A new framework for modeling decentralized low impact developments using Soil and Water Assessment Tool

Environmental Modelling & Software 96 (2017) 305e322

Contents lists available at ScienceDirect

Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft

A new framework for modeling decentralized low impact developments using Soil and Water Assessment Tool Younggu Her a, *, Jaehak Jeong b, Jeffrey Arnold c, Leila Gosselink d, Roger Glick d, Fouad Jaber e a

Department of Agricultural and Biological Engineering & Tropical Research and Education Center, University of Florida, United States Department of Agricultural and Biological Engineering & Texas A&M AgriLife Blackland Research and Extension Center, Texas A&M University, United States c Grassland Soil and Water Research Laboratory, USDA-ARS, United States d Watershed Protection Department, City of Austin, United States e Department of Agricultural and Biological Engineering & Texas A&M AgriLife Research and Extension Center at Dallas, Texas A&M University, United States b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 26 August 2016 Received in revised form 5 April 2017 Accepted 15 June 2017

Assessing the performance of LID practices at a catchment scale is important in managing urban watersheds. Few modeling tools exist that are capable of explicitly representing the hydrological mechanisms of LIDs while considering the diverse land uses of urban watersheds. In this paper, we propose computational modules that simulate the hydrological processes of LIDs including green roof, rain garden, cistern, and porous pavement. The applicability of the modules was evaluated using plot scale experimental monitoring data. The effectiveness of LIDs was investigated in a highly urbanized watershed located in Austin, TX. Results indicate that the performance of LIDs is sensitive to LID configurations, application areas, and storm event characteristics, suggesting the need for studies on spatial optimization of LIDs and critical storm events to maximize the utility of LIDs. The LID modules offer a comprehensive modeling framework that explicitly simulates the water quantity processes of the LIDs considering landscape heterogeneity. © 2017 Elsevier Ltd. All rights reserved.

Keywords: Low impact development Soil and Water Assessment Tool Green roof Rain garden Cistern Porous pavement

1. Introduction Low impact development (LID) practices are designed and implemented to reduce stormwater runoff at the source level, which subsequently leads to decreased velocity, prolonged travel time of downstream runoff, and then reduced pollutant loading to downstream areas. LIDs have been widely employed as measures to mitigate urbanization impacts on water quantity and quality (Dietz, 2007; Roy et al., 2008; Ahiablame et al., 2012). Estimating effectiveness of LIDs is a process necessary for developing stormwater management plans to improve urban water environment (Gilroy and McCuen, 2009). Although many studies reported effectiveness of LIDs with wide variations in reducing the amount of runoff and pollutants (Ahiablame et al., 2012; Li and Babcock, 2014), Only

* Corresponding author. 18905 SW 280 St., Homestead, FL 33031-3314, United States. E-mail address: yher@ufl.edu (Y. Her). http://dx.doi.org/10.1016/j.envsoft.2017.06.005 1364-8152/© 2017 Elsevier Ltd. All rights reserved.

few research has been conducted on watershed-scale effects due in part to the lack of tools for simulating LIDs using process-based modeling techniques at this scale (Elliott and Trowsdale, 2007; Roy et al., 2008; Gilroy and McCuen, 2009). LID modeling tools are expected to be refined to better address the increasing complexity of urban landscape and stormwater infrastructure, let alone the growing popularity of LIDs implementation to control stormwater in urban watersheds (Brander et al., 2004; Elliott and Trowsdale, 2007; Hood et al., 2007; Freni et al., 2010; Ahiablame et al., 2013; Loperfido et al., 2014; Palla and Gnecco, 2015). Mathematical models are critical and efficient tools that help better understand hydrological processes occurring at various spatial and temporal scales, and they have been commonly used to investigate effects of land use changes on watershed hydrology and water quality. Each model employs its own unique approaches and mechanisms to represent landscapes and simulate hydrological processes of a watershed, thus incorporating LID practices into the models would require different strategies. Technical Release 20/55 (TR-20/55) and Hydrologic Engineering Center e Hydrologic

306

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

Modeling System (HEC-HMS) are event-based models developed for simulating surface runoff hydrographs of large storms without continuous soil water accounting (USDA, 1986; USACE, 2000). Since they represent the watershed landscape in a lumped way at a subwatershed level, it is not allowed to place LIDs in specific land uses. Long-Term Hydrologic Impact Assessment-Low Impact Development (L-THIA-LID) is a screening model to assess the relative effectiveness of LIDs quickly by comparing watershed responses to land use change or LID scenarios represented using runoff curve numbers in a highly lumped way (USDA, 1986; Ahiablame et al., 2012). Hydrologic Simulation Program e Fortran (HSPF) can consider various land use types in modeling, but it does not explicitly consider watershed-level runoff routing that is an important transport process controlling hydrological responses of a watershed to a storm event (Bicknell et al., 1996). Also, its lumped representation of watershed landscape restricts its capability to consider the spatial heterogeneity of the landscape in hydrological modeling. Storm Water Management Model (SWMM) is one of the most widely used models to simulate urban runoff and water quality of a small lot or block considering the sewer system (Rosa et al., 2015). However, a study area is represented with only two land types, pervious and impervious covers, and detailed hydrological processes are not explicitly incorporated in the model (Rossman, 2010; Burszta-Adamiak and Mrowiec, 2013). Thus, its capability of representing complicated urban watershed landscapes consisting of several levels of urbanizations would be greatly limited, and subsequently, it becomes difficult to show different hydrologic responses to the variety of urban land uses using the model. Simulation strategies of SWMM and HSPF were integrated into the simulation modules of System for Urban Stormwater Treatment and Analysis INtegration (SUSTAIN; Shoemaker et al., 2009). However, the oversimplified land use representation of SWMM was adopted in SUSTAIN without an improvement (Shoemaker et al., 2009). Useful reviews on current models for simulating LIDs can be found in the literature (Zoppou, 2001; Obropta and Kardos, 2007; Elliott and Trowsdale, 2007; Li and Babcock, 2014). SWAT was developed to simulate the impacts of land use management practices on hydrology and water quality processes, initially for an agricultural watershed (Arnold et al., 1998; Gassman et al., 2007). The model is capable of simulating hydrological processes at multiple spatial scales, from HRUs to subwatersheds and watersheds, which allows detailed descriptions of heterogeneous watershed landscape and processes and placements of LIDs at the HRU level (Her et al., 2015). In addition, SWAT allows continuous simulation of watershed hydrology by accounting soil water with consideration of infiltration, evapotranspiration, and percolation between storm events (Arnold et al., 1998; Neitsch et al., 2011). Recent enhancements for sub-hourly simulation expanded its utility to urban stormwater modeling (Jeong et al., 2010, 2011, 2013; Kannan et al., 2014), and the model has become a good alternative to other models in assessing long-term urban watershed processes in a distributed and process-based manner. Thus, it would be encouraged to take a benefit of the utilities and strengths of SWAT in developing modeling tools for spatial and temporal analysis of urban watershed hydrology. The objectives of this study are to develop algorithms for (1) simulating hydrological processes of spatially distributed LIDs such as green roof, rain garden, cistern, and porous pavement and (2) assessing their effectiveness in reducing stormwater discharge volume and rate at the field and watershed scales. Runoff hydrographs simulated using the LID modules and observed from field experiments were compared to assess the validity of the LID modules. Also, the sensitivity of simulated discharge to LID parameters was investigated to assess the hydrological behavior of

the modules. Twenty-six LID implementation scenarios were developed and evaluated to find a correlation between the watershed-scale effectiveness of LIDs and their application areas.

