

2- Rainfall-runoff model SWMM for runoff quantity simulation

3- Urban runoff management Selecting appropriate LID controls

4- Climate change impact modeling 4-1- GCM outputs for rainfall

4-2- Rainfall downscaling

4-3- Identifying extreme climate scenarios

5- Optimization model 5-1- Objective function, constraints and decision variables

5-2- Genetic Algorithm (GA) optimization model

Fig. 10. Flowchart of the proposed methodology for urban runoff management under climate change impacts

40

12 20-year

Rainfall (mm/h)

10

2-year

8 6 4 2 0 0

2

4

6

8

10

12

14

16

18

20

22

24

Time (h) Fig. 11. Design rainfalls hyetographs with return period of 2-year and 20-year (Ghodsi et al. 2016-a)

41

Objective function value F(x)

0.315

S2

0.31 0.305

S1

0.3

S5 S3

0.295

S9

S7

S6

S11, S14 S4 S10 S8 S13 S12

0.29 0.285 20

30

40

50

60

70

80

Total rainfall (mm)

Fig. 12. Objective function value (F(x) in Equation 4) for each rainfall scenario

42

90

100

705

600 S13

Cost (103 $)

Volume (106 lit)

800 S11, S14

S8 S12 S3 S6 S4 S10 S5 S7 S2 S1 S9

400 200

700 695

S9

690 685

S3

70

120

S6 S8 S10 S4

20

S13

70

Total rainfall (mm)

120

Total rainfall (mm)

(a) Volume reduction ratio (%)

S11, S14

S5

675 20

S7

S12

680

0

S1

S2

(b)

20% S2

S1 S9

15%

S5 S3

S6

S7

S8 S12

10%

S13

S4 S10

S11, S14

5%

0% 20

30

40

50

60

70

80

90

100

Total rainfall (mm) (c) Fig. 13. a) Runoff volume (106 lit), b) Cost of LIDs’ implementation (103$), c) Runoff volume reduction rate (%) associated with each rainfall scenario

43

Objective function value F(x)

0.300

Optimal Solution I Design I Design II Multi-event optimization

0.250 0.200 0.150

S1

0.100 S8

0.050 0.000 20

30

40

50

60

70

80

90

100

Rainfall (mm) Fig. 14. Scenario S1 and S8 optimum solution (Design I and Design II) comparison with the optimal solution I (the objective function F(x) corresponds to the optimal solution for each rainfall scenario)

44

Change in LIDs' cost (%)

6% 5% 4% 3% 2% 1% 0% -1% -2% -3% S1

S2

S3

S4

S5

S6

S7

S8

S9

S10

S11

S12

S13

S14

S11

S12

S13

S14

Rainfall scenario

Change in runoff volume (%)

Design I

Design II

Multi-event optimization

0.5% 0.4% 0.3% 0.2% 0.1% 0.0% -0.1% -0.2% -0.3% -0.4% S1

S2

S3

S4

S5

S6

S7

S8

S9

S10

Rainfall scenario Design I

Design II

Multi-event optimization

Fig. 15. Change in LIDs’ cost and runoff volume from optimal solution I (LIDs’ optimum design for each scenario separately), for Design I (LIDs’ optimum design based on rainfall scenario 1), Design II (LIDs’ optimum design based on rainfall scenario 8) and the multi-event optimization (LIDs’ optimal design considering all rainfall scenario together)

45

5 7 9 10

16 19

4000

4000

3000 2000

VeS BRc PrPv InTr

VeS BRc PrPv InTr

Sub-catchment 12

18 17 21

20

24 23

22

25

Sub-catchment 13

5000

5000

4000

4000

3000 2000 1000

26

2000 1000 VeS BRc PrPv InTr

VeS BRc PrPv InTr

27

Sub-catchment 14

1000 m

Sub-catchment 15

2500

2500

2000

2000

2000

1500 1000 500

Area (m2)

2500 Area (m2)

1500 1000

1000 0

0

VeS BRc PrPv InTr

1500 500

500

0

VeS BRc PrPv InTr

VeS BRc PrPv InTr

Sub-catchment 17

Sub-catchment 18

Sub-catchment 20

5000

5000

4000

4000

4000

Area (m2)

5000

3000 2000 1000

Area (m2)

Area (m2)

0 100 300 500

3000

0

0

Area (m2)

2000 0

0

14

3000 1000

1000

12 15

5000

Area (m2)

11 13

5000 Area (m2)

Area (m2)

8

Area (m2)

6

2 1 4 3

3000 2000 1000

0

2000 1000

0 VeS BRc PrPv InTr

3000

0 VeS BRc PrPv InTr

VeS BRc PrPv InTr

Sub-catchment 22 Sub-catchment 25 Sub-catchment 27 Fig. 16. Multi-event optimization results, representing LIDs’ type (VeS: Vegetative swale, BRc: Bioretention cell, PrPv: Permeable pavement, InTr: Infiltration trenches) and area in each sub-catchment

46

Table 6. List of the climate models used to define the rainfall scenarios Original resolution Model number

Model name

Scenarios

Country Long × Lat levels (L)