2. Methods and materials 2.1. Study watershed and input data preparation A SWAT model was prepared to simulate the dynamic runoff hydrographs in the Brentwood watershed, which is a highly urbanized catchment (149.8 ha) in Austin, TX (Fig. 1). Austin is the capital city of Texas, which has recently been growing as fast as any other cities in the United States. The city of Austin is keen on protecting the environment and mitigating flood risks in populated urban areas. The watershed has ten land uses based on an in-house database (“LANDUSE 12”: City of Austin, 2012) of the City of Austin (Table 1). A fraction of each land cover directly connected to the impervious area was calculated using citywide sampling data for the total impervious area (TIA) of each land use type and using the method Sutherland (2000) proposed to estimate effective impervious areas (EIA) from TIA (Table 1). Soil properties were obtained from the SSURGO database (USDA, 2015). Streamflow has been monitored by the City of Austin, where they installed two

Fig. 1. Land uses elevation, channel networks, and SWAT subbasins of the study watershed.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

2.2. Modules for LIDs simulation

Table 1 Land uses of the Brentwood watershed. Codes

Description

Fraction of directly connected impervious areas

Area (ha)

A120 A200 A300 A400 A500 A600 A710 A800

Res-High Density Multi-Family Commercial Office Industrial Civic Parks & Recreation Transportation & Infrastructure Right-of-Way Undeveloped

0.259 0.542 0.645 0.536 0.514 0.248 0.009 0.135

79.8 6.5 10.8 1.6 2.4 15.7 2.0 3.0

0.650 0.001

27.7 0.2

A860 A900

307

monitoring stations, one at the watershed outlet and one at a subbasin outlet in the middle of the watershed, since February 2012 to present (Fig. 1). Each station is equipped with an automatic ISCO sampler and bubbler stage sensors. There are two 1/100-inch tipping-bucket rain gauges installed at the locations of the flow gages. Daily temperature data were obtained from a temperature gauge of the Mueller airport located six kilometers away from the watershed. Topographic information was obtained from 0.6-m (2ft) LIDAR DEM created by City of Austin, which was used to delineate the watershed boundaries and to identify the drainage network (Fig. 1) in the watershed. According to the LIDAR DEM, the length of the longest flow path is 3.1 km, and the average slope is 3.42%. In the SWAT model, the watershed was divided into 137 subbasins according to the drainage networks and further split into 1212 HRUs with the 0-5-5 threshold: 0% for land uses, 5% for soils, and 5% for slopes (Fig. 1). The HRU threshold (0-5-5) applied would keep all the land use types but ignore minor soils and slopes in defining HRUs for improved modeling efficiency (Her et al., 2015). About 22.5% of the annual rainfall (¼ 918 mm) discharges to the watershed outlet, varying between zero to 47.1% between months (Fig. 2). There is little baseflow or groundwater contribution to streamflow due in part to the small size of the watershed and also to the high fraction of impervious cover. Comparison of 15-min rainfall and streamflow exhibits 15-min to 30-min lag time, implying that the time of concentration ranges from 25 to 50 min (USDA, 2010).

2.2.1. Understanding of LID processes In this study, we focus on the four most commonly implemented LIDs in the study watershed, including a green roof, rain garden, cistern, and porous pavement. A green roof is a vegetation layer planted on a building rooftop to detain stormwater in its soil layer, then slowly release the stormwater through its bottom (seepage). After saturating the soil layer, all raindrops on the green roof bypass seepage and run off. The moisture content of the soil layer is reduced with plant uptake and soil evaporation. The evapotranspiration and seepage rates are functions of soil moisture content varying from field capacity to wilting point of the soil. A cistern is a container installed on (or in) the ground to reduce the amount of stormwater volume by storing a part of stormwater at the source's location. Computationally, a cistern is represented with a single tank that does not allow seepage and evaporation of stored water. Irrigation is the only way to drain the water. A cistern can be connected to the green roof, and receive water seeped from the amended soil layer of a green roof. A rain garden is an artificial surface depression where stormwater can be stored and infiltrate. A typical rain garden consists of two storage components (tanks): storage on the garden surface (upper tank) and an amended soil layer (lower tank). Water stored in the surface storage infiltrates into the soil layer and then percolates into the native soil. Since a rain garden receives water draining from its upstream impervious drainage areas within an HRU as well as raindrops that fall directly on its surface, water ponding may occur frequently. Some of the water stored in the upper tank can discharge through an orifice pipe. A discharge orifice pipe allows quick drainage of the stored water and reduces the time of submergence of vegetation plants in the garden. Water stored in the upper and lower storage can be evapo-transpired into the air between storm events. A porous pavement is an alternative paved-surface that allows infiltration of rainfall and surface water to reduce the volume of stormwater generated on site. There are many different types of porous pavement (Mullaney and Lucke, 2014), so the system is simplified to a two-layer system consisting of a gravel layer and a soil base located between the pavement and the native soil. Raindrops falling on the pavement surface infiltrate into the gravel layer and then move down to the soil base, and water stored in the tanks can be percolated into the native soil

Fig. 2. Monthly rainfall, runoff, and water yield of the Brentwood watershed.

308

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

or discharged through a drainage pipe installed in the soil base. As evaporation from a porous pavement is not significant, its contribution to water balance can be ignored (Brown and Borst, 2015). 2.2.2. Representation of hydraulic processes In this study, we implement four LIDs including a green roof, rain garden, cistern, and porous pavement. In the SWAT LID module, each LID is conceptualized with tanks connected to each other or isolated, representing its function as storage or to temporally hold and/or infiltrate stormwater (Fig. 3). Regarding the scale of spatial representation, these LIDs are formulated at the HRU level. Therefore, the hydraulic mechanism of LIDs is represented with details such as disproportional distributions over different land uses or variable spatial densities in the watershed. It must be worth noting that topography, land use, and soil are assumed uniform in an agricultural HRU, but an urban HRU consists of pervious and impervious parts that are connected or disconnected to the pervious part. In the LID modules, a LID is located in the impervious area of an existing urban HRU and connected to its pervious areas. Thus, the impervious areas covered or treated by the LID would be a part of the pervious areas in the HRU, and HRU-level outputs represent the overall responses of an urban HRU comprising of the heterogeneous parts, pervious and impervious areas. The depth of surface runoff generated in the treated areas of an HRU is calculated considering the mechanisms of LIDs (Fig. 3) placed in the HRU, and it is added to the depth of surface runoff calculated for the untreated impervious and pervious parts of the HRU considering their fractions to the entire HRU area so as to calculate the overall depth of surface runoff generated in the HRU. Based on the continuity equations developed for the ground surface and soil profile of an HRU, then, amount of infiltrated water is calculated from the overall depth (or volume) of the surface runoff

generated in the HRU, and soil water is updated with the infiltrated water volume on a daily basis (Neitsch et al., 2011). The LID modules are integrated into the sub-daily SWAT to capture dynamic hydrographs of urban catchments where LIDs are implemented. The modules require physical properties of LID practices including the areal fraction of LIDs installed to each urban land use, storage characteristics such as hydraulic conductivity of amended soils. In each time step, runoff depths in an urban HRU is calculated considering the fractions of pervious cover and impervious cover, and the contributing drainage area of LIDs. Specifically, the depth of runoff generated in the pervious part of an urban HRU is calculated using the Green-Ampt equation. In the case of the impervious part, the depth of precipitation, excluding the initial abstract, is regarded as the runoff depth. Once both runoff depths are calculated separately, the overall runoff depth of the HRU is calculated proportionately to the areas of the parts (Fig. 4). When a LID is placed in the impervious part, the runoff depth of the part is proportionally decreased by the size of areas treated by the LID (Fig. 4). In SWAT, surface runoff lag is calculated using a first order equation, which is modified to enable subhourly surface runoff simulation (Neitsch et al., 2011; Jeong et al., 2010). Then, hydrographs are constructed based on the runoff volumes estimated from HRUs at every time step by multiplying runoff volume ðQ Þ to the ordinates of a dimensionless unit hydrograph estimated by either the SCS triangular unit hydrograph method, or a gamma function distribution method. Once all contributing runoff volumes are distributed forward the time step, surface runoff hydrograph for the subbasin is then calculated by superimposing these distributed runoff volumes that are allocated to the same time step (Jeong et al., 2010). Details of simulation processes and equations adapted in the LID modules are presented in the following sub-sections.

Fig. 3. Storages of the LID practices (ET: evapotranspiration, INF: infiltration, Excess R: excess rainfall, Seepage: seepage of soil water, Bypass: bypass of stormwater, Orifice: discharge of stormwater through an orifice, PRC: percolation of soil water, IRR: irrigation and DRN: lateral drainage of soil water).

Fig. 4. Example calculation of runoff depth for an urban HRU with a LID. The upper figure represents an urban HRU that does not have a LID, and the lower figure shows a case of having a LID that treats 20% of the impervious part (12% of the HRU). Numbers and symbols in bold signify calculations related to the LID.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

2.2.3. Green roofs Infiltration, percolation, and evapotranspiration are the main processes involved with green roofs. Infiltration rate is calculated using the Green-Ampt Mein-Larson (GAML) equation (Jeong et al., 2010),


Df IðtÞ ¼ Ke 1 þ J$ Iacc ðtÞ

(1)

where I is the infiltration rate (mm/hour), t is the current simulation time step, Ke is an effective hydraulic conductivity in which land cover impact (curve number) is considered (mm/hour; Nearing et al., 1996), J is the wetting front matric potential (mm), and Df is a change in soil water content (mm/mm). The GAML equation makes the original Green-Ampt (GA) equation practically applicable by assuming that the infiltration rate equals the rainfall intensity until surface ponding occurs. Thus, the GAML equation does not include the term of water ponding depth found in the original form of the GA equation. In SWAT, the GAML equation is modified to use effective hydraulic conductivity ðKe Þ instead of saturated hydraulic conductivity ðKsat Þ in order to consider unsaturated soil conditions in the calculation of infiltration rate. Here, Ke is determined based on soil water content so that unsaturated soil conditions are properly reflected in the calculation of infiltration rates. When soil is saturated, Ke becomes equal to the saturated hydraulic conductivity ðKsat Þ. Otherwise, it is less than Ksat as a function of daily CN and soil water content. In the GAML equation, infiltration rate varies over time with the wetting front progresses into deeper soil. When infiltration rate is greater than rainfall intensity, the total rainfall volume during the time step infiltrates into the soil. If rainfall intensity is greater than infiltration rate, the current infiltration rate determines the infiltration volume and the rainfall volume that exceeds the infiltration rate tends to make surface runoff. Cumulative volume of water infiltrated after ponding (Iacc in mm) is calculated using the following equation:

Iacc ðtÞ ¼ Iacc ðt  1Þ þ Ke $Dtþ   ðIacc ðtÞ þ J$DfÞ J$Df$ln ðIacc ðt  1Þ þ J$DfÞ

(2)

where t  1 is the previous simulation time step. Equation (2) is solved using a successive substitution technique, and then the infiltration rate is calculated based on the cumulative infiltration volume using Equation (1) for each time step. Evapotranspiration occurs on the soil surface and through vegetated grasses so that soil water content would decrease. An actual evapotranspiration rate is calculated using a crop coefficient method:

ET ¼ k$PET

(3)

where ET the actual evapotranspiration rate (mm/hour), k is the crop coefficient, and PET is the potential evapotranspiration rate (mm/hour). The crop coefficient is determined by a user considering types of vegetation planted, and it is set to a constant parameter in the model for the simplicity assuming that typical green roofs drain well with amended soils used as soil beds. Soil water content is updated using a continuity equation:

SWðtÞ ¼ SWðt  1Þ þ ðIðtÞ  Q ðtÞ  ETðtÞÞ$Dt

(4)

where SW is the water content of an amended soil layer (mm) and Q is the surface runoff discharge rate (mm/hour), and Dt is the simulation time interval (hour). We assume that field capacity is the maximum storage capacity of the soil pores in the green roof.

309

Rain water infiltrates into the soil and accumulates in the soil storage while percolating to the bottom of the soil bed if the soil water content is smaller than the field capacity. Once the soil water content reaches the field capacity, the soil has no more storage capacity so the entire rainfall volume drains while soil moisture content does not change during the time step. Water seeped out of amended soil layer is the only source of surface runoff in the area covered by the green roof. Assuming shallow ponding depth and high hydraulic conductivity of a drainage media laid underneath the amended soil layer, the seepage rate becomes the same as the vertical hydraulic conductivity of the amended soil:

Q ¼ SP ¼ PC ¼ CK $K

(5)

where SP is the seepage rate (or percolation rate, PC) of the soil water, CK is the coefficient introduced to consider anisotropy of the soil and other uncounted factors controlling hydraulic conductivity, and K is the hydraulic conductivity of the amended soil layer (mm/ hour). The value of CK may be determined by a model calibration. When the soil is not saturated, a hydraulic conductivity is estimated using the van Genuchten equation (van Genuchten, 1980) based on soil water content, field capacity, and wilting point:

 2  l m K ¼ Ksat $Sle 1  1  Sem

(6)

where Ksat is the saturated hydraulic conductivity (mm/hour), Se is the effective (or relative) saturation (mm/mm), l and m are coefficients. The effective saturation is calculated based on the hyQQw Þ , where Q is the current water draulic features of soil: Se ¼ ððQ Q Þ s

w

content (mm/mm), Qw is the residual water content and approximated to the wilting point (mm/mm) for the simplicity, and Qs is the saturated water content (or porosity) (mm/mm).

2.2.4. Rain garden A rain garden receives flow from its drainage area and retains surface runoff. The ponding water is promoted to infiltrate into lower soils through an amended highly conductive soil layer. The rate of infiltration occurring at the bottom of the surface storage is calculated following the same steps as those developed for green roofs (Equations (1) and (2)). Similarly, the infiltration and percolation processes in a rain garden are calculated using those equations used for the green roof. Finally, the water content in the amended soil layer is updated using the continuity equation (Equation (4)), and the amount of the percolated water is added to soil water content of the native soil. Evaporation occurs from the surface of detention water, and percolation of the stored water to the soils continues. Once the storage is empty, evapotranspiration may still happen on the soil surface until soil water content of the amended soil reaches the wilting point. Once the stage of ponding water becomes higher than that of an orifice installed in the surface storage, the ponding water starts to drain the orifice. When the surface storage is filled with water, any runoff from upstream bypasses the rain garden. The summation of the bypassing runoff, discharging water, and rainfall directly deposited in the rain garden storage becomes the total surface runoff. The amount of water stored in the upper tank (surface storage) of a rain garden is calculated using a continuity equation:

h STRRG ðtÞ ¼ STRRG ðt  1Þ þ PðtÞ$AGardenSurface  IðtÞ$AGardenBottom  i þ 3600$ QRG;IN ðtÞ  ODðtÞ $Dt

(7)

310

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

where STRRG is the volume of water stored in the surface storage of a rain garden (m3), P is the rainfall depth (m), AGardenSurface is the rain garden surface area (m2), AGardenBottom is the area of the rain garden bottom where percolation occurs (m2), QRG;IN ðtÞ is the rate of flow coming in the rain garden storage (m3/s), and ODðtÞ is the rate of flow being discharged from an orifice pipe (m3/s), which is calculated using the orifice equation:

ODðtÞ ¼ Cd $Apipe

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi  2g hðtÞ  Hpipe

(8)

where Cd is a dimensionless runoff coefficient (0.611: Swamee and Swamee, 2010), Apipe is the cross section area of the orifice pipe (m2), g is the gravitational acceleration (9.81 m/s2), Hpipe is the height of the center of the pipe from the storage bottom (m), and Dpipe is the diameter of the pipe (m). The ponding depth hðtÞ is determined using STRRG ðtÞ and a stage-volume relationship of the storage. Detained water in the upper tank discharges in a controlled way through the orifice. In calculating the infiltration rate using the GAML equation, the depth of the stored water is considered as the ponding depth that increases hydraulic head in calculating the matric potential at the wetting front ðJÞ (Equations (1) and (2)). 2.2.5. Cistern The storage of a cistern is updated at every time interval using a continuity equation:

STRCS ðtÞ ¼ STRCS ðt  1Þ þ 3600$QCS;IN ðtÞ$Dt

(9)

where STRCS is the volume of water stored in a cistern (m3) during the time step t and QCS;IN is the inflow to the cistern (m3/s). If a cistern is serially connected to a green roof, QCS;IN is equal to the storm water draining from the green roof (SP). Once stormwater fills up the cistern, incoming water bypasses and drains. The detained stormwater in the cistern is consumed for irrigation after the storm ceases based on management input by user, which contributes to maintaining soil moisture content of the HRU. 2.2.6. Porous pavement Porous pavement system is a multi-layer system that promotes stormwater infiltration while securing structural stability. A typical application of porous pavement includes parking lots and roadway pavements. In porous pavement, rainfall infiltrates through the pavement surface into the gravel layer and to the soil base and then drains through lateral underdrain pipes or percolates into native soils. Infiltration rate is a function of the hydraulic characteristics of pavement system that vary widely (Drake et al., 2013). The gravel layer is assumed not to limit infiltration, but detention capacity is determined by porosity and hydraulic conductivity. The overall strategy for hydraulic simulation of porous pavement is similar to that of rain gardens (Fig. 3). The storage capacity of the gravel layer is calculated based on the porosity of gravels:

VolGRVEL ¼ APV $DPTPV $PORPV

(10)

where VolGRVEL is the maximum capacity of a gravel layer to store water (m3), APV is the area of porous pavement in the current HRU (m2), DPTPV is the depth of the gravel layer (m), and PORPV is the porosity of the gravel layer. Then, the amount of water stored in the gravel layer is calculated using a continuity equation:

STRPV ðtÞ ¼ STRPV ðt  1Þ þ IðtÞ$APV $Dt

(11)

where STRPV is water volume stored in the gravel layer of a porous pavement (m3). Once the gravel layer is fully filled with water,

surface runoff will be generated at the rate of the rainfall intensity on the porous pavement surface. The same procedures as those of the rain garden are applied to calculate the infiltration rate of stormwater stored in the gravel layer (Equations (1) and (2)) and the percolation rate of water stored in the soil base (Equation (5)). The rate of soil water drainage through a drainage pipe is calculated using Equation (12).