1

CSIRO Mk3.0

A2, A1B, B1

Australia

192×96 L18

2

ECHOG-G

A2, A1B

Germany

96×48 L19

3

GFDL CM2.0

A2, A1B, B1

United States

144×90 L24

4

GFDL CM2.1

A2, A1B, B1

United States

144×90 L24

5

GISS-ER

A2, B1

United States

72×46 L20

6

HadGEM1

A2, A1B

United Kingdom

192×145 L38

7

INM-CM3.0

A2, A1B, B1

Russia

72×45 L21

8

IPSL-CM3

A2, A1B, B1

France

96×72 L19

9

MIROC3.2

A2, A1B, B1

Japan

128×64 L20

10

ECHAM5

A2, A1B, B1

Germany

192×96 L31

11

MRI-CGCM2.3.2

A2, A1B, B1

Japan

128×64 L30

12

CCSM3

A2, A1B, B1

United States

156×128 L26

13

PCM

A2, A1B

United States

128×64 L26

14

HadCM3

A2, A1B, B1

United Kingdom

96×73 L19

47

Table 7. Total runoff, impervious area, and the defined index for the sub-catchments in the initial condition, regarding finding the potential sub-catchments to implement LIDs Runoff Sub*

Runoff % Imperv**

Index***

Sub

(mm)

% Imperv

Index

(mm)

1

0

0

0

15

19.08

62

1183

2

0

0

0

16

11.39

37

421

3

0

0

0

17

22.02

72

1585

4

0

0

0

18

20.56

67

1378

5

0

0

0

19

11.38

37

421

6

0

0

0

20

25.14

82

2061

7

0

0

0

21

2.47

32

79

8

0

0

0

22

25

82

2050

9

0

0

0

23

1.98

32

63

10

2.17

7

15

24

1.97

32

63

11

0

0

0

25

21.84

72

1572

12

22.17

72

1596

26

11.43

37

423

13

20.59

67

1380

27

20.5

67

1374

14

23.66

77

1822

* Sub: Sub-catchment

48

Table 8. Rainfall scenarios; cumulative amount of rainfall (mm) for design scenarios based on the current status of the watershed (BAU) and under the impact of climate change Future scenarios

Baseline Return period scenarios

Min P A-Min P

A-Mean P

Mean P

Max P

A-Max P MRCGCM (B1)

HADCM3 (A1B)

GFCM21 (B1)

S1 (BAU)

S2

S3

S4

S5

S6

S7

36.6

24.5

37.1

55.0

33.4

39.6

49.3

S8 (BAU)

S9

S10

S11

S12

S13

S14

54.0

37.6

58.1

90.6

56.5

63.6

90.6

2-year

20-year

49

Table 9. The maximum and minimum values of runoff volume for each rainfall scenario Rainfall

Rainfall

Vmin

Vmax

Rainfall

Rainfall

Vmin

Vmax

scenario

(mm)

(106 L)

(106 L)

scenario

(mm)

(106 L)

(106 L)

S1

36.6

214.1

263.7

S8

54

355.9

413.1

S2

24.5

120.3

160.3

S9

37.6

222

272.3

S3

37.1

218

267.9

S10

58.1

389.2

448

S4

55

363.7

421.6

S11

90.6

662.2

727.7

S5

33.4

189

236.6

S12

56.5

376.5

434.9

S6

39.6

238.6

289.9

S13

63.6

434.6

495.4

S7

49.3

317.4

372.9

S14

90.6

662.2

727.7

50

Table 10. Specifications of LIDs type and area for each rainfall scenario; output of the optimization process LID area (m2)

Percent of the LIDs’ area to the:

Scenario VeS*

BRc*

PrPv*

InTr*

Total

Area I **

Area II ***

Area III ****

S1

2245

40453

1231

1105

45034

0.23

0.45

20.8

S2

8014

38246

699

1734

48693

0.25

0.49

22.5

S3

1278

40277

697

1409

43661

0.22

0.44

20.2

S4

1289

39993

1112

1282

43677

0.22

0.44

20.2

S5

1435

39473

892

1582

43382

0.22

0.43

20.0

S6

1479

40618

904

1386

44386

0.22

0.44

20.5

S7

1809

40619

1807

1206

45441

0.23

0.45

21.0

S8

2556

40088

629

1380

44652

0.23

0.45

20.6

S9

1738

40479

1042

1328

44588

0.23

0.45

20.6

S10

1830

39762

1372

1706

44670

0.23

0.45

20.6

S11

1221

40796

1343

1223

44583

0.23

0.45

20.6

S12

1182

39916

724

1576

43398

0.22

0.43

20.1

S13

1130

39851

983

1252

43217

0.22

0.43

20.0

S14

1221

40796

1343

1223

44583

0.23

0.45

20.6

Average (%)

4.5

90.1

2.4

3.1

100

0.23

0.45

20.6

* VeS: Vegetative swale, BRc: Bio-retention cell, PrPv: Permeable pavement, InTr: Infiltration trenches ** Area (I): watershed area=19,776,000 m2 *** Area (II): 10 potential sub-catchments areas= 9,991,000 m2 **** Area (III): maximum possible area of LID implementation= 216400 m2

51

Analyzing climate change impacts on watershed runoff management 1- Data collection and processing 1-1- Drainage system characteristics

1-2- Historical rainfall

1-3- Determination of design rainfall

1-4- Historical runoff

1-5- Information of LIDs

2- Rainfall-runoff model SWMM for runoff quantity simulation

3- Urban runoff management Selecting appropriate LID controls

4- Climate change impact modeling 4-1- GCM outputs for rainfall

4-2- Rainfall downscaling

4-3- Identifying extreme climate scenarios

5- Optimization model 5-1- Objective function, constraints and decision variables

5-2- Genetic Algorithm (GA) optimization model

Find an optimal solution for LIDs’ design under climate scenarios. Satisfaction of watershed management goals subjected to climate change impacts. 18% runoff volume reduction by allocating 0.23% of the area to implement LIDs. The unique solution meets the goals for all considered climate scenarios at once. 52

53