DRPV ðtÞ ¼ CDR $KðtÞ

(12)

where DRPV is the rate of draining water through a underdrain pipe (mm/hour), CDR is a coefficient representing the drainage efficiency of the pipe, depending on the characteristics of the pipe. Then, a continuity equation is used to account water content of the soil base:

SWPV ðtÞ ¼ SWPV ðt  1Þ þ ðIðtÞ  DRPV ðtÞ  PCðtÞÞ$Dt

(13)

where SWPV is soil water content of the soil layer (mm) and PC is the percolation rate in the soil base to native soils (mm/hour). 2.3. Strategies for LID module evaluation The proposed LID modules were tested by investigating model sensitivity to LID configuration, by comparing simulated flow out of the LIDs with field monitoring data, and by performing LID implementation scenario analysis. A sensitivity analysis was conducted for LID parameters newly introduced in the LID modules to evaluate the significance of the parameters to simulation output and understand the hydrologic behaviors of the modules. Then, simulated discharge from LIDs was evaluated at two experimental sites, one in Austin and another place in Dallas. Finally, LID implementation scenarios represented by the unique combinations of LID types and application areas were modeled in the Brentwood watershed in Austin, TX, to assess the watershed-scale effectiveness of stormwater management using the LIDs. 2.3.1. Sensitivity analysis Changes in simulated runoff hydrographs from LIDs were tracked while varying parameter values by up to 20% using a oneat-a-time (OAT) sampling approach. Then, the relative amount of changes in the outputs to that of parameter values is calculated using Equation (14), representing the sensitivity of the outputs to the parameters.





ðQþD QD Þ Qbase

2D

(14)

where Q±D is runoff volume calculated when increased or decreased a parameter value by D% and Qbase is runoff volume calculated with the baseline LID configuration. The OAT method was selected since it facilitates a simple way to have a clear idea about the roles of each parameter on simulation processes and results. As a cistern is simply represented with a storage tank in the module, sensitivity of surface runoff simulated with a cistern is quite straightforward (proportional to the storage dimension), thus the cistern module was not further investigated. The sensitivity analysis focused on the hydraulic characteristics of LIDs and their amended soil layers, including the hydraulic conductivity, porosity, field capacity, wilting point, and soil depth. Depending on the types of LIDs, surface storage and orifice sizes and gravel layer features were also included in the analysis. The sensitivity of runoff simulation using the LID modules must be a function of many other factors including size and temporal patterns

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

of storm events, initial conditions of the amended soil layer and storage, size of treated upstream areas and the configuration of the orifice in the case of the rain garden, and the configurations of the drainage pipe in the case of the porous pavement. For simplicity of the sensitivity analysis, these factors were set to constants in this study. Specifically, the same uniform rainfall of 38.1 mm as one used in the green roof experiments was employed for all the cases in the sensitivity analysis. The initial moisture content of the soil layer and the initial water volume of the gravel base were set to the field capacity and 30% of the void volume of the base, respectively. In the case of the rain garden, the treated upstream area of the rain garden was set to 13.91 m2, which is five times larger than its surface area of 2.78 m2. The orifice pipe was assumed to have the diameter of 0.05 m with the roughness coefficient of 0.012 s/m1/3 and the coefficient of discharge of 0.611, and the height of the orifice pipe bottom from the bottom of the surface storage was set to 0.05 m. It was assumed that 70% of water percolated into the amended soil layer of the porous pavement is removed through the drainage pipe system. 2.3.2. Field experiments and module testing Subdaily discharge measurements made at experimental green roof and rain garden sites were obtained from the City of Austin and Dallas Urban Solutions Center, respectively, and compared with field-scale (HRU-level) outputs of the modules to evaluate the performance of the LID modules (Fig. 5). An experimental green roof comprised 1.83 m by 1.52 m metal platform, and the roof system assembly was 22 gauge galvanized metal deck. Runoff was collected through a single (10 cm by 10 cm) scupper and directed into a catchment pan, and water level monitored with an ISCO flow meter. A rainfall event was generated by using brass nozzles calibrated to supply 38.1 mm/h over the entire roof platform surface. The nozzle spray cone was protected from water loss due to splashout or wind drift by 1 m high Plexiglas screen (Fig. 5a). The soil layers located under the green roof surface was classified as fine sands (Table 2). The experimental green roof was located in Austin, Texas, and the test was conducted in 2011. An experimental rain garden was built in Dallas Texas in 2012. The rain garden in a stadium shape was designed to collect runoff from surface runoff routed from its upstream drainage areas of 0.34 ha (Fig. 5b). The rain garden surface is 174-m2 wide and 0.186-

311

Table 2 LID configurations used for the field-scale validation and the baseline of the sensitivity analysis. Configuration

LID Types

Component

Items

Green Roof

Rain Garden

Amended soil layer

Hydraulic conductivity (mm/hr) Porosity (mm/mm) Field capacity (mm/mm) Wilting point (mm/mm) Soil depth (m) Depth (m) Area (m2) Area (ha)

130.95

130.00a

0.492 0.235 0.124 0.100 e e e

0.500 0.450 0.125a 0.900 0.186 174.0 0.34

Surface storage Drainage area

a The hydraulic conductivity and wilting point of the amended soil layers of the rain garden were adapted from those of the rain garden.

m deep, which allows storing stormwater of 32.4 m3. A 30-cm diameter corrugated plastic pipe (CPP) installed in the rain garden drains stored stormwater into its underdrain outlet box and eventually downstream ditch when the depth of stored stormwater becomes greater than 1 cm. The rain garden has a 0.91-m deep soil layer beneath the surface, and soil water can percolate (drain)

Table 3 LID implementation scenarios. Scenario

LID types Green Roof

Rain garden

Cistern

Porous pavement

Types of treated areas

Residential, commercial, industrial, and civic

Types of treated areas for green roof & parks and transportation

Residential, commercial, industrial, and civic

Parking lots

Baselinea All GR-5c GR-10 GR-15 GR-25 GR-50 GR-Full RG-5 RG-10 RG-15 RG-25 RG-50 RG-Full CS-5c CS-10 CS-15 CS-25 CS-50 CS-Full PV-5 PV-10 PV-15 PV-25 PV-50 PV-Full

0% 25%b 5% 10% 15% 25% 50% 100% 0% 0% 0% 0% 0% 0% 5% 10% 15% 25% 50% 100% 0% 0% 0% 0% 0% 0%

0% 25% 0% 0% 0% 0% 0% 0% 5% 10% 15% 25% 50% 100% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%

0% 25% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 100%d 100% 100% 100% 100% 100% 0% 0% 0% 0% 0% 0%

0% 25% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 5% 10% 15% 25% 50% 100%

a

The baseline scenario does not have any LID. The percentage means a fraction of treated area to impervious areas of an HRU where a LID placed. c The first two letters represent the type of a LID (GR e green roof, RG e rain garden, CS e cistern, and PV e porous pavement), and the following number signifies the fraction of treated area to the impervious areas of an HRU where a LID is placed. d A cistern is connected to a green roof in an HRU, and 100% means that a cistern receives all runoff coming out of the corresponding green roof. b

Fig. 5. Experimental (a) green roof (Austin, TX) and (b) rain garden (Dallas, TX).

312

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

through the bottom of the layer. Outflow including overflow and perforated pipe flow was measured with an ISCO flow meter at the downstream ditch. Outflow was measured for several storm events, and events that produced clear peaks were used for the test in this study. 2.3.3. Parameter calibration The hydrology simulation of the SWAT model prepared for the study watershed was calibrated to 15-min stream flow measurements made at the watershed outlet between Mar 2012 and Dec 2013, and then validated with the measurements made in 2014. Calibration parameters were selected based on the understandings of the SWAT model and the hydrologic characteristics of the study watershed. Then, the calibrated SWAT model was used to represent the baseline in the following scenario analysis for storm events and implementation areas. The Nash-Sutcliffe efficiency (NSE) coefficient (Nash and Sutcliffe, 1970) was used to measure the goodnessof-fit between predicted and observed 15-min streamflow hydrographs. The use of the NSE coefficient enabled evaluating the model performance regarding high flows as well as the overall agreement between the simulated and observed mean flow (Krause et al., 2005; Engel et al., 2007). In the calibration, the NSE coefficient was maximized by adjusting parameter values using a samplingbased heuristic optimization algorithm, AMALGAM (Vrugt and Robinson, 2007). Though AMALGAM has been used for SWAT calibration (Zhang et al., 2010; Her and Chaubey, 2015; Her et al., 2015), this was the first attempt to calibrate the SWAT model simulating subdaily dynamic hydrographs. 2.3.4. LID scenario analysis On-site effects of LIDs may not be directly proportional to their watershed-scale performance due to routing processes along flow paths on overland areas and in channels (Her et al., 2016). In addition, LID effects would be responsive to the characteristics of storm events such as rainfall intensity. For quantifying watershedscale effectiveness of the LIDs, this study investigated how watershed-scale LID effectiveness varied while (1) expanding LID

Table 4 Design storm sizes (mm) according to return periods and duration. Return period (yr)

2

5

10

25

50

100

250

Duration

43.7 58.9 77.7 87.4

57.9 79.5 103.4 126.7

68.1 94.2 122.2 154.9

83.3 115.6 149.9 194.1

96.3 134.1 174.2 225.3

111.0 155.2 202.2 259.1

133.6 187.5 245.6 304.8

1h 3h 12 h 24 h

application areas from 0% to 100% of the connected impervious areas of associated land covers (HRUs) and (2) changing the return periods of storm events from 2-year to 250-year (Tables 3 and 4). The sizes of design storms for the study watershed were obtained from the City of Austin (Drainage Criteria Manual: Section 2 e Determination of Storm Runoff, https://www.municode.com/libra ry/tx/austin/codes/drainage_criteria_manual?nodeId¼S2DESTRU, Accessed in 2015; Table 4). Two types of temporal distributions, uniform and SCS Type III, of a storm event were considered in constructing hyetographs of hypothetical design storms (Fig. 6). According to USDA NRCS (2015), Austin, TX belongs to the rainfall type III areas. The uniform rainfall was selected to show the responses of simulated storm water to different rainfall types. In the scenario analysis, the SWAT model calibrated for the Brentwood watershed was used as the baseline representing the current status of the study watershed that does not have any LID. The baseline configurations of the LIDs for the scenario analysis were determined based on the LID design guideline of the City of Austin and the settings of the field experiments (Table 5). It was assumed that cisterns were coupled with green roofs, and 30% of stored water in a cistern was irrigated into the adjacent pervious area of the corresponding HRU after four consecutive dry days in the scenario analysis. 3. Results and discussion 3.1. Sensitivity analysis on the effectiveness of LID configurations Hydrologic simulation using the LID modules showed unique responses to the soil and storage features of the LIDs (Fig. 7). Porosity was most influential to runoff depth simulated using the green roof module, followed by field capacity and depth of the amended soil layer, reflecting the significance of soil properties on the stormwater holding capacity of a green roof. Since the initial soil water content was set to field capacity in the sensitivity analysis, total runoff depth simulated increased with increase in field capacity. Runoff simulation of the rain garden module was most sensitive to hydraulic conductivity. Contrary to the case of the green roof, the depth of runoff decreased with increase in hydraulic conductivity. In the rain garden module, soil water in the amended soil layer percolates into the native soil rather than directly contributing to runoff. Thus high hydraulic conductivity leads to increase in storage capacity of the layer and therefore the runoff discharging from the rain garden decreases. Runoff depths simulated using the rain garden module responded to other soil features, including porosity, field capacity, wilting point, and soil depth, in the same directions as those of the green roof (Fig. 7a, b,

Fig. 6. SCS Type-III temporal patterns of a 25-year 24-hr storm design storm event.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

313

Table 5 The baseline LID configurations used in the scenario analysis. Configuration

LID types

Component

Items b

Amended soil layer

Surface storage Orificed

Cistern Gravel layer

a

Hydraulic conductivity (mm/hr) Porosity (mm/mm) Field capacity (mm/mm) Wilting point (mm/mm) Soil depth (m) Depth (m) Area (fraction to HRU area)c Distance between the orifice pipe bottom and the storage bottom (m) Diameter of the orifice pipe (m) Surface runoff depth captured (mm) Depth (m) Porosity (mm/mm)

Green roof

Rain garden

Cistern

Porous pavement

7.50 0.50 0.40 0.15 0.25 e e e

7.50 0.50 0.40 0.15 0.25 0.10 0.10 0.05

e e e e e e e e

7.50 0.50 0.40 0.15 0.25 e e e

e e e e

0.05 e e e

e 2.5e e e

e e 0.13 0.35

a The design guideline for a rain garden of the City of Austin regulates a minimum hydraulic conductivity of the soil layer to be equal to or greater than 5.08 mm/h, and 7.50 mm/h was applied to the cases of the green roof and rain garden for consistency of the analysis. b The hydraulic conductivity is the only regulation of the design guideline for the amended soil layer, thus, the other soil features were set to close to those of the field experiments. c In the SWAT LID module, the surface storage area of a rain garden is set to a fraction of treated area to the HRU area. d The orifice configurations were set based on professional experience. e In the SWAT LID module, the capacity of a cistern is set to the depth of surface runoff generated in treated areas draining to the cistern.

Fig. 7. Responses and sensitivity of runoff depth simulated using the LID modules to LID configurations. (a) green roof, (b) rain garden, (c) porous pavement, and (d) relative sensitivity of runoff depth simulated using the LID modules to the characteristics of amended soil layers.

314

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

and 7d). Runoff simulation result with the rain garden module was insensitive to the maximum retention capacity of the rain garden. Obviously, the depth of runoff coming out of the rain garden storage was nonlinearly proportional to the diameter of the orifice pipe (Fig. 7b). Simulated runoff from porous pavement was responsive to the features of gravel base and amended soil layers, and it was most sensitive to parameters (depth and porosity) defining the storage capacity of the gravel base layer (Fig. 7c and d). Since the rain garden and porous pavement were similarly represented with upper (surface storage and gravel base) and lower (amended soil) layers (Fig. 3) in the modules, these LIDs showed similar patterns of runoff responses but were different in magnitudes reflecting different characteristics of amended soil layers used in these LIDs (Fig. 7b, c and 7d). 3.2. Comparison of LID model outputs with field measurements Simulated runoff the LID modules was compared with field data measured at experimental sites in Dallas and Austin, to assess the validity of the LID simulation (Figs. 8 and 9). In the Dallas site, the green roof was underlain by a shale soil (fine sand) layer (Fig. 5 and Table 2), and the model over-predicted peak runoff though the overall runoff volume was close to the measured amount (Fig. 8a). In the case of the perlite soil (loamy fine sand; Table 2), both simulated and observed runoff hydrographs showed constant peak rates representing the equilibrium condition of the green roof system (Fig. 8b). Overall, the model well predicted falling limb with the shale soil and peak discharge with the perlite soil. The observed rising limbs came earlier than did the simulated in the both cases, indicating the initial soil water content might be underestimated in the simulation. The discrepancy was also attributed to the process representation simplified at the HRU scale in the module. In the case of Fig. 8a, for instance, we can see that the green roof system did not reach its equilibrium, as the measured runoff did not achieve a steady state at its maximum rate, but the simulation did. Although the soil was classified as sand based on its clay, sand, and silt content, the classification must be fuzzy and continuous rather

than distinct and discrete. Thus, the representation of the soil feature might be inaccurate to some extent in the modeling. In addition, the macro poles in the soil might accelerate the drainage of soil water at the beginning and prevent the system reaching its equilibrium, which could not be considered in the HRU scale modeling. The rain garden produced runoff hydrographs quite different from those of the green roof (Figs. 8 and 9). The model provided peak runoff rates close to the observed but underestimated the overall runoff volume, which was attributed to the underestimation of a hydrograph tail. Since an orifice installed in the surface storage is the main route for runoff to come out of the rain garden, unlike the green roof case where infiltrated water percolates out of the bottom of an amended soil layer, rising and falling limbs of the rain garden runoff hydrograph should be much steeper than those of the green roof. Also, it is reasonably speculated that water drained out of the surface soil might contribute to the runoff hydrograph even after the surface storage was emptied, which was not considered in the LID module. The simulated and observed hydrographs presented complicated responses to a storm event with multiple peaks (Fig. 9b). Similar to the green roof case, the runoff hydrograph simulated with the rain garden rose earlier than did the observed. Inaccurate initial soil water content estimation would be attributed to the earlier start of the simulated rising limb than that of the observed. 3.3. Evaluation of SWAT subhourly hydrological simulation Hydrologic simulation of the SWAT model prepared for the study watershed was calibrated with acceptable model performance statistics (NSE and R2) in the both calibration and validation periods (Table 6; Krause et al., 2005; Engel et al., 2007), and calibrated parameter values are provided in Table 7. Daily and monthly runoff hydrographs simulated using the calibrated subhourly SWAT model are compared with the observed in Fig. 10. Overall, the 1:1 line goes through the center of the daily runoff rate points implying that the model predicted the central trends of actual flow rates, but overestimations in low flows and underestimations in high flows

Fig. 8. Comparison of runoff observed and simulated with the green roof: (a) shale soil and (b) perlite soil.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

315

Fig. 9. Comparison of runoff observed and simulated with the rain garden: (a) 2 September 2013 and (a) 28 September 2013.

are found (Fig. 10a). Simulated monthly runoff time series well follow the trends of observed flow, but large overestimations were found in a few summer months during the calibration period (Mar 2012 to Dec 2013) as well as in September and November of the validation period (2014) (Fig. 10b). The model overestimated runoff volume by 15.2% and 27.9% in the calibration and validation periods, respectively. The large bias was mainly contributed to the baseflow overestimation (Fig. 10). Table 6 Goodness-of-fit statistics of the calibrated model. R2

Period

NSE 15-min

Daily

Monthly

15-min

Daily

Monthly

Calibration Validation

0.88 0.71

0.91 0.84

0.91 0.73

0.89 0.71

0.91 0.86

0.94 0.89

The study area was a small catchment which is highly urbanized with many impervious areas. As is typical in an urban catchment, the amount of baseflow observed in this catchment was small, especially during dry periods between storm events. The subhourly SWAT model provided tiny baseflow, which was still greater than the observed. Overall, the amount of total baseflow simulated was about two times greater than that of the observed, even though SWAT parameters (ALPHA_BF, GWQMN, REVAPMN, and GW_REVAP) related to baseflow simulation were calibrated (Table 7). However, the overestimation was rather expected since the goodness-of-fit measures used as the objective functions (NSE and R2) in the parameter calibration tend to emphasize on high flow than low flow, which is a required feature of a performance measure for urban stormwater modeling concerning about flooding. The large relative errors in the validation period were mainly attributable to the overestimated runoff for a storm event on

Table 7 Default and calibrated values of the SWAT parameters for hydrological simulation. Parameter

Inputa

Description

Range

Default

Calibrated

SURLAG TIMP ESCO OV_N SLOPE DEP_IMP ALPHA_BF GWQMN REVAPMN GW_REVAP CN_F CH_NII CH_SII SOL_AWC SOL_z CH_NI CH_SI

bsn bsn hru hru hru hru gw gw gw gw mgt rte rte sol sol sub sub

Surface runoff lag time (days) Snow pack temp. lag factor Soil evaporation comp. factor Manning's n for overland flow Average slope steepness (SFb) Depth of the impervious layer (mm) Baseflow alpha factor Threshold depth of shallow aquifer (mm) Threshold depth for ‘revap’ (mm) Groundwater revap coeff. Curve number (SF) Manning's n for the main channels Average slope of the main channel (SF) Available water capacity (mm/mm, SF) Depth of the soil layers (SF) Manning's n for the tributary channels Average slope of the tributary channels (SF)

0.01e1.00 0.01e1.00 0.01e1.00 0.01e0.30 50%-100% 0-3000 0.01e1.00 0-5000 0.1e5000 0.010e0.200 50%-35% 0.01e0.30 50%-100% 30%-100% 5%-5% 0.008e0.065 50%-100%

0.50 1.00 0.01 0.14 0% 3000 0.05 0 1.0 0.020 0% 0.014 0% 0% 0% 0.014 0%

0.95 0.42 0.02 0.23 32% 473 1.00 4741 46.2 0.195 16.6% 0.070 2.4% 67% 2.8% 0.063 42.6%

a b

SWAT input file associated with a parameter. Scale factor (SF) proportionally changes values of spatially distributed parameters. Detailed descriptions of the parameters can be found in Neitsch et al. (2011).

316

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

Fig. 10. Comparison of runoff hydrographs observed and simulated with the calibrated SWAT model: (a) daily and (b) monthly hydrographs.

September 8, 2014. This storm gave 74.9 mm rainfall, but no runoff was recorded at the stream gage while the model predicted 30.4 cms of discharge. According to the field notes, the flow level logger was turned off for a gage test performed on the day, and small runoff without rainfall was occasionally found in the raw datasets, which was likely to affect the quality of the flow measurements on this particular date. The event occurred on November 5th, 2014 provided a rainfall depth of 30.7 mm, but runoff of only 5.4 m3/s was observed while runoff of 18.3 m3/s was predicted. Considering another rainfall event of 24.9 mm producing runoff rate of 5.8 m3/s on November 22nd, 2013, the observed flow rate of 5.4 m3/s seems reasonable. In the 15-min interval simulation of SWAT, the amount of water infiltrated into the soil is calculated using the Green-Ampt equation where infiltration capacity is a function of soil properties and water content (Jeong et al., 2013). As soil water content increases with the infiltration of rainwater, infiltration rate calculated using the equation decreases over time as the depth of capillary head increases, and ultimately the infiltration process turns to Darcy flow. In response, estimated surface runoff increases over time. Thus, the overestimations of runoff could be attributed to overestimation of initial soil moisture conditions of storm events. This is particularly

evident during September and November of 2013 and 2014, suggesting further investigation is needed on soil water accounting procedure and/or sensitivity of infiltration rate to soil water content in the Green-Ampt equation of SWAT especially when soils are dry (Kannan et al., 2007). Surface runoff rate is influenced by many factors, such as the amount of rainfall, spatiotemporal variation of rainfall, and initial soil hydrological conditions. The observations show that a rainfall event of 32.5 mm produced runoff volume of 16,262 m3 on May 12th, 2014, and the following event of 33.0 mm produced only 11,405 m3 on the next day. For these storms, the model predicted runoff volumes of 14,787 m3 and 19,196 m3, respectively. The soil should be wetter on the second day than the first day, and the study watershed is small enough to assume the uniform spatial distribution of rainfall. Thus, the higher runoff rate of the first day was likely to result from its higher rainfall intensity. The rainfall measurements show that the maximum rainfall intensities of the first and second days were 36.6 mm/h and 11.2 mm/h, respectively. Therefore, it can be reasonably said that the Green-Ampt equation may need further improvements to reflect rainfall intensity on its infiltration rate calculation in SWAT more accurately.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

3.4. LID scenario analysis 3.4.1. LID performance on design storm events Runoff peaks and volumes simulated using the LID modules showed different levels of sensitivity according to storm sizes and temporal patterns depending on types of the LIDs incorporated (Fig. 11). In the LID simulation of SWAT, twelve combinations of storm sizes (2, 5, 10, 25, 100, and 250 years) and patterns (SCS Type III and uniform) were incorporated. The porous pavement was more effective in reducing runoff peaks of small storms (of the 2 and 5-year return periods) than large ones, but rain garden better handled large storms than small ones. On constant rainfall intensity, the rain garden held all of the rain water fell into it plus the runoff routed from its upstream impervious drainage areas within an HRU, except for the 250-year storm. The runoff peak increased with increase in the storm sizes, and the increase rates (the slopes of the lines in Fig. 11a) were relatively large for the porous pavement and small for the rain garden. The gravel layer of the porous pavement can hold rainfall infiltrated until its storage (pore spaces between gravels) is filled with water. Water infiltrated into the gravel layer slowly moves down to the bed soil layer, and some of the soil water discharges to downstream through the drainage pipe. Once the gravel layer is saturated, any subsequent rainfall will become direct runoff and bypass the

317

pavement. Thus, the porous pavement is less effective in treating storm events producing large amount of rainfall that quickly fill the retention storage. On the other hand, it produces little runoff discharging through the drainage pipe for small storms that does not give much rainfall enough to fill the storage. The rain garden produced runoff whenever the level of water stored in the surface storage went over the height of the orifice pipe. Unlike green roof or porous pavement, the rain garden receives stormwater discharging from upstream drainage areas. Therefore, even a small rainfall event can yield much runoff for a rain garden to treat depending on the size of drainage areas. Thus, the rain garden is less effective in reducing peak runoff than is the porous pavement when the size of a storm event is small. Water stored in the surface storage of the rain garden is quickly drained through its orifice pipe so that the surface storage can maintain its runoff retention capacity, which makes it effective in reducing peak runoff of large storm events. The green roof was effective in treating small storm events. When rainfall intensity was constant over time, however, the effectiveness of the green roof was relatively steady as the storm size increased (Fig. 11b). The rain drops falling onto the green roof infiltrates into the layer of amended soils and then eventually comes out of the layer. Only when rainfall intensity is greater than infiltration rate or the soil layer is completely saturated, rainfall will

Fig. 11. Sensitivity of runoff peak and depth simulated using the LID modules to the size of storm events: (a) SCS Type III rainfall and (b) uniform rainfall.

318

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

become direct runoff on the surface or bypass the green roof. An event that has constant and small rainfall intensity can lead the green roof system into the equilibrium condition where rate of inflow is the same as that of outflow, which is evident in the

responses (constant runoff peak and constant increase in runoff volume) of the green roof to the storm events (Fig. 11b). The LIDs showed unique hydrologic responses to a storm event designed for 100-year return period (Fig. 12). When a 100-year 24-h

Fig. 12. Hydrologic behaviors of the LIDs at the HRU scale: (a) green roof, (b) rain garden connected to upstream drainage areas, (c) rain garden disconnected from upstream drainage areas, and (d) porous pavement.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

storm event of 259.1 mm was assumed to come in the SCS Type III pattern (Table 4 and Fig. 6), the green roof captured 68% of rainfall fell on its surface. When the rainfall intensity becomes greater than the infiltration capacity of the soil, excess rainfall occurred and became direct runoff and bypassed the green roof. Although the soil layer got saturated as rainfall continued, its moisture retention capacity maintained over time because the soil water continued to percolate through the bottom of the amended soil layer (Fig. 3). Thus, the green roof could be still effective even after the storm peak, as shown by the shaded area between noon and 18:00 in Fig. 12 (a). The rain garden retained all stormwater at the beginning of the storm event, but it produced runoff depths greater than that of rainfall (Fig. 12b). Since the rain garden is designed to treat stormwater runoff draining from its impervious drainage area within an HRU as well as direct raindrops falling onto it, the total discharge is likely to go beyond the stormwater holding capacity of its surface storage. When the rain garden was not connected to its upstream drainage areas, its surface storage and soil layer could accommodate all stormwater generated from the 259.1-mm rainfall event on its surface (Fig. 12c). Similar to the case of the green roof, water stored in the soil layer percolates into the base soil, and then water ponded in the surface storage further infiltrates into the soil layer. Thus, its stormwater holding capacity can be recovered after the peak of a storm event (Fig. 12b). Response from the porous pavement to the storm event was similar to that of a simple storage tank (Fig. 12d). Since the gravel layer of the pavement does not limit infiltration, rainfall fallen on the porous surface immediately reached the gravel layer, which then held infiltrated water until the voids of the layer are filled. The water stored in the gravel layer further percolated into the base soil and then discharged through a porous pipe installed in the soil

319

base, which made the porous pavement still effective even after the amount of water stored in the gravel layer reached the maximum retention capacity. Overall, the performances of the green roof and the porous pavement were significantly influenced by peak rainfall intensity and the total volume of the rainfall, respectively (Fig. 12a and d). On the other hand, the size of upstream drainage area was the key determinant regarding the effectiveness of the rain garden (Fig. 12b and c). SWAT provides another option to represent urban BMPs in its hydrologic simulation, which is called ‘blanket BMPs’. If configured with a certain percentage reduction, then sediment, N, or P loading to runoff from an HRU is reduced by the given percentage. This is the simplest way to represent urban BMPs and model their effectiveness in hydrologic simulation, but it does not explicitly consider hydrologic processes happening in a LID. For instance, the unique responses of LIDs to storm events cannot be modeled with the current options due to their oversimplified BMP representation. On the other hand, the process-based approach of the proposed SWAT LID module describes physical processes that control the hydrologic behavior of LIDs even within a storm event as seen in Fig. 12. Such capacity will help LID modeling be more explicit and objective. 3.4.2. LID implementation scenarios In general, the implementation of LIDs demonstrated 1) decrease in water yield and the amount of surface runoff flowing out of the watershed and 2) increase in soil moisture and actual evapotranspiration (AET) of the HRUs where the LIDs were simulated (Figs. 13 and 14). Field scale effectiveness of the LIDs varied across HRUs due to spatial variability in topography, land uses, and soils. At the watershed scale, the effectiveness of LIDs was proportional to the treated areas (Figs. 13 and 14). With the default LID configuration (Table 5), the rain gardens provided the greatest

Fig. 13. Variations in runoff and AET simulated with the LID implementation scenarios across urban HRUs: (a) water yields (total runoff or streamflow), (b) surface runoff (or direct runoff), and (c) actual evapotranspiration.

320

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

reduction in water yield and surface runoff volume at the HRU scale, followed by porous pavement and green roof with/without cistern (Fig. 13). Rain garden and porous pavement both are configured with two storage tanks as depicted in Fig. 3, which

provides greater retention capacity than other LIDs with a single storage tank. The implementation of LIDs increased the soil water content and AET (Fig. 13). For instance, the green roof increases AET of the HRU even though soil water in the green roof's amended soil

Fig. 14. Runoff hydrographs simulated with the LID implementation scenarios at the watershed outlet. (a) and (e) green roof, (b) and (f) rain garden, (c) and (g) cistern, and (d) and (h) porous pavement; (a), (b), (c), and (d): the 2012 storm, and (e), (f), (g), and (h): the 2013 storm.

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

321

Fig. 15. Relationship between treated areas and reductions in runoff peaks and volumes. (a) the 2012 storm and (b) the 2013 storm.

layer is not allowed to directly contribute to increasing soil water content of the native soil layers. Since the impervious areas of the HRU were converted to the pervious areas represented by the green roof in the LID simulation, the overall soil water of the HRU should increase with an increase in pervious areas. In the case of the cistern, the default module was formulated to irrigate 30% of water stored in the tank into pervious areas of the HRU when the soil becomes dry (Table 5), which makes a cistern be linked to the overall HRU hydrology. It is worth noting that a cistern was assumed to be hydraulically connected to a green roof in the watershed-scale scenario analysis (Table 5). This reflects common implementation practices of green roofs and cisterns in the study watershed so that the water-logging effects with the green roof could be enhanced with cisterns (Sce4-1 to 5). Watershed-scale impacts of the LIDs on stormwater hydrographs with different LID implementation scenarios are presented in Fig. 14 for selected storm events that occurred on 12 to 13 May 2012 (one peak) and 12 to 13 October 2013 (two peaks). The LIDs were effective in reducing surface runoff volume and peaks, and the effectiveness of the LIDs was proportional to the size of treated areas (Fig. 15). The rain garden provided the greatest reductions in the volume and peaks because of its largest treated areas (or wide ranges of land use types applicable) (Table 3) and highest fieldscale effectiveness (Fig. 11). The green roof was more effective than was the porous pavement due to its wide applicability (Table 3). In the scenario analysis, the porous pavements were placed in only parking lots, while the green roofs were distributed into residential, commercial, industrial, and civic areas (Table 3), indicating the tangible watershed-scale effectiveness of the LIDs would require wide application areas as well as high field-scale efficiency. The LID effectiveness exhibited in the runoff hydrographs shows different temporal patterns depending on spatial scales (Figs. 12 and 14). The HRU-scale runoff depth hydrographs promptly reacted to the LID implementation, and the runoff reduction rate decreased with time as the storages got filled with stormwater during the storm event (Fig. 12). At the watershed scale, the LIDs effectiveness was proportional to discharge rates, and the peak rainfall was followed by the peak runoff with the lag time being about two hours. The delayed responses of the streamflow to the

LID implementation was caused by landscape routing including overland flow and channel flow. The watershed-scale effectiveness was not significantly diminished in the second event as LIDs’ stormwater holding capacities were recovered by continuing water percolation and drainage processes (Fig. 14e and h). 4. Summary and conclusions Modules for simulating LID practices including a green roof, rain garden, cistern, and porous pavement were developed and integrated into the subhourly simulation components of SWAT. The modules were tested through sensitivity analysis, field-scale assessment, and watershed-scale assessment of LID effectiveness. Each SWAT LID module demonstrated unique sensitivity to the characteristics of soils that were used to represent embedded soil layers in LIDs. The LID modules well reproduced the hydrologic features of subhourly runoff hydrographs observed at fields where the LIDs were installed. Subhourly runoff hydrographs simulated using the LID modules were reasonably responsive to the size and temporal distribution of storm events. Rain garden was more effective in controlling large storm events compared to other LIDs evaluated. On the other hand, porous pavement did not produce runoff on small storm events. All LIDs showed better performance in controlling uniform rainfall compared with the SCS type III rainfall. At the watershed scale, rain garden provided most efficient measures in reducing peaks and volumes of runoff in the Brentwood watershed. Porous pavement was least effective due to its smallest application area. These application results demonstrated that functions of LIDs could be effectively conceptualized with storage tanks at HRU level so that land-use heterogeneity of an urban watershed could be represented will in dynamic stormwater simulations. Thus, the SWAT LID module is expected to provide a new framework and tool to simulate LIDs. Cost-benefit analysis, the spatial optimization of LID placement, and water quality processes of LIDs are left for future research. References Ahiablame, L.M., Engel, B.A., Chaubey, I., 2012. Effectiveness of low impact development practices: literature review and suggestions for future research. Water,

322

Y. Her et al. / Environmental Modelling & Software 96 (2017) 305e322

Air, & Soil Pollut. 223 (7), 4253e4273. Ahiablame, L.M., Engel, B.A., Chaubey, I., 2013. Effectiveness of low impact development practices in two urbanized watersheds: retrofitting with rain barrel/ cistern and porous pavement. J. Environ. Manag. 119, 151e161. Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic modeling and assessment part I: model development. J. Am. Water Resour. Assoc. 34 (1), 73e89. Bicknell, B.R., Imhoff, J.C., Kittle Jr., J.L., Donigian Jr., A.S., Johanson, R.C., 1996. Hydrological Simulation Program-FORTRAN. User's Manual for Release 11. USEPA, Washington, DC. Brander, K.E., Owen, K.E., Potter, K.W., 2004. Modeled impacts of development type on runoff volume and infiltration performance. J. Am. Water Resour. Assoc. 40 (4), 961e969. Brown, R.A., Borst, M., 2015. Quantifying evaporation in a permeable pavement system. Hydrol. Process. 29, 2100e2111. Burszta-Adamiak, E., Mrowiec, M., 2013. Modelling of green roofs' hydrologic performance using EPA's SWMM. Water Sci. Technol. 68 (1), 36e42. City of Austin, 2012. PLANNINGCADASTRE.land_use_2012. City of Austin Planning and Development Review Department. Dietz, M.E., 2007. Low impact development practices: a review of current research and recommendations for future directions. Water, Air, Soil Pollut. 186 (1e4), 351e363. Drake, J.A.P., Bradford, A., Marsalek, J., 2013. Review of environmental performance of permeable pavement systems: state of the knowledge. Water Qual. Res. J. Can. 48, 203e222. Elliott, A.H., Trowsdale, S.A., 2007. A review of models for low impact urban stormwater drainage. Environ. Model. Softw. 22 (3), 394e405. Engel, B., Storm, D., White, M., Arnold, J., Arabi, M., 2007. A hydrologic/water quality model application. J. Am. Water Resour. As. 43 (5), 1223e1236. Freni, G., Mannina, G., Viviani, G., 2010. Urban storm-water quality management: centralized versus source control. J. Water Resour. Plan. Manag. 136 (2), 268e278. Gassman, P.W., Reyes, M.R., Green, C.H., Arnold, J.G., 2007. The soil and water assessment tool: historical development, applications, and future research directions. Trans. ASABE 50 (4), 1211e1250. Gilroy, K.L., McCuen, R.H., 2009. Spatio-temporal effects of low impact development practices. J. Hydrol. 367 (3), 228e236. Her, Y., Frankenberger, J., Chaubey, I., Srinivasan, R., 2015. Threshold effects in HRU definition of soil and water assessment tool. Trans. ASABE 58 (2), 367e378. Her, Y., Chaubey, I., 2015. Impact of the numbers of observations and calibration parameters on equifinality, model performance, and output and parameter uncertainty. Hydrol. Process. 29 (9), 4220e4237. Her, Y., Chaubey, I., Frankenberger, J., 2016. Effect of conservation practices implementation by USDA programs at field and watershed scales. J. Soil Water Conserv. 71 (3), 249e266. Hood, M.J., Clausen, J.C., Warner, G.S., 2007. Comparison of stormwater lag times for low impact and traditional residential development. J. Am. Water Resour. Assoc. 43 (4), 1036e1046. Jeong, J., Kannan, N., Arnold, J., Glick, R., Gosselink, L., Srinivasan, R., 2010. Development and integration of sub-hourly rainfall-runoff modeling capability in a watershed model. Water Resour. Manag. 24 (15), 4505e4527. Jeong, J., Kannan, N., Arnold, J., Glick, R., Gosselink, L., Srinivasan, R., Harmel, D., 2011. Development of subdaily erosion and sediment transport models in SWAT. Trans. ASABE 54 (5), 1685e1691. Jeong, J., Kannan, N., Arnold, J., Glick, R., Gosselink, L., Srinivasan, R., Barrett, M.E., 2013. Modeling sedimentation-filtration basins for urban watersheds in SWAT. J. Environ. Eng. 139 (6), 838e848. Kannan, N., Santhi, C., Williams, J.R., Arnold, J.G., 2007. Development of a continuous soil moisture accounting procedure for curve number methodology and its behavior with different evapotranspiration methods. Hydrol. Process. 22 (13), 2114e2121. Kannan, N., Jeong, J., Arnold, J.G., Gosselink, L., Glick, R., Srinivasan, R., 2014. Hydrologic modeling of retention irrigation system. J. Hydrol. Eng. 19 (5), 1036e1041. €se, F., 2005. Comparison of different efficiency criteria for Krause, P., Boyle, D.P., Ba

hydrological model assessment. Adv. Geosci. 5, 89e97. Li, Y., Babcock, R.W., 2014. Green roof hydrologic performance and modeling: a review. Water Sci. Technol. 69 (4), 727e738. Loperfido, J.V., Noe, G.B., Jarnagin, S.T., Hogan, D.M., 2014. Effects of distributed and centralized stormwater best management practices and land cover on urban stream hydrology at the catchment scale. J. Hydrol. 519, 2584e2595. Mullaney, J., Lucke, T., 2014. Practical review of pervious pavement designs. Clean. Soil Air Water 42 (2), 111e124. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I - A discussion of principles. J. Hydrol. 10, 282e290. Nearing, M.A., Liu, B.Y., Risse, L.M., Zhang, X., 1996. Curve numbers and Green-Ampt effective hydraulic conductivities. J. Am. Water Resour. Assoc. 32 (1), 125e136. Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2011. Soil and Water Assessment Tool, Theoretical Documentation, Version 2009. Texas Water Resources Institute Technical Report No. 406. Texas A&M University System, College Station, TX. Obropta, C.C., Kardos, J.S., 2007. Review of urban stormwater quality models: deterministic, stochastic, and hybrid approaches. J. Am. Water Resour. Assoc. 43 (6), 1508e1523. Palla, A., Gnecco, I., 2015. Hydrologic modeling of Low Impact Development system at the urban catchment scale. J. Hydrol. 528, 361e368. Rosa, D.J., Clausen, J.C., Dietz, M.E., 2015. Calibration and verification of SWMM for low impact development. J. Am. Water Resour. Assoc. http://dx.doi.org/10.1111/ jawr.12272. Rossman, L.A., 2010. Storm Water Management Model User's Manual, Version 5.0. National Risk Management Research Laboratory, Office of Research and Development. US Environmental Protection Agency. Roy, A.H., Wenger, S.J., Fletcher, T.D., Walsh, C.J., Ladson, A.R., Shuster, W.D., Thurston, H.W., Brown, R.R., 2008. Impediments and solutions to sustainable, watershed-scale urban stormwater management: lessons from Australia and the United States. Environ. Manag. 42 (2), 344e359. Shoemaker, L., Riverson, J., Alvi, K., Zhen, J., Paul, S., Rafi, T., 2009. SUSTAIN - a Framework for Placement of Best Management Practices in Urban Watersheds to Protect Water Quality. National Risk Management Research Laboratory. United States Environmental Protection Agency. Report: EPA/600/R-09/095. Sutherland, R.C., 2000. Methods for Estimating the Effective Impervious Area of Urban Watersheds, Article 32. In the Practice of Watershed Protection, pp. 193e195. Swamee, P.K., Swamee, N., 2010. Discharge equation of a circular sharp-crested orifice. J. Hydraul. Res. 48 (1), 106e107. USACE, 2000. Hydrologic Modeling System, HEC-HMS, Technical Reference Manual. U.S. Army Corps of Engineers, Hydrologic Engineering Center. USDA, 1986. Urban Hydrology for Small Watersheds. USDA-NRCS-conservation Engineering Division. Technical Report 55. U.S. Department of AgricultureNatural Resources Conservation Service. USDA, 2010. National Engineering Handbook, Part 630 Hydrology, Chapter 15: Time of Concentration. U.S. Department of Agriculture -Natural Resources Conservation Service. USDA, 2015. Description of SSURGO database. Soil survey staff, natural resources conservation service, United States department of agriculture. Web soil survey. Available online at: http://www.nrcs.usda.gov/wps/portal/nrcs/detail/soils/su rvey/?cid¼nrcs142p2_053627. (Accessed in November 2015). USDA NRCS, 2015. National Engineering Handbook: Part 630 Hydrology e Chapter 4: Storm Rainfall Depth and Distribution. USDA Soil Conservation Service, Washington, DC, USA. van Genuchten, M.T., 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Sci. Soc. Am. J. 44, 892e898. Vrugt, J.A., Robinson, B.A., 2007. Improved evolutionary optimization from genetically adaptive multimethod search. P. Natl. Acad. Sci. USA 104 (3), 708e711. Zhang, X., Srinivasan, R., Liew, M.V., 2010. On the use of multi-algorithm, genetically adaptive multi-objective method for multi-site calibration of the SWAT model. Hydrol. Process. 24 (8), 955e969. Zoppou, C., 2001. Review of urban storm water models. Environ. Model. Softw. 16 (3), 195e231